Manus is out of control

00:44:32
https://www.youtube.com/watch?v=D6jxT0E7tzU

Ringkasan

TLDRManis AI est considéré comme le premier agent d'intelligence artificielle entièrement autonome, capable d'effectuer des recherches, de créer des sites web et d'accomplir diverses tâches de programmation sans intervention humaine. Les utilisateurs expriment leur étonnement face à ses capacités et comparent positivement ses performances à celles d'autres entreprises leader dans le domaine de l'IA. Lancé le 7 mars, Manis AI pourrait transformer le développement de logiciels en rendant les processus plus efficaces et accessibles.

Takeaways

  • 🚀 Première démonstration d'un agent IA autonome.
  • 💡 Manis AI offre des capacités avancées de recherche et de création.
  • 💻 Utilise une architecture cloud avec des agents multiples.
  • 📅 Lancement le 7 mars, avec une forte attente d'accès.
  • 💬 Exempt de sponsoring, basé sur des expériences authentiques.

Garis waktu

  • 00:00:00 - 00:05:00

    Introduction de Manis AI, présenté comme le premier agent autonome d'IA, avec des commentaires de personnes influentes faisant l'éloge de cette technologie révolutionnaire. Un test utilisateur exprime une transformation de la perspective sur l'IA et souligne que cela pourrait être le point de départ de l'ère des agents autonomes.

  • 00:05:00 - 00:10:00

    Exécution de divers tests sur Manis en mode standard, avec une recherche sur les jeux vidéo créés à l'aide de l'IA. Le test implique la collection d'informations sur des jeux récents et la création d'un site internet pour les présenter, en mettant l'accent sur l'utilisation de technologies modernes dans le développement de jeux.

  • 00:10:00 - 00:15:00

    Détails sur la configuration technique de Manis, en utilisant une machine virtuelle sous Linux, permettant des opérations complexes. Manis montre sa capacité à gérer des tâches multiples, y compris le découpage de projets en sous-tâches et l'organisation des processus de recherche.

  • 00:15:00 - 00:20:00

    Évaluation des capacités de codage de Manis, en lui demandant de créer un cours en ligne sur l'utilisation de Linux avec des ressources d'IA. Manis est capable de produire un plan de leçon et de créer un site web avec un thème spécifique, en s'assurant que les utilisateurs puissent suivre sans expérience préalable.

  • 00:20:00 - 00:25:00

    Manis illustre ses capacités via l'utilisation de terminaux et de codage en temps réel, démontrant comment les projets sont gérés avec précision. Les performances et la structure de code sont admirées pour leur intelligence et leur efficacité dans la création d'un environnement de développement personnalisé.

  • 00:25:00 - 00:30:00

    Le test inclut une exploration de l'idée d'utiliser Manis pour coder un jeu de simulation de combat aérien, en suivant des instructions précises sur la création d'éléments de jeu. Manis est capable de générer des sous-tâches et de gérer des commandes complexes sans intervention humaine.

  • 00:30:00 - 00:35:00

    Des tests sur la création d'un script vidéo via des APIs montrent que Manis rencontre des obstacles. Malgré certains échecs dans la génération audio, il réussit à créer une interface utilisateur, à mettre en place une logique et à gérer des erreurs de façon autonome, ce qui souligne son potentiel dans le développement de contenu.

  • 00:35:00 - 00:44:32

    Conclusion avec une discussion sur les performances globales de Manis et des prévisions sur l'avenir du développement d'IA. L'utilisateur exprime une surprise constante face aux capacités de Manis, et indique un intérêt croissant pour l'avenir de cette technologie, mentionnant un potentiel d'open source à venir.

Tampilkan lebih banyak

Peta Pikiran

Video Tanya Jawab

  • Qu'est-ce que Manis AI?

    Manis AI est un agent AI autonome que les utilisateurs considèrent comme étant en avance sur les technologies de pointe d'autres entreprises comme OpenAI.

  • Est-ce que Manis AI est sponsorisé?

    Non, l'utilisateur qui a testé Manis AI précise qu'il n'y a pas de sponsoring ni de compensation financière.

  • Comment fonctionne Manis AI?

    Manis AI utilise une architecture multi-agent et fonctionne sur un environnement cloud, lui permettant de réaliser des tâches complexes de manière autonome.

  • Quels types de tâches Manis AI peut-il accomplir?

    Il peut réaliser des recherches, créer des sites web, proposer des leçons sur Linux et automatiser des processus de développement.

  • Quand Manis AI a-t-il été lancé?

    Manis AI a été lancé le 7 mars.

  • Quels langages de programmation Manis AI supporte-t-il?

    Manis AI prend en charge plusieurs langages, y compris ceux utilisés pour le développement de jeux et d'applications web.

  • Comment Manis AI gère-t-il les erreurs?

    Lorsqu'il rencontre une erreur, Manis AI continue d'essayer de résoudre le problème et d'optimiser les résultats.

  • Est-ce que Manis AI est accessible à tous?

    Actuellement, il y a une liste d'attente avec un million de personnes pour obtenir l'accès à Manis AI.

  • D'où tire-t-il ses informations?

    Manis AI utilise des recherches sur le web pour collecter des informations pertinentes pour les tâches qu'il effectue.

  • Manis va-t-il devenir open-source?

    Oui, il est prévu que Manis AI devienne open-source en 2025.

Lihat lebih banyak ringkasan video

Dapatkan akses instan ke ringkasan video YouTube gratis yang didukung oleh AI!
Teks
en
Gulir Otomatis:
  • 00:00:00
    so you might have heard about Manis AI
  • 00:00:02
    this I think is fair to say is the
  • 00:00:04
    world's first autonomous AI agent I got
  • 00:00:07
    a chance to test it out and uh you need
  • 00:00:10
    to see this really fast here are some of
  • 00:00:11
    the other people that I follow online
  • 00:00:13
    that have tested it here's McKay rley
  • 00:00:15
    saying it's legitimately impressive and
  • 00:00:17
    as a reminder I don't do paid posts he's
  • 00:00:20
    kind of Blown Away by it starting to
  • 00:00:22
    freak out a little bit I may have
  • 00:00:24
    undersold this what in the world did I
  • 00:00:26
    get access to Here Without exaggeration
  • 00:00:28
    I'm genuinely being a super Earnest
  • 00:00:29
    about about this I think this experience
  • 00:00:31
    has shifted my worldview a bit that was
  • 00:00:34
    basically 80% of what I imagine
  • 00:00:36
    experiencing AGI will be like and I
  • 00:00:39
    agree with what he's saying and I do
  • 00:00:40
    have a need to tell you that no this is
  • 00:00:42
    not sponsored no one's paying me to say
  • 00:00:44
    these things this is I think the first
  • 00:00:47
    example of a fully autonomous AI agent
  • 00:00:50
    this seems way ahead of open AI of
  • 00:00:53
    anthropic I think this is what openi is
  • 00:00:56
    planning to release as their you know
  • 00:00:58
    $20,000 a month or for $10,000 or $2,000
  • 00:01:01
    month agents I think this is that
  • 00:01:04
    there's a couple things that it did that
  • 00:01:06
    I don't understand how it did that it
  • 00:01:08
    seems very smart across the board but
  • 00:01:10
    yeah I think March 7th this thing was
  • 00:01:12
    released I think that is the day we
  • 00:01:14
    entered the era of autonomous AI agents
  • 00:01:17
    so I was able to run five tests on it
  • 00:01:19
    before hitting my limit I've requested
  • 00:01:21
    an increase there's like a million
  • 00:01:23
    person waiting list so we'll see if
  • 00:01:25
    that's a possibility but take a look and
  • 00:01:27
    let me know what you think about 5
  • 00:01:28
    minutes ago I got access to to Mattis
  • 00:01:30
    let's go all right so first and foremost
  • 00:01:32
    we have standard High effort so high
  • 00:01:35
    effort mode employs a longer Chain of
  • 00:01:36
    Thought and performs more detailed tasks
  • 00:01:38
    breakdown though it requires more
  • 00:01:40
    processing time during the experimental
  • 00:01:41
    period this mode is limited to one use
  • 00:01:43
    per day all right we'll we'll save that
  • 00:01:45
    for last we'll stick with standard for
  • 00:01:47
    now so first and foremost let's uh let's
  • 00:01:50
    do some research on the AI made games
  • 00:01:52
    that are kind of going nuts you might
  • 00:01:54
    have seen fly Peter right that game that
  • 00:01:57
    was coded up with grock and CLA 2 by
  • 00:01:59
    Peter levels think last I checked it's
  • 00:02:01
    pulling in something like 100,000 per
  • 00:02:03
    month based on this fairly simple game
  • 00:02:06
    that was coded up fairly quickly using
  • 00:02:08
    AI technology so so we're going to say
  • 00:02:10
    research video games that were created
  • 00:02:11
    recently using AI for example using
  • 00:02:13
    cursor Gro 3 Cloud 3.7 Etc one example
  • 00:02:17
    is flyer.com basically whatever the
  • 00:02:19
    whatever they're doing however they're
  • 00:02:21
    making the code as long as it's
  • 00:02:22
    utilizing AI to build it that's what
  • 00:02:24
    we're looking for so so let's research
  • 00:02:26
    the top five to 10 games like that must
  • 00:02:28
    have been built in the last 3 months or
  • 00:02:30
    so get details about these games like
  • 00:02:32
    how they were built who is the creator
  • 00:02:34
    list any resources the Creator posted
  • 00:02:36
    online talking about how this was built
  • 00:02:38
    like videos X posts Etc let's say
  • 00:02:40
    YouTube videos so if they have YouTube
  • 00:02:42
    videos talking about how they built this
  • 00:02:43
    thing or posts on X about it uh we want
  • 00:02:46
    to see those then create a website that
  • 00:02:48
    showcases all of this with links to the
  • 00:02:49
    game to the Creator and a summary about
  • 00:02:51
    the most important things about the game
  • 00:02:53
    make the website have the aesthetic feel
  • 00:02:55
    of a '90s video game and away we go so
  • 00:02:59
    I've covered some this in the previous
  • 00:03:00
    video but what completely blew away my
  • 00:03:03
    mind about Manis is the way that they've
  • 00:03:06
    built it so so far I've only seen other
  • 00:03:08
    people using it this is the first time
  • 00:03:10
    that I I'm I'm using it Hands-On but as
  • 00:03:13
    far as I could tell from the previous
  • 00:03:15
    videos posted online madus itself so
  • 00:03:17
    they it has its own sort of computer its
  • 00:03:19
    own cloud computer that's running Linux
  • 00:03:22
    specifically it looks like Ubuntu and
  • 00:03:24
    it's got access to the command line
  • 00:03:26
    meaning it's kind of a super user on an
  • 00:03:28
    open source operat system which if done
  • 00:03:31
    correctly basically gives it tons of
  • 00:03:33
    power to do a lot of different stuff so
  • 00:03:36
    as you can see here the layout is a
  • 00:03:38
    little bit similar to Devon by cognition
  • 00:03:40
    AI to where you kind of have these
  • 00:03:42
    different windows and you can kind of
  • 00:03:44
    look through to see what it's doing so
  • 00:03:46
    here is a list of all of the sort of
  • 00:03:49
    websites that it's looking to and we can
  • 00:03:51
    scroll forward to see where it is in the
  • 00:03:54
    process all right so as you can see here
  • 00:03:55
    it created a to-do MD so in Linux MD is
  • 00:04:00
    a mark down file so you can kind of
  • 00:04:01
    think of it as a text editor but with
  • 00:04:03
    like a little bit more functionality so
  • 00:04:05
    it can be used for code for text Etc and
  • 00:04:07
    as you can see here it broke down the
  • 00:04:09
    tasks I gave it into subtasks we have
  • 00:04:11
    the search phase the research phase the
  • 00:04:13
    website creation Etc and for each one of
  • 00:04:15
    those it has subtasks and it shows it
  • 00:04:18
    completing the subtasks as it goes along
  • 00:04:20
    so as you can see here it's searching
  • 00:04:22
    YouTube I'm trying to find a shot of
  • 00:04:25
    where it actually is using the terminal
  • 00:04:28
    because we saw it pretty clearly in in
  • 00:04:30
    in another video that I saw online
  • 00:04:31
    interestingly we can also use mantis's
  • 00:04:34
    computer looks like we use vs code or
  • 00:04:36
    browser we'll do that later I don't want
  • 00:04:38
    to mess it up right now looks like it
  • 00:04:39
    does get blocked by Reddit so a lot of
  • 00:04:41
    the times when you're using these um you
  • 00:04:43
    know virtual machines it's not like an
  • 00:04:45
    actual desktop sitting somewhere it's
  • 00:04:46
    just a virtual machine that's kind of
  • 00:04:48
    simulating a desktop experience but um
  • 00:04:51
    as you can see here we can take control
  • 00:04:52
    so they've basically recreated operator
  • 00:04:55
    opening eyes operator they've also
  • 00:04:57
    recreated a lot of the functionality in
  • 00:05:00
    clad coder the brand new thing the
  • 00:05:02
    research preview that they've released
  • 00:05:03
    so here is the live view so is it is it
  • 00:05:07
    just sitting there watching a video I
  • 00:05:09
    wonder that's very interesting I'd be
  • 00:05:11
    surprised if it actually watches the
  • 00:05:13
    videos I don't think anything right now
  • 00:05:15
    currently actually does that most of
  • 00:05:16
    them just use the the transcript and use
  • 00:05:19
    the text in the transcript meanwhile I
  • 00:05:21
    will do another test on it so one thing
  • 00:05:23
    that I really wanted to test with this
  • 00:05:25
    is to see how well it's able to do
  • 00:05:27
    coding now currently my understanding is
  • 00:05:29
    this isn't its main thing that it's
  • 00:05:31
    supposed to be doing I don't expect good
  • 00:05:33
    results which is fine this is not going
  • 00:05:35
    to lower my opinion of it this is more
  • 00:05:37
    like a web surfing agent like a
  • 00:05:40
    knowledge worker something they can do
  • 00:05:42
    some coding like creating HTML websites
  • 00:05:45
    Etc but it's not the software engineer
  • 00:05:49
    quite yet but this is what we're going
  • 00:05:51
    to do we're going to say lots of people
  • 00:05:52
    want to get started using a Linux to
  • 00:05:54
    create things with AI let's create a
  • 00:05:56
    full online course about how to do that
  • 00:05:57
    including videos lessons links to
  • 00:06:00
    resources Etc start by creating a
  • 00:06:02
    walkthrough of how to install Ubuntu so
  • 00:06:04
    walk them through the basics of using
  • 00:06:06
    Ubuntu they'll need to install something
  • 00:06:08
    called Cloud coder I will go ahead and
  • 00:06:10
    drop the link in there just so it
  • 00:06:11
    doesn't get confused later I'm going to
  • 00:06:13
    test it specifically on how well it's
  • 00:06:14
    able to find this information online but
  • 00:06:16
    currently I'm I'm testing something
  • 00:06:18
    different I don't want it to trip up on
  • 00:06:19
    that particular thing because I'm
  • 00:06:20
    testing something else and I'm I guess
  • 00:06:22
    I'm kind of assuming that it will have
  • 00:06:24
    some weaknesses or mistakes we'll see it
  • 00:06:26
    might not but teach them the basics the
  • 00:06:28
    need to install Cloud coder and the
  • 00:06:29
    commands that they need so there's like
  • 00:06:31
    just a handful of commands that you have
  • 00:06:32
    to be able to run in order to install
  • 00:06:35
    clot coder so we don't want to like
  • 00:06:36
    overwhelmed in the person but at the
  • 00:06:39
    same time they need like three or four
  • 00:06:41
    different things that they have to type
  • 00:06:42
    in in order to proceed assume little to
  • 00:06:45
    no Linux experience prior to this then
  • 00:06:47
    explain what GitHub is and how to use it
  • 00:06:49
    and then find one simple project that
  • 00:06:51
    you can try on GitHub by using Cloud
  • 00:06:52
    coder to clone the repo this will be a
  • 00:06:54
    lesson that demonstrates how to use
  • 00:06:55
    GitHub and Cloud coder in conjunction
  • 00:06:58
    okay so this is where it's sort of like
  • 00:07:00
    generates the lesson plan and Next Step
  • 00:07:02
    will be actually creating the website
  • 00:07:05
    with the lesson plan make it in dark
  • 00:07:06
    mode so create the website that has all
  • 00:07:07
    the lessons and walkthroughs make it in
  • 00:07:09
    dark mode with a black and yellow theme
  • 00:07:11
    all right so we're going to click go so
  • 00:07:13
    for the people that may be not familiar
  • 00:07:15
    with what I'm talking about here so
  • 00:07:16
    basically a lot of the stuff that gets
  • 00:07:18
    released in the sort of the AI space
  • 00:07:21
    often gets released on Linux and apple
  • 00:07:23
    first because they are sort of similar
  • 00:07:26
    Apple's OS is sort of built on the same
  • 00:07:28
    sort of foundation as Linux and so if
  • 00:07:30
    you're a Windows user very often you're
  • 00:07:33
    going to be waiting for the updates or
  • 00:07:35
    stuff is not going to work as intended
  • 00:07:37
    Cloud coder I could not get it to run on
  • 00:07:39
    Windows and I do believe more and more
  • 00:07:41
    people get interested in using Linux
  • 00:07:43
    especially since things like this run on
  • 00:07:45
    Linux I think a lot more things will
  • 00:07:46
    start running on Linux things like
  • 00:07:48
    Ubuntu Etc Ubuntu is sort of a a flavor
  • 00:07:50
    a distribution of Linux kind of a
  • 00:07:52
    version of it customized for certain
  • 00:07:54
    specific things and back in the days you
  • 00:07:56
    know 5 10 years ago it was fairly
  • 00:07:59
    difficult at least for people that don't
  • 00:08:00
    have a lot of experience with it it had
  • 00:08:02
    a bit of a learning curve now not only
  • 00:08:04
    is it getting a lot better to use
  • 00:08:05
    because of all the AI tools that you can
  • 00:08:07
    use with it but also it's getting a lot
  • 00:08:09
    easier to learn because the AI can
  • 00:08:12
    assist you in using it if you have Linux
  • 00:08:14
    open you know the terminal in one sort
  • 00:08:16
    of window and then something like Chad
  • 00:08:18
    PT in another Chad PT will just walk you
  • 00:08:20
    through all the commands and exactly
  • 00:08:21
    what to do and it's pretty good at doing
  • 00:08:23
    so so I think this is going to be I'm
  • 00:08:25
    kind of calling it now I think there's
  • 00:08:27
    going to be a lot more interesting stuff
  • 00:08:28
    like this so what I'm trying to do here
  • 00:08:30
    is to create a beginner friendly guide
  • 00:08:33
    to how to get into it right so and and
  • 00:08:35
    it's going to cover a lot of you know
  • 00:08:37
    before you had things like Chad gbt to
  • 00:08:39
    help you out doing this would be you
  • 00:08:41
    know fairly difficult for a beginner you
  • 00:08:44
    know figuring out how GitHub Works
  • 00:08:45
    figure out how Cloud coder Works how to
  • 00:08:47
    use the command line all these things
  • 00:08:49
    would have a steep learning curve but as
  • 00:08:52
    you hopefully you'll see now it's it's a
  • 00:08:54
    lot easier to get in there and start
  • 00:08:56
    building and start tweaking things and
  • 00:08:57
    figuring out how they work which I think
  • 00:08:59
    is going to be a lot more interesting
  • 00:09:00
    for people so let's leave that to do its
  • 00:09:02
    thing oh and here we can see Manis using
  • 00:09:04
    the terminal okay so we're still able to
  • 00:09:06
    see it so let me move this over here so
  • 00:09:08
    this is Manis using the Ubuntu terminal
  • 00:09:10
    right so as you can see here it's
  • 00:09:11
    actually typing in you know change
  • 00:09:13
    directory go to this directory and so it
  • 00:09:15
    usually what it does when you give a
  • 00:09:17
    task a project it starts by creating a
  • 00:09:19
    new directory that that it's going to
  • 00:09:21
    work on so each project it's own
  • 00:09:23
    directory and whatever assets whatever
  • 00:09:26
    files it creates to complete that
  • 00:09:28
    project go into that directory so this
  • 00:09:30
    is very similar to Cloud coder and as
  • 00:09:32
    you can see here on the previous page
  • 00:09:34
    actually I can probably go back so as
  • 00:09:35
    you can see here so it needs Cloud flare
  • 00:09:38
    workers so this is likely for creating
  • 00:09:41
    the website that we asked it for so as
  • 00:09:43
    you can see here it's initializing a get
  • 00:09:45
    repository so GitHub is a place where a
  • 00:09:48
    lot of people store code that anybody
  • 00:09:50
    can use right so I can go there and if
  • 00:09:52
    somebody created a piece of software
  • 00:09:54
    that does something I can just clone
  • 00:09:56
    that to my computer and then I can use
  • 00:09:57
    it or work on it to upgrade in some ways
  • 00:10:00
    I can either Fork it meaning I can kind
  • 00:10:02
    of create a different version and take
  • 00:10:03
    it into sort of whatever Direction I
  • 00:10:05
    want to or contribute to the original
  • 00:10:08
    it's kind of this open source ecosystem
  • 00:10:09
    so here it's using that to I'm assuming
  • 00:10:13
    this is for building the website so
  • 00:10:14
    looks like it's using uh it's talking
  • 00:10:16
    about database config it's uh using a
  • 00:10:19
    SQL file to reset the D database or
  • 00:10:22
    rather this is probably some
  • 00:10:23
    documentation that it's reading about
  • 00:10:24
    how to do that and so if we go live here
  • 00:10:27
    is the layout. TSX which includes the
  • 00:10:31
    HTML text for the website that we've
  • 00:10:33
    asked for noex is adding all the you
  • 00:10:35
    know the copyright and all that stuff in
  • 00:10:37
    the footer so it kind of like knows how
  • 00:10:39
    to do that stuff this is the cool thing
  • 00:10:41
    about kind of the um the agents like
  • 00:10:43
    this because you know when you're
  • 00:10:44
    dealing with a chatbot you know there's
  • 00:10:46
    a certain context window so it it tries
  • 00:10:48
    to shorten things up right so if you go
  • 00:10:50
    a project that's massive they can't run
  • 00:10:52
    it in one thing by creating these kind
  • 00:10:53
    of agentic structures you kind of create
  • 00:10:55
    these like Loops that it's able to do
  • 00:10:57
    like now do this now do this now do it
  • 00:10:59
    now do this right and you can do one
  • 00:11:01
    thing at a time and here it explains the
  • 00:11:04
    Showcase right so this is the AI game
  • 00:11:06
    development thing so it's saying this
  • 00:11:07
    website highlights the incredible
  • 00:11:09
    progress in AI assistant game
  • 00:11:10
    development all these games were built
  • 00:11:12
    in the last 3 months so this is we're
  • 00:11:14
    beginning to see the power of these AI
  • 00:11:15
    agents right so you kind of give it this
  • 00:11:17
    top level goal and it has enough
  • 00:11:19
    intelligence to not only figure out how
  • 00:11:21
    to break it into little pieces but when
  • 00:11:22
    it gets to for example you know creating
  • 00:11:24
    the website knows okay like I need to
  • 00:11:26
    describe what this website is right
  • 00:11:28
    because it has to appear in for example
  • 00:11:31
    Google you know search results plus the
  • 00:11:33
    people that are looking at the website
  • 00:11:34
    have to know what it is so it figures
  • 00:11:36
    out how to describe it based on the
  • 00:11:38
    description I gave to it right so it
  • 00:11:40
    takes kind of the words that I've used
  • 00:11:42
    translates them into the description of
  • 00:11:43
    the website it's it's little things like
  • 00:11:46
    that like normally you would have to
  • 00:11:47
    have a human being that sat there and
  • 00:11:49
    thought about that and rewrote it right
  • 00:11:51
    there's a little bit of a cognitive load
  • 00:11:53
    involved with that but it's lowlevel
  • 00:11:55
    it's you know kind of boring if that's
  • 00:11:57
    all you had to do it's not high LEL
  • 00:11:59
    tasks the huge thing about AI agents
  • 00:12:02
    like Manis is that a lot of that stuff
  • 00:12:04
    it can do for you right and then if if
  • 00:12:06
    it's off if you want to improve it you
  • 00:12:08
    can step in and manually adjust things
  • 00:12:10
    or just tell it to try again or specify
  • 00:12:13
    how it should do it Etc so we have a
  • 00:12:14
    space shooter by Alex Finn Frogger with
  • 00:12:17
    rocket launch by Micah flight simulator
  • 00:12:19
    by Peter levels shooter game by Jagger
  • 00:12:21
    Barda and a 3D seen by AI oriented Dev I
  • 00:12:25
    think I I'm familiar with like two or
  • 00:12:27
    three of those so so as you see here
  • 00:12:29
    it's creating a kind of a development
  • 00:12:32
    environment to run to render the website
  • 00:12:35
    locally so we'll come back to this but
  • 00:12:37
    so far this little robot is doing a
  • 00:12:39
    great job very very proud so looks like
  • 00:12:42
    going back to the Linux AI course it's
  • 00:12:45
    now on GitHub and it's checking out okay
  • 00:12:48
    so it found a python artificial
  • 00:12:49
    intelligence projects for beginners
  • 00:12:52
    which is probably the thing that it's
  • 00:12:53
    going to use as its demonstration
  • 00:12:55
    project wow so for a second there if you
  • 00:12:58
    saw that let me see if I can go back so
  • 00:13:00
    for a second there I can't quite find
  • 00:13:01
    that frame um but for a second there you
  • 00:13:03
    could see the the bounding boxes around
  • 00:13:06
    kind of all the usable objects on screen
  • 00:13:09
    so that's how it seems like it's like
  • 00:13:11
    able to navigate stuff like that which
  • 00:13:13
    is interesting so that's not quite how
  • 00:13:15
    the open aai operator does it the open
  • 00:13:17
    aai operator if I understand it actually
  • 00:13:20
    uses like real Vision uh the the vision
  • 00:13:23
    model looking at the screen and then
  • 00:13:24
    moving the mouse to click on it this
  • 00:13:26
    still has some sort of a like it it
  • 00:13:27
    finds a different objects on there and
  • 00:13:29
    then goes to click on them and there are
  • 00:13:31
    sort of pros and cons to both from what
  • 00:13:33
    I've seen kind of the advantages to
  • 00:13:34
    using something like this is you're
  • 00:13:36
    going to be much faster and much more
  • 00:13:38
    accurate within the things that that
  • 00:13:41
    approach can do but there's going to be
  • 00:13:43
    some like weird use cases where it it
  • 00:13:45
    doesn't quite work like that right so if
  • 00:13:47
    you if it's having a hard time defining
  • 00:13:49
    certain little things on screen as
  • 00:13:50
    objects it might miss those but for most
  • 00:13:52
    of the stuff online this is going to be
  • 00:13:54
    the faster more accurate just easier
  • 00:13:56
    approach so as you can see here you know
  • 00:13:58
    one of the course is that it walks you
  • 00:14:00
    through is setting up your environment
  • 00:14:02
    so if you're going to code of python
  • 00:14:04
    it's very good to use certain virtual
  • 00:14:05
    environments to make sure that kind of
  • 00:14:07
    everything works well within that
  • 00:14:08
    environment where you have all the tools
  • 00:14:10
    that you need the correct version right
  • 00:14:12
    it's just a way to manage the complexity
  • 00:14:14
    a little bit so they're walking you
  • 00:14:16
    through exactly how to do that so you
  • 00:14:17
    have your sort of your terminal bash you
  • 00:14:20
    create a directory you navigate to that
  • 00:14:22
    directory you create a virtual
  • 00:14:24
    environment you activate the virtual
  • 00:14:26
    environment and then within that
  • 00:14:27
    environment you install the required
  • 00:14:29
    packages this looks terrific so far then
  • 00:14:33
    it walks you through how to use cloth
  • 00:14:35
    coder to clone the the repo this this
  • 00:14:37
    looks great I got to say so far I'm very
  • 00:14:39
    very impressed all right but let's test
  • 00:14:41
    its coding abilities is it a coder so
  • 00:14:44
    we're going to create a 3Gs game where
  • 00:14:45
    you control a plane make the plane a
  • 00:14:48
    World War II fighter plane create
  • 00:14:50
    several environments to fly over like
  • 00:14:52
    ocean Forest City at a day and night
  • 00:14:56
    cycle create an attack function that
  • 00:14:57
    fires a machine gun and we're going to
  • 00:14:59
    say add lots of visual effects to make
  • 00:15:02
    it realistic smoke lens flares Trace
  • 00:15:05
    rounds for the machine gun clouds Etc so
  • 00:15:07
    let's start there and see where this
  • 00:15:09
    takes us if it Nails this that I got to
  • 00:15:11
    say would be fairly impressive so
  • 00:15:13
    there's this new thing so it says
  • 00:15:15
    knowledge suggestions when pending so
  • 00:15:17
    let's see what this is okay so this is a
  • 00:15:20
    way to store like permanent knowledge or
  • 00:15:22
    memory as open AI cause it um so when I
  • 00:15:25
    ask for a 3js WW2 fighter plane it's
  • 00:15:29
    saying should we just assume you want
  • 00:15:31
    all of those things in there which is
  • 00:15:33
    pretty cool not quite there yet because
  • 00:15:37
    um I want to see how it how it works on
  • 00:15:39
    this first attempt and then later maybe
  • 00:15:40
    we'll add that as in addition to every
  • 00:15:42
    single prompt this is going to be very
  • 00:15:44
    effective if you're doing the same thing
  • 00:15:46
    over and over again so but we're going
  • 00:15:47
    to reject it for now all right but that
  • 00:15:48
    was a child's play well not really it's
  • 00:15:51
    it's kind of complicated but let's see
  • 00:15:52
    if we can uh crank up the difficulty
  • 00:15:55
    just a little bit we're going to ask it
  • 00:15:56
    to do this so we're going to create a
  • 00:15:58
    tool that takes a prompt like a woman is
  • 00:16:00
    talking about how amazing this hamster
  • 00:16:02
    cage product is and an llm of my choice
  • 00:16:05
    writes out a script for that then that
  • 00:16:07
    script gets voiced over in 11 Labs with
  • 00:16:10
    the voice of my choosing then that voice
  • 00:16:12
    over is fed into haen to create a video
  • 00:16:15
    Avatar that says the thing that says
  • 00:16:17
    that script and use the apis for
  • 00:16:20
    everything I will provide API keys and
  • 00:16:23
    by the way keep in mind that if it fails
  • 00:16:26
    at any one of these then we can always
  • 00:16:28
    bump it into that extended thinking mode
  • 00:16:30
    so once we figure out like what where
  • 00:16:31
    the limit is we'll we'll test that out
  • 00:16:33
    as well all right so here we got a
  • 00:16:35
    notification Manis is attempting to
  • 00:16:37
    deploy a service located at you know
  • 00:16:39
    whatever so basically it's asking me if
  • 00:16:41
    it's okay to deploy the application to
  • 00:16:43
    the public network yeah why not
  • 00:16:45
    unleashing this egi onto the internet
  • 00:16:48
    what could possibly go wrong all right
  • 00:16:50
    so here we have so it's hitting an M
  • 00:16:53
    error which is a which is a package
  • 00:16:56
    manager for for JavaScript but as you
  • 00:16:59
    can see here so when it encounters an
  • 00:17:00
    error it just completes it it it keeps
  • 00:17:04
    troubleshooting and figuring out how to
  • 00:17:05
    get around it Etc creating optimized
  • 00:17:07
    production build so not that long ago if
  • 00:17:10
    you recall we were sitting here kind of
  • 00:17:13
    like dreaming anticipating the day then
  • 00:17:16
    something like this would exist where
  • 00:17:17
    you tell it to do something and it just
  • 00:17:19
    goes off because I mean some of these
  • 00:17:20
    things have been running for 20 minutes
  • 00:17:22
    now right so for 20 minutes from one
  • 00:17:25
    prompt this thing is sitting there and
  • 00:17:27
    just thinking and work and doing and
  • 00:17:30
    troubleshooting optimizing with almost
  • 00:17:33
    zero input from me now if it's able to
  • 00:17:36
    create these World War II fighter
  • 00:17:38
    playing games by the way look at this
  • 00:17:41
    this subtask right list this this game
  • 00:17:43
    development right look at how many
  • 00:17:45
    subtasks it created right environments
  • 00:17:48
    ocean Forest City environment switching
  • 00:17:50
    mechanism day night cycle machine like
  • 00:17:52
    like look at that so we're about I don't
  • 00:17:54
    know 20 minutes in maybe and we're less
  • 00:17:56
    than a third of the way maybe a third of
  • 00:17:58
    the way into it completing all the
  • 00:18:00
    things by the way really fast so if it
  • 00:18:02
    can do something like this and keep on
  • 00:18:04
    adding various functionality to it in an
  • 00:18:07
    effective manner why is that important
  • 00:18:09
    right is it just that oh now you have a
  • 00:18:10
    little shooting plane game that you can
  • 00:18:12
    mess around with well you know again
  • 00:18:14
    this is the fly Peter game that he
  • 00:18:16
    created very very quickly so you know
  • 00:18:18
    you can have the various different
  • 00:18:20
    things you can purchase different planes
  • 00:18:22
    like the F16 but we'll just start with
  • 00:18:24
    the Cessna all right so you know I I'll
  • 00:18:26
    have the sound off so it doesn't get too
  • 00:18:27
    annoying but you know this this game is
  • 00:18:30
    multiplayer so you're able to fly around
  • 00:18:32
    with a lot of different people all over
  • 00:18:34
    the world and as you can see here like
  • 00:18:36
    right this blimp says a synth flow on it
  • 00:18:38
    Etc freelancer whatever right so these
  • 00:18:41
    are people that chose to pay money to
  • 00:18:43
    advertise within this game and this game
  • 00:18:46
    is pulling in quite a bit of money from
  • 00:18:49
    these advertisements even if I see it
  • 00:18:51
    correctly an advertisement on the moon
  • 00:18:54
    can't quite tell what it is but it's
  • 00:18:55
    there right and that game rapidly scaled
  • 00:18:57
    up to where it's making 50,000 a month
  • 00:19:00
    and I even been seeing uh potentially
  • 00:19:02
    got up to 100,000 a month in monthly
  • 00:19:05
    recurring revenue or maybe it's not
  • 00:19:06
    recurring if it's like a sponsorship
  • 00:19:08
    placements I'm not sure but the point is
  • 00:19:10
    it's it's a lot from something that was
  • 00:19:12
    created very quickly so initially it was
  • 00:19:14
    made in just 30 minutes with AI tools so
  • 00:19:16
    it kind of gives you a glimpse into how
  • 00:19:18
    quickly these things are proceeding
  • 00:19:19
    because when you know claw 3.7 grock 3
  • 00:19:22
    when they came out it really
  • 00:19:23
    supercharged developers to create a lot
  • 00:19:26
    of stuff things that could be scaled up
  • 00:19:29
    very very quickly to make significant
  • 00:19:31
    amounts of money so people used
  • 00:19:33
    something like cursor which is a
  • 00:19:34
    development environment that's
  • 00:19:36
    integrated with these large language
  • 00:19:38
    models like Claude or 03 mini High
  • 00:19:41
    whichever one you prefer but you still
  • 00:19:43
    needed to know some things in order to
  • 00:19:46
    kind of use it you still have to have
  • 00:19:47
    some technical background with this and
  • 00:19:50
    again we have to see how well it's going
  • 00:19:51
    to complete these tasks but it does seem
  • 00:19:54
    more and more that these large language
  • 00:19:56
    models these AI tools they're kind of
  • 00:19:57
    separating the human from like the code
  • 00:20:00
    the command line because you know in the
  • 00:20:03
    very beginning a long long time ago
  • 00:20:05
    everything had to be written out by hand
  • 00:20:07
    right so each word each letter of the of
  • 00:20:10
    the code had to be done by a human each
  • 00:20:12
    command line had to be written out you
  • 00:20:14
    know on a keyboard Etc and now as you
  • 00:20:16
    can see here this AI tool is running the
  • 00:20:19
    commands in the command line it's
  • 00:20:21
    creating the documents is populating it
  • 00:20:23
    with code so far I didn't do anything I
  • 00:20:26
    didn't write one bit of code I wrote
  • 00:20:29
    everything out in natural language in
  • 00:20:31
    English and the important thing to keep
  • 00:20:33
    in mind is so I don't know how well this
  • 00:20:35
    thing is going to come out like the
  • 00:20:36
    coding how well it's going to come out
  • 00:20:38
    but here's the thing that the large
  • 00:20:41
    language models that are being used by
  • 00:20:42
    this I don't know exactly what they're
  • 00:20:43
    using but they're saying that they've
  • 00:20:45
    been fine-tuned so they took some
  • 00:20:46
    probably open- Source large language
  • 00:20:48
    model they've fine-tuned it to be good
  • 00:20:51
    at you know being Manis Etc and that's
  • 00:20:54
    what's being used and they said that
  • 00:20:56
    they're going to open source this thing
  • 00:20:58
    later this here 2025 so think about that
  • 00:21:00
    for a second so in 2025 let's say you
  • 00:21:03
    have this entire sort of scaffolding
  • 00:21:05
    right the the what Manis is so if you
  • 00:21:07
    think of this thing as scaffolding plus
  • 00:21:10
    a large language model that's kind of
  • 00:21:13
    what this is by scaffolding I mean the
  • 00:21:15
    thing that's you know kind of like that
  • 00:21:16
    kind of says okay so first thing you
  • 00:21:18
    know create these subtasks and then run
  • 00:21:20
    these commands right so it kind of like
  • 00:21:21
    turns this into an agent right allows it
  • 00:21:24
    to surf the web Etc but I think the
  • 00:21:26
    important thing to understand here is
  • 00:21:27
    that if it is open source then and and
  • 00:21:31
    the whatever large language models that
  • 00:21:32
    they're using under the hood isn't the
  • 00:21:34
    top of the line they can very simply
  • 00:21:37
    switch in the top of the line right you
  • 00:21:39
    can put in Cloud 3.7 Sonet or grock 3 or
  • 00:21:42
    03 mini high or any time the next big
  • 00:21:45
    thing comes out you just put it in there
  • 00:21:47
    I'm curious to know if they're using
  • 00:21:49
    deep seek for this that would certainly
  • 00:21:51
    make sense all right so here it's saying
  • 00:21:53
    it's waiting for user all right so this
  • 00:21:55
    is Manis it says thank you for your
  • 00:21:57
    patience a very very polite I like that
  • 00:22:00
    okay so it says we've completed the
  • 00:22:01
    Ubuntu installation Basics Etc and it's
  • 00:22:04
    saying should I just continue working
  • 00:22:06
    making the adjustments or show you a
  • 00:22:07
    preview I'm going to say preview
  • 00:22:09
    actually now that I'm thinking about it
  • 00:22:10
    I'm going to say preview please this
  • 00:22:12
    thing is a little bit too advanced I I
  • 00:22:14
    feel like I should be polite back
  • 00:22:15
    preview please so I going to deploy a
  • 00:22:17
    temporary version so you can see how it
  • 00:22:19
    looks with the Dark theme using black
  • 00:22:21
    and yellow colors meanwhile our AI
  • 00:22:23
    generate video games so it's also asking
  • 00:22:26
    if it should deploy to the public
  • 00:22:27
    network let's say go for it and our
  • 00:22:29
    automated video creation with llm and
  • 00:22:31
    apis it's asking for the API Keys now
  • 00:22:35
    interestingly in anthropics and Cloud
  • 00:22:37
    coder when it's time to get to the point
  • 00:22:39
    where you know you give up your API keys
  • 00:22:41
    to the thing I mean there's obviously
  • 00:22:43
    security issues with that so Cloud coder
  • 00:22:45
    refuses to do it but then if you just
  • 00:22:47
    you know you copy and paste your uh API
  • 00:22:49
    key in there it kind of like it like
  • 00:22:51
    lectures you about it kind of grumbles
  • 00:22:53
    about it's like well you really
  • 00:22:54
    shouldn't do that young young sir but it
  • 00:22:56
    it does still take the API keys and put
  • 00:22:58
    it where it's supposed to go again you
  • 00:23:00
    know don't do this or at least be aware
  • 00:23:03
    of the risks of doing this I'm willing
  • 00:23:05
    to be your giddy pig to test this thing
  • 00:23:07
    out but you know there are security
  • 00:23:09
    issues with just doing something like
  • 00:23:11
    this all right so I added all of my API
  • 00:23:13
    keys this is one of those things where
  • 00:23:15
    you have to be careful what you're doing
  • 00:23:16
    I plan to delete those API keys after
  • 00:23:19
    testing this thing you know and before
  • 00:23:20
    uploading this video but I've created
  • 00:23:23
    unique API keys for testing this thing
  • 00:23:26
    and later if I wanted to use it long
  • 00:23:28
    term I would actually have to go in
  • 00:23:30
    myself and add those keys in a secure
  • 00:23:32
    way but let's take a look at our AI
  • 00:23:35
    generated video game research preview so
  • 00:23:37
    we're going to click to view and wow it
  • 00:23:41
    is glorious absolutely glorious oh wow
  • 00:23:45
    and it's got the little scrolling text
  • 00:23:47
    absolutely incredible look at that games
  • 00:23:49
    created with AI tools like cursor Gro 3
  • 00:23:51
    and Cloud 3.7 in the last 3 months so it
  • 00:23:53
    kind of literally copied what I said but
  • 00:23:55
    that's okay no coding experience
  • 00:23:57
    required the future of game development
  • 00:23:59
    is here okay and it added those this is
  • 00:24:01
    I mean this is good so far so good right
  • 00:24:04
    wow all right recently created with AI
  • 00:24:06
    we got the space shooter game by Alex
  • 00:24:08
    Finn it's got oh and it like has a
  • 00:24:11
    little popup thing wow okay this is very
  • 00:24:14
    good and look at that it writes out uh
  • 00:24:16
    how it was created we can click on more
  • 00:24:19
    info and it gives us more info I wow wow
  • 00:24:23
    wow wow used grock's deep research for
  • 00:24:25
    research and planning used grock's think
  • 00:24:28
    or Cloud 3.7 to execute the code
  • 00:24:30
    deployed via GitHub all the resources it
  • 00:24:34
    it gives how many followers this person
  • 00:24:36
    has look at that it's got the YouTube
  • 00:24:38
    channel a newsletter a website notable
  • 00:24:41
    results it got multiple Elon Musk
  • 00:24:43
    retweets very very cool but okay but
  • 00:24:45
    where's the game dude okay so it doesn't
  • 00:24:46
    have a link to the game but but not
  • 00:24:49
    everybody hosts their game online but I
  • 00:24:51
    know Peter fly Peter does so let's see
  • 00:24:54
    if it has the oh okay so it has a play
  • 00:24:56
    game and more info okay and this this
  • 00:24:59
    winter 3D game okay so if I click play
  • 00:25:01
    game okay it takes me to this thing and
  • 00:25:03
    if I click play game it takes me to the
  • 00:25:06
    game this is terrific okay so that's
  • 00:25:08
    that's good okay okay so this is the
  • 00:25:10
    game you can click to move the bear
  • 00:25:12
    around zoom in and out okay very very
  • 00:25:14
    cool all right and it provides the AI
  • 00:25:16
    tools used as well as the Creator
  • 00:25:19
    resources it tells you how to recreate
  • 00:25:21
    this thing so basically what I was
  • 00:25:23
    trying to do with this uh particular
  • 00:25:24
    prompt is you know right now people are
  • 00:25:27
    super interested in all these games
  • 00:25:28
    people want to learn how to create them
  • 00:25:30
    people want to see who's creating what
  • 00:25:33
    and it would be kind of cool to have
  • 00:25:34
    like one resource one website that has
  • 00:25:37
    you know an Ever updating list of all
  • 00:25:40
    these games with any resource that we
  • 00:25:43
    know about how to create these games and
  • 00:25:45
    if you wanted to recreate it how would
  • 00:25:47
    you go about recreating that now
  • 00:25:49
    obviously you know if you were doing it
  • 00:25:51
    yourself that would be kind of a pain in
  • 00:25:53
    the butt right it would take a lot of
  • 00:25:55
    time a lot of research but as you can
  • 00:25:57
    see here this was all done with you know
  • 00:25:59
    me being AFK basically right I give it a
  • 00:26:02
    prompt I go away now the next step would
  • 00:26:04
    be to see how well we're able to add on
  • 00:26:06
    to this right so can we include more
  • 00:26:08
    resources can we include more games I
  • 00:26:11
    mean you can imagine just having that
  • 00:26:13
    thing kind of run every 24 hours or
  • 00:26:15
    whatever updating this database uh in
  • 00:26:18
    real time basically creating a perfect
  • 00:26:21
    sort of ever updated resource for people
  • 00:26:24
    looking for this stuff I I'm just I'm
  • 00:26:26
    blown away I am absolutely Blown Away
  • 00:26:28
    okay I mean this is A+ right this is
  • 00:26:31
    kind of mindblowing from one prompt keep
  • 00:26:33
    in mind we're going to test sort of
  • 00:26:35
    iterating on this but so far wow okay so
  • 00:26:38
    it presented the final deliverables to
  • 00:26:40
    user okay we're going to come back to
  • 00:26:43
    this but wow all right so our buun
  • 00:26:45
    course is ready to be deployed to the
  • 00:26:46
    public network go for it did we not say
  • 00:26:48
    that already not sure oh okay so it's
  • 00:26:50
    this is the preview it's asking us do we
  • 00:26:52
    want to continue working it or do we
  • 00:26:54
    want to get the preview and the WW2
  • 00:26:57
    fighter plane still working automated
  • 00:26:59
    video creation still working and so it
  • 00:27:02
    looks like there's a Manis dospace where
  • 00:27:05
    it hosts these things for its user so it
  • 00:27:07
    kind of generates a random username for
  • 00:27:08
    you and then creates it so mine is Gizzy
  • 00:27:11
    yazel Manis dospace so I'll post this
  • 00:27:14
    link down below if you wanted to check
  • 00:27:15
    it out this this is throwing me a little
  • 00:27:18
    bit here this this feels like a little
  • 00:27:19
    bit too good you know what I'm going to
  • 00:27:21
    start a new session here and uh I'm
  • 00:27:23
    going to say research everything you can
  • 00:27:25
    about Manis AI what llm are they using
  • 00:27:27
    for it who's behind it what Vision model
  • 00:27:29
    are they using Etc anything and
  • 00:27:30
    everything about how this thing is made
  • 00:27:32
    let's see where that takes us all right
  • 00:27:34
    and it looks like our World War 2
  • 00:27:36
    fighter plane game is ready to be tested
  • 00:27:38
    out it has been submitted for our
  • 00:27:40
    approval all right so we're getting a
  • 00:27:42
    little uh error so we got an error
  • 00:27:44
    message I'm just going to post it right
  • 00:27:45
    back in there you know that Meme where
  • 00:27:47
    Fry from Futurama is like Shut Up And
  • 00:27:49
    Take My Money uh this might be one of
  • 00:27:52
    those products where it's like I feel
  • 00:27:54
    like I need a subscription to this I
  • 00:27:55
    want to be testing out like the extended
  • 00:27:57
    version of it this this thing is kind of
  • 00:27:59
    Bonkers all right let's do one more
  • 00:28:01
    thing so all right let's say I invested
  • 00:28:02
    $1,000 each into the following stocks
  • 00:28:05
    Nvidia Microsoft Google Tesla Amazon
  • 00:28:07
    meta on January 01 2020 so create a cool
  • 00:28:10
    looking display of my investments
  • 00:28:11
    include a way to customize the view to
  • 00:28:13
    see the data in detail so let's see how
  • 00:28:16
    it sort of interprets that I literally
  • 00:28:18
    was just saying how like just you know
  • 00:28:20
    shut up and take my money I want to keep
  • 00:28:22
    using it I ran out of my standard but I
  • 00:28:24
    still have sort of one high effort task
  • 00:28:27
    remaining and and ideally I'm going to
  • 00:28:28
    use that for whatever it fails to do or
  • 00:28:31
    for whatever it sort of like struggles
  • 00:28:33
    with here is our fighter plane jet
  • 00:28:37
    fighting thing and it's not showing up
  • 00:28:40
    okay so right off the bat it doesn't
  • 00:28:42
    seem to be functioning so I load a
  • 00:28:44
    screenshot and I said the heads up
  • 00:28:45
    display shows up but nothing else so I
  • 00:28:48
    think it should be able to figure that
  • 00:28:49
    out and so it's on it so it looks like
  • 00:28:51
    we can still continue going back and
  • 00:28:53
    forth with the previous projects we just
  • 00:28:56
    can't start new ones interestingly it
  • 00:28:58
    has these little sort of uh icons
  • 00:29:00
    displaying so this one's like stocks or
  • 00:29:02
    charts right this one's gaming so it's
  • 00:29:03
    pretty good at at naming and Visually
  • 00:29:05
    kind of referencing these so it's a
  • 00:29:07
    little bit easier to find all right I'll
  • 00:29:08
    go ahead and delete this one for the
  • 00:29:09
    time being and here's our Linux AI
  • 00:29:12
    development course that's looking pretty
  • 00:29:13
    neat so we might mess around to change
  • 00:29:16
    the visuals but so far so good so you
  • 00:29:19
    can start with start learning so it
  • 00:29:21
    walks you through understanding Ubuntu
  • 00:29:23
    what you need to do before you begin
  • 00:29:25
    backing up your data Etc it's got you
  • 00:29:28
    links to where you need to go creating a
  • 00:29:30
    bootable USB drive with link to rofus
  • 00:29:34
    phenomenal it seems perfect I'm not
  • 00:29:36
    going line by line but man yeah
  • 00:29:38
    everything is just like it's snailing
  • 00:29:39
    all the points uh the various you know
  • 00:29:41
    stuff you need to to get started just
  • 00:29:43
    the basics right graphic drivers Etc
  • 00:29:47
    walks you through the terminal how to
  • 00:29:48
    start the terminal gives you the
  • 00:29:50
    keyboard shortcut perfect how to
  • 00:29:52
    navigate through the terminal this is
  • 00:29:55
    terrific I mean I guess that's not that
  • 00:29:56
    impressive I mean cuz this is available
  • 00:29:58
    in like 50 million different places
  • 00:29:59
    online so this this shouldn't be that
  • 00:30:01
    impressive but notice that it did do a
  • 00:30:03
    very good job what I wanted what I
  • 00:30:05
    described because in this one I gave it
  • 00:30:07
    very specifically here's what I want it
  • 00:30:09
    did that I would say flawlessly yeah
  • 00:30:12
    it's going through node.js and then how
  • 00:30:14
    to install clot coder if I was going to
  • 00:30:18
    nitpick just a little bit this does not
  • 00:30:21
    belong here so this is they've copied
  • 00:30:23
    this from the installation page on
  • 00:30:25
    anthropic and saying if you're using
  • 00:30:27
    Windows subsist systems so yeah I mean
  • 00:30:29
    that's an important thing to know but in
  • 00:30:31
    the context of this thing we're not so
  • 00:30:32
    this is literally how to install Linux
  • 00:30:35
    um and then use it so but again I can't
  • 00:30:37
    say this is wrong it's just maybe not
  • 00:30:39
    needed probably not needed but otherwise
  • 00:30:42
    terrific then we have a a brief GitHub
  • 00:30:44
    tutorial and a project demo okay so this
  • 00:30:48
    is where I was curious if I was going to
  • 00:30:50
    drop the ball or not so basically what
  • 00:30:51
    we're doing is we are asking you to do
  • 00:30:54
    something that the documentation for
  • 00:30:56
    which isn't really available online
  • 00:30:59
    because you know normally if you want to
  • 00:31:01
    install a project from GitHub there's
  • 00:31:04
    documentation on that if you want to
  • 00:31:06
    install clot coder there's documentation
  • 00:31:07
    on that but if you wanted to use clot
  • 00:31:11
    coder to have clot coder install the or
  • 00:31:14
    you know clone the GitHub repository
  • 00:31:16
    there's not going to be a lot of
  • 00:31:17
    information on that it's kind of a brand
  • 00:31:18
    new thing so I was curious to see if it
  • 00:31:21
    would figure it out this to me shows
  • 00:31:24
    that whatever model they're using under
  • 00:31:27
    the hood is is smart is powerful this
  • 00:31:30
    isn't some half-baked model they're
  • 00:31:32
    using it figured it out based on context
  • 00:31:35
    cuz again there's not a lot of like data
  • 00:31:37
    out there on this so to figured out that
  • 00:31:39
    in order to clone the repository with
  • 00:31:42
    Cloud coder you don't run the commands
  • 00:31:44
    yourself you tell CLA coder I I want to
  • 00:31:47
    clone this project from this person or
  • 00:31:50
    whatever on GitHub so I hope you
  • 00:31:52
    understand kind of what I'm saying here
  • 00:31:52
    like this this this thing that it did
  • 00:31:54
    right here this whole thing this took
  • 00:31:57
    intelligence took learning this is not
  • 00:31:59
    available online there's not a lot of
  • 00:32:01
    resource that spell this out this is
  • 00:32:03
    kind of crazy I got to say I'm curious
  • 00:32:05
    what they're using what model they're
  • 00:32:07
    using for this then they're saying clot
  • 00:32:08
    coder will help you with the process
  • 00:32:09
    suggesting the appropriate git command
  • 00:32:12
    how do they know this as far as you can
  • 00:32:14
    tell it it went it read read the manual
  • 00:32:16
    it read anthropics manual this is way
  • 00:32:20
    better than than than I would have
  • 00:32:21
    expected I am kind of Blown Away this
  • 00:32:23
    thing seems smarter than it should be
  • 00:32:25
    yeah walks you through how to use CLA
  • 00:32:27
    coder exploring the database adding
  • 00:32:29
    visualizations and it's literally
  • 00:32:31
    telling you how to commit changes to
  • 00:32:33
    GitHub again with Claude coder how did
  • 00:32:36
    it this is this seems like it's an
  • 00:32:38
    output from Claude I'm wondering if it
  • 00:32:41
    like actually installed coder on its own
  • 00:32:44
    thing to then run it it's something this
  • 00:32:47
    is weird this is weird I'm I'm a little
  • 00:32:49
    bit blown away this this this is
  • 00:32:50
    throwing me for a loop here um okay I'm
  • 00:32:53
    not 100% sure how it did that I so I
  • 00:32:57
    know that it read anthropics
  • 00:32:59
    documentation on Claude coder so as far
  • 00:33:02
    as I can tell it like understood what
  • 00:33:04
    what Claude coder was kind of like and
  • 00:33:06
    it was able to kind of imagine what it
  • 00:33:08
    would say and what it would do in those
  • 00:33:10
    scenarios maybe I'm missing something
  • 00:33:12
    but this this feels this whole thing
  • 00:33:14
    kind of feels Next Level somehow next
  • 00:33:16
    we're testing our fighter plane but
  • 00:33:19
    sandbox not found so we just paste that
  • 00:33:21
    right back in there so it did a lot of
  • 00:33:23
    research on itself basically and it's
  • 00:33:25
    saying that Manis likly employs
  • 00:33:27
    technology similar to deep seek R1 and
  • 00:33:29
    V3 models while the specific Vision
  • 00:33:32
    model isn't explicitly named Manis has
  • 00:33:34
    confirmed multi modal capabilities
  • 00:33:36
    including image processing and Analysis
  • 00:33:38
    as part of its system so we've tested
  • 00:33:40
    uploading a screenshot but now that I
  • 00:33:42
    think about it I actually kind of told
  • 00:33:44
    it what's in the screenshot as as one of
  • 00:33:46
    our next tests will be kind of like just
  • 00:33:48
    uploading visual information to see how
  • 00:33:49
    well it's able to do that because I did
  • 00:33:51
    kind of spell it out in the text I said
  • 00:33:54
    this is what this is what's rendering
  • 00:33:55
    this is what is not rendering so it it
  • 00:33:57
    didn't even need a screenshot really and
  • 00:34:00
    uh Manis uses a multi-agent architecture
  • 00:34:02
    where specialized AI sub agents operate
  • 00:34:05
    independently under executive oversight
  • 00:34:07
    it features cloud-based asynchronous
  • 00:34:09
    operation and sandbox virtual
  • 00:34:10
    environments for safety so there was a
  • 00:34:12
    little bit of a confusion here because
  • 00:34:14
    it's um created a readme file it's like
  • 00:34:16
    oh clone the repository but there's no
  • 00:34:19
    repository it's just a bunch of files
  • 00:34:21
    that it has on its computer so we kind
  • 00:34:24
    of had to go back and forth and have to
  • 00:34:25
    say okay just zip all the files and and
  • 00:34:27
    G them to me now what I think might be
  • 00:34:29
    happening is since we ran out of our
  • 00:34:31
    sort of quota for the day and now we're
  • 00:34:33
    still going back and forth and talking
  • 00:34:35
    within a certain project maybe
  • 00:34:37
    downgrades you to a different model
  • 00:34:38
    that's not quite escapable I'm going to
  • 00:34:40
    need to do more testing to like figure
  • 00:34:42
    it out but this seems like a fairly
  • 00:34:44
    simple problem to solve basically it
  • 00:34:47
    thinks that all the files are that you
  • 00:34:49
    have all the files but it's actually on
  • 00:34:51
    its computer so so you have to kind of
  • 00:34:53
    connect the dots for it but uh we'll see
  • 00:34:55
    if it works all right but next we have
  • 00:34:57
    our video Creation with LM and apis 11
  • 00:35:00
    Labs so the haen API it could not get it
  • 00:35:03
    to work there seemed to be some issues
  • 00:35:05
    we're going to look at that later so for
  • 00:35:06
    the time being I just said use open AI
  • 00:35:09
    to create a script and then 11 labs to
  • 00:35:12
    voice that script so it completed it it
  • 00:35:14
    gave me the project file here and it
  • 00:35:16
    explained how to install it so I just
  • 00:35:18
    copi and pasted the command it told me
  • 00:35:20
    to and then installed all the
  • 00:35:22
    requirements we run the command and we
  • 00:35:24
    just type in whatever prompt we want
  • 00:35:26
    into it so we're going to say Manis is
  • 00:35:28
    the best AI Agent app for world
  • 00:35:30
    domination so let's see what it does
  • 00:35:32
    with that and so what it's doing is it's
  • 00:35:34
    sending that through an API to the open
  • 00:35:37
    AI client and the open AI whatever model
  • 00:35:39
    we're using 40 probably it sort of
  • 00:35:42
    creates a little script of a person kind
  • 00:35:44
    of like talking about how awesome the
  • 00:35:45
    thing is that I requested then it takes
  • 00:35:48
    that script it sends it to the 11 Labs
  • 00:35:50
    API so that it's able to produce it in
  • 00:35:54
    an AI voice and then it takes that and
  • 00:35:56
    saves it in this folder so here we have
  • 00:35:58
    the generated script and the speech
  • 00:35:59
    output let's take a listen hey everyone
  • 00:36:02
    if you've ever dreamed of taking over
  • 00:36:03
    the world or at least felt like you
  • 00:36:06
    could manage things a bit better than
  • 00:36:07
    the powers that be then you've got to
  • 00:36:09
    check out Manis the ultimate AI Agent
  • 00:36:12
    app designed for well world domination
  • 00:36:16
    now I know how that sounds a bit like a
  • 00:36:18
    plot out of a sci-fi movie right but
  • 00:36:20
    hear me out Manis isn't just any app
  • 00:36:23
    it's designed with the kind of features
  • 00:36:25
    that not only help you make smarter
  • 00:36:27
    decisions but actually anticipate
  • 00:36:29
    problems before they arise it goes on
  • 00:36:31
    for a while so again rating this I mean
  • 00:36:34
    it's excellent again it wasn't able to
  • 00:36:36
    troubleshoot the hay video generation
  • 00:36:38
    but I actually haven't used it in a
  • 00:36:40
    while so maybe there's some issues that
  • 00:36:41
    I'm not aware of so I'm going to try to
  • 00:36:43
    after this see if there's like some
  • 00:36:45
    issue that's caused by haen and if
  • 00:36:48
    that's the case I'll just provide a
  • 00:36:49
    solution to this thing cuz I feel like I
  • 00:36:51
    feel like it should be able to do it if
  • 00:36:53
    this is caused by haen I would give it
  • 00:36:55
    an a for this uh for nailing the prompt
  • 00:36:57
    if this was something that it failed at
  • 00:37:00
    then that certainly marks it down to a b
  • 00:37:02
    or a C+ but I still want to give it a
  • 00:37:04
    very high rating because so much of it
  • 00:37:06
    was was done really well I mean created
  • 00:37:08
    all the files it it taught me how to set
  • 00:37:11
    up those files right how to run the
  • 00:37:13
    application so I gave it the API keys
  • 00:37:15
    and it sort of put them in the config
  • 00:37:17
    slnv file it has a read me file it even
  • 00:37:20
    created like a like a web interface
  • 00:37:22
    here's kind of uh what that looks like
  • 00:37:25
    right if you say rats are not pets I if
  • 00:37:27
    you have a rat then we can select
  • 00:37:29
    whatever voice that we want and click
  • 00:37:31
    generate audio so it's processing our
  • 00:37:33
    request and here's the generate audio
  • 00:37:35
    okay let's talk about rats now when you
  • 00:37:37
    hear the word rat what comes to mind
  • 00:37:39
    maybe you think of dark alleys or
  • 00:37:41
    perhaps old cartoons with villains
  • 00:37:42
    surrounded by these creatures there's a
  • 00:37:44
    good reason for these associations rats
  • 00:37:46
    in the wild so as you can see there's a
  • 00:37:48
    lot of good stuff Happening Here the
  • 00:37:50
    reason it has that failed to generate
  • 00:37:52
    audio is some of these voices are like
  • 00:37:53
    premium voices and I guess I've and I
  • 00:37:55
    think I've downgraded my subscription
  • 00:37:57
    recently so I can use the default voice
  • 00:38:00
    but not some of the other voices but it
  • 00:38:02
    throws up that error message but it's
  • 00:38:04
    still completes the task with the
  • 00:38:06
    default voice all right so here towards
  • 00:38:08
    the end so we kept getting console
  • 00:38:11
    errors I'm putting it back in there and
  • 00:38:12
    it says matus performance May decline so
  • 00:38:14
    it it stopped so I think the reason that
  • 00:38:17
    is happening is they're getting hit with
  • 00:38:19
    a lot of people trying to test this
  • 00:38:20
    these things out I'm sure they're
  • 00:38:22
    melting their servers so we ran into
  • 00:38:25
    we've hit a limit a while ago and we
  • 00:38:27
    still continued you know updating and
  • 00:38:29
    and going under each of the tasks that
  • 00:38:31
    we were allowed to start we kept doing
  • 00:38:33
    follow-up questions digging deeper and
  • 00:38:35
    deeper so I think probably we I would
  • 00:38:38
    assume maybe we got downgraded to a
  • 00:38:40
    lesser model and eventually this thing
  • 00:38:43
    just sto because we've hit some limit so
  • 00:38:45
    I I don't quite want to give it a rating
  • 00:38:47
    on this for a number of reasons one
  • 00:38:49
    first of all it's not a software
  • 00:38:50
    development agent it's kind of a general
  • 00:38:52
    well-rounded agent two I should probably
  • 00:38:56
    use the extended version version of this
  • 00:38:57
    cuz this is a fairly complicated prompt
  • 00:38:59
    that requires a lot of coding a lot of
  • 00:39:00
    back and forth and some of these issues
  • 00:39:02
    I ran into with other coders as well so
  • 00:39:05
    I don't want to give it a final score
  • 00:39:07
    without really testing it just on this
  • 00:39:09
    prompt on its own when I have my full
  • 00:39:11
    sort of quota available to me Etc but on
  • 00:39:14
    all the other tasks I got to say I'm
  • 00:39:16
    kind of blown away so the website about
  • 00:39:19
    the AI generated video games made with a
  • 00:39:22
    kind of 90s video game theme I love it I
  • 00:39:25
    love everything about it it is
  • 00:39:26
    phenomenal right it did it really
  • 00:39:28
    captured exactly what we wanted to gives
  • 00:39:30
    a lot of info about the games a lot of
  • 00:39:32
    resources everything it's obviously
  • 00:39:34
    doing its research just phenomenal it
  • 00:39:36
    also links to the games themselves also
  • 00:39:39
    gives you more info this is all I could
  • 00:39:41
    have hoped for and more and now if I'm
  • 00:39:44
    able to I can run follow-up commands to
  • 00:39:46
    say Okay add five more games or add
  • 00:39:48
    these features that you want to talk
  • 00:39:50
    about to every single game or every
  • 00:39:51
    single listing Etc we're at a point
  • 00:39:53
    where if you're designing websites or
  • 00:39:55
    you want a website designed for you you
  • 00:39:57
    don't have to hire somebody you can just
  • 00:39:59
    tell this thing what to do and it'll go
  • 00:40:01
    and and do it it'll update it uh
  • 00:40:02
    eventually if you can even schedule
  • 00:40:04
    tasks like every morning go out there
  • 00:40:06
    and find if there's a new game and add
  • 00:40:08
    it to the resource you're basically just
  • 00:40:09
    kind of like queuing these things up to
  • 00:40:11
    happen at certain intervals that feature
  • 00:40:14
    as far as I know is not available but
  • 00:40:15
    that's super simple everything else just
  • 00:40:17
    seems to be there and and ready to go
  • 00:40:20
    this is like a plus to me just
  • 00:40:22
    mind-blowing so this is the Linux AI
  • 00:40:25
    development course it designed like a
  • 00:40:26
    little V graphic for it and like I said
  • 00:40:29
    I was kind of Blown Away here because I
  • 00:40:32
    was sure I was going to get some of
  • 00:40:33
    these things wrong because some of the
  • 00:40:34
    stuff isn't really like available online
  • 00:40:36
    yet like how do you use CLA coder to
  • 00:40:39
    work with GitHub it it nailed it and I'm
  • 00:40:41
    not 100% sure how it did that I'd be
  • 00:40:43
    curious to know like it seemed like it
  • 00:40:45
    just like read the manual and then just
  • 00:40:46
    just it's a really smart model
  • 00:40:48
    underneath it that figured out how to
  • 00:40:49
    sort of like put that information
  • 00:40:51
    together just kind of mind-blowing if I
  • 00:40:53
    had to nitpick there was like one or two
  • 00:40:55
    things in there specifically like this
  • 00:40:57
    section that like I would not have put
  • 00:40:59
    in there because again the whole point
  • 00:41:01
    of this is to install Linux right and
  • 00:41:02
    saying here if you're using Windows
  • 00:41:04
    subsystems for Linux you can do this I
  • 00:41:07
    understand why I put it in there it's
  • 00:41:08
    not needed so kind of like if I had to
  • 00:41:10
    nitpick out say this this section would
  • 00:41:12
    not be in there but everything else is
  • 00:41:14
    just just perfect A+ then we did a very
  • 00:41:17
    simple just overview of Manus Ai and it
  • 00:41:20
    brought back some really good
  • 00:41:21
    information and it gave me this whole
  • 00:41:23
    massive research that it conducted like
  • 00:41:26
    just really deep dive onto every single
  • 00:41:29
    member of the company where it's coming
  • 00:41:31
    from what open source architecture it's
  • 00:41:33
    using deep seek Etc I mean just really
  • 00:41:36
    really huge in-depth things and then
  • 00:41:39
    also and gave me this really simple like
  • 00:41:41
    four bullet point description that kind
  • 00:41:43
    of gives me like the top level overview
  • 00:41:45
    of what I need to know again A+ probably
  • 00:41:47
    a simpler prompt but I was impressed
  • 00:41:50
    with the sort of depth of knowledge the
  • 00:41:52
    Little Gems of wisdom that I was able to
  • 00:41:54
    find online and fairly quickly too again
  • 00:41:57
    A+ blown away next thing is the
  • 00:41:59
    automated you know using openai API to
  • 00:42:03
    then write a script that script uses the
  • 00:42:05
    11 Labs API to to do a voice over that
  • 00:42:09
    voiceover gets popped into haen to
  • 00:42:11
    create a video it struggled a lot with
  • 00:42:13
    setting up haen what I had to do is
  • 00:42:15
    actually I realized that I didn't have a
  • 00:42:17
    haen subscription at the time so I
  • 00:42:19
    re-enabled it I regenerate the API key I
  • 00:42:22
    gave it to it it still wasn't able to do
  • 00:42:23
    it so it didn't complete a part of the
  • 00:42:25
    prompt but I don't to call it as that
  • 00:42:28
    was its mistake it might have been just
  • 00:42:30
    something to do with my account at the
  • 00:42:32
    time or how the API is I still have to
  • 00:42:35
    see why specifically it failed so I
  • 00:42:37
    don't want to gr it on that aspect of it
  • 00:42:39
    but the rest of it it handled flawlessly
  • 00:42:42
    it gave me all the files it created a
  • 00:42:44
    read me it told me how to like install
  • 00:42:45
    it and run it locally it even created a
  • 00:42:48
    little web UI that it could use it put
  • 00:42:50
    the API keys in the right place both for
  • 00:42:53
    functionality and just like security
  • 00:42:54
    basically if it put the API Keys kind of
  • 00:42:57
    just in the main code right and then you
  • 00:42:59
    wanted to upload it to GitHub you'd sort
  • 00:43:00
    of expose your API Keys publicly but
  • 00:43:03
    there's certain locations where you
  • 00:43:05
    you're supposed to put your API key so
  • 00:43:06
    that if you're you know sharing it with
  • 00:43:08
    G GitHub and stuff like that those don't
  • 00:43:10
    get shared it put them in the right
  • 00:43:11
    place so not only does it understand how
  • 00:43:14
    to like use them functionally but also
  • 00:43:16
    like where to put them for security
  • 00:43:18
    purposes and again I would say this is
  • 00:43:19
    an A+ assuming that the issue of the
  • 00:43:22
    haen was on haen side not on mana's side
  • 00:43:25
    but still it's very impressive in any
  • 00:43:28
    case and our World War II simulation I'm
  • 00:43:30
    not ready to call it yet I'm going to
  • 00:43:31
    have to like rerun it and see what the
  • 00:43:34
    issue is looks like we hit the context
  • 00:43:36
    window on here so and this makes sense
  • 00:43:39
    uh what I maybe should have done is
  • 00:43:41
    actually once it completed it I could
  • 00:43:42
    have just throw it in a cursor or
  • 00:43:44
    something else and had that
  • 00:43:45
    troubleshooted because that's
  • 00:43:47
    specifically made to be a software
  • 00:43:50
    engineer you know an AI used for coding
  • 00:43:52
    this isn't so this might might not have
  • 00:43:55
    been like the best way to test it but
  • 00:43:56
    based on everything I've seen this is
  • 00:43:58
    just extremely exciting extremely
  • 00:44:00
    capable multiple times where I was kind
  • 00:44:02
    of like Blown Away by what it was able
  • 00:44:04
    to do and they're talking about open
  • 00:44:06
    sourcing some of it so I'm not sure
  • 00:44:08
    exactly if they're doing everything all
  • 00:44:10
    of it or just certain portions of it so
  • 00:44:11
    we'll see but man this thing feels very
  • 00:44:14
    powerful and it feels ahead of open Ai
  • 00:44:16
    and anthropic in terms of like the AI
  • 00:44:18
    agents sort of structure that they've
  • 00:44:20
    created so let me know what you think
  • 00:44:22
    but from my experience from what I've
  • 00:44:24
    seen online people are losing their
  • 00:44:26
    minds over this thing if you've made it
  • 00:44:28
    this far thank you so much for watching
  • 00:44:29
    my name is Wes rth and I'll see you next
  • 00:44:31
    time
Tags
  • Manis AI
  • Intelligence Artificielle
  • Agent Autonome
  • Open Source
  • Technologie
  • Développement Web
  • Recherche
  • Linux
  • Jeux Vidéo
  • API