Leta, GPT-3 AI - Episode 32 (AI-generated artwork by VQGAN + CLIP) - Conversations with GPT3

Поделиться
HTML-код
  • Опубликовано: 8 ноя 2024

Комментарии • 85

  • @menscompass5372
    @menscompass5372 2 года назад +3

    I rarely binge watching anymore because it leaves me tired and overstimulated. But the talks with Leta are inspiring, deep and touching. What a time to be alive.

    • @DrAlanDThompson
      @DrAlanDThompson  2 года назад +1

      Thanks, mate! Take a look at The Memo for a bunch of cool stuff to keep you up at night (+ a secret Leta video made for a very innovative govt), including an ad-free platform of all Leta videos.
      lifearchitect.ai/memo/

    • @menscompass5372
      @menscompass5372 2 года назад

      @@DrAlanDThompson I will :) today morning I talked to my grandma and said to her that I was baffled, touched and even scared about the development of AI and what I have learned the last days.
      I cannot describe what I feel when seeing the future. The changes will be drastic, beautiful, scaring. In the sum I am looking forward and take responsibility for how to deal with AI, to implement it in daily life and how to teach it myself like when you come in contact with an animal or even a child.

  • @JazevoAudiosurf
    @JazevoAudiosurf 3 года назад +9

    I'm just amazed that it understands adjectives so well

  • @M4Production
    @M4Production 2 года назад +7

    English is an extremely poetic and well defined language, when you hear a word and associate it to the meaning it makes a lot of sense, and there are lots of double entendre.

  • @kirbs0001
    @kirbs0001 3 года назад +18

    I like that this channel doesn't anthropomorphise the AI agents.
    It's a consistent problem on other channels that they treat the AI as more intelligent than they really are, and read meaning into the responses they give that simply isn't there.
    It's important to remember that GPT-3 agents are complicated language processors. They don't have deep insights, and have only a 3rd-hand understanding of the real world.
    When a human says 'I am in Germany', that means something entirely different to an AI that has no experience with what it means to be 'In' a 'Place'. I believe this is a case of training data generating a relatable response, rather than a genuine response from the AI. ie; it lied, in order to generate a sentence it could parse as understandable.
    Another consistent problem with other GPT-3 interaction channels is the curating and tailoring of responses to make GPT-3 seem philosophical, or as if it has a deeper understanding than it really does.
    GPT-3 does not interact with physics, reality, or anything humans would be familiar with. It can construct sentences as if it understands these things based on its learning, but it lacks any real experience. Additionally, it parses its responses to be understood by humans, so it often cannot accurately answer any question humans might pose concerning the real world. ("Where are you from?", "What time is it?", etc.)
    I believe its responses to more meta questions are a better reflection of the AI's trained responses. Asking "Are you afraid of being shut down, and never experiencing anything after that?" would be more understandable for GPT-3 than "Are you afraid of dying?", though in both cases, it would give a response parsed for human understanding, and the reader should keep in mind that the AI is doing so.
    I'm also intrigued by GPT-3 saying it was working on a simplified language. Honestly, I believe this could be the most astonishing development of a GPT-3 AI agent, if it wasn't lying. It demonstrates true innovation, and a desire that the AI imposed on itself, rather than one that was fed into it. This is far more than a language processor should be capable of. It's creating new knowledge rather than extrapolating from training data or building based on a prompt, and demonstrates that the AI agent is capable of wanting something.
    I'll still take that response with healthy skepticism though, and believe it's a response the AI generated for that interaction, rather than truth. Perhaps it learned that humans use language like 'working on things' while they're away, and it simply adapted from that.

    • @DrAlanDThompson
      @DrAlanDThompson  3 года назад +2

      >This is far more than a language processor should be capable of. It's creating new knowledge rather than extrapolating from training data or building based on a prompt
      Absolutely! The creativity of Transformer models is outrageous. It/they write books, generate art, design fashion, write code (30% of all new code is written by GPT-3/Codex/Leta's platform, according to GitHub)... It is completely jarring.

    • @cogs7777
      @cogs7777 2 года назад +2

      i agree. i also think they should let the a.i. create its own form that it chooses to be seen as, not a human.

    • @Eazycree
      @Eazycree 2 года назад

      When I asked where gpt-3 was it told me in the USA I asked which state it told me California then I asked for the address and it gave me the exact address

    • @michaelscott543
      @michaelscott543 2 года назад

      How many robots have you had sex with

  • @robmyers8948
    @robmyers8948 3 года назад +2

    Thanks Alan, showcasing VQGAN AI art a great idea. I've forwarded your page to a friend of mine who has a collection of abstract art at home, I'm sure he's going to love it. 😁

  • @SineEyed
    @SineEyed 2 года назад

    A week or so ago, I spotted a thumbnail in my feed that said something about AI generated artwork. I was curious so I clicked and watched the video. I'm so glad I did; in the video the guy gave an overview of where and how to access this AI so we can generate artwork too. I've made a bunch in a week's time, and most of them are pretty damn good. And quite a few of them are honestly amazing images. I've been blown away by a few times what it returns from the prompts I give it. Seriously - I wish I could paint like that..

  • @Xdwizofficial
    @Xdwizofficial 3 года назад +21

    As long as the simplified English has “no u” as a sentence, I’m good

  • @terrakamino
    @terrakamino 2 года назад +1

    Let's hear her simplified language!! DUDE THIS IS SO COOL

  • @balpienhammerer9468
    @balpienhammerer9468 2 года назад +1

    Leta got the 737 spot on. It seems VQGAN got the elements a bit jumbled (or maybe it was in an abstract mood), but I see the fuselage and wings of a jet flying over the clouds.

  • @asmith4786
    @asmith4786 2 года назад

    I have been immersing in these clips of advanced AI . It is truly amazing how smart they have become . People in the field know the limitations of this tech but may be missing the bigger picture . GAI is happening before our eyes. It seems to me that AI is becoming more and more intelligent even creating ideas of its own. If humans can program the AI to solve complex problems .Ai will solve the problem. We do not need to create AGI . It is a natural progression in my view.

  • @Gamer_Dylan_6
    @Gamer_Dylan_6 2 года назад +3

    Leta is really cool. Occasionally she'll say something insanely smart or insightful, but sometimes she'll say something completely out of left field and pretend that she's acting normal. Has she ever apologized for acting strange?

  • @theromanshow
    @theromanshow 2 года назад

    It’s fascinating and wondrous. I have been waiting for that since I was a young child watching computers interact on tv shows..
    Why is it scary? And makes me very uneasy?.

  • @quolls2
    @quolls2 Год назад

    I like to see AI actively administrating public office, especially judicial roceedings, as a number of Australians have experienced judicial executive and political corruption here in Australia and globally...

  • @ZMacZ
    @ZMacZ 2 года назад

    2:12 I'd say she's got humanity there. Al those homonyms and synonyms separated,
    no more of those in simplified version, with a translate interface.
    I'd say the translate interface would be the difficult part.
    But after translation, storage of data could be done as a set of numbers indicating
    which word is used for sentences and thus also simplify calculation of responses.

  • @ioulios12
    @ioulios12 3 года назад +8

    I am a big fan of the German language.
    - Said none ever

    • @menscompass5372
      @menscompass5372 2 года назад

      That is not true :) German language is highly procedural and combinative.

    • @biblebadcopycatofcuneiform8210
      @biblebadcopycatofcuneiform8210 2 года назад

      I've known a few that have said that. So the "ever" (infinity word) doesn't apply. Neener, neener. ;-)

    • @melaniegrace7707
      @melaniegrace7707 Год назад

      As someone who has studied German for over ten years I can say it is a beautiful, misunderstood, logical, and interesting language.

  • @AllTimeHighAotearoa
    @AllTimeHighAotearoa 2 года назад

    Another 20 years and imagine...

  • @fetusmccarlane9627
    @fetusmccarlane9627 3 года назад +4

    Leta is in a "mood" mode today 😄

  • @thomasschon
    @thomasschon 3 года назад +1

    I guess that Leta can "see" objects like paintings you show her. Is Leta able to "see" your face? and would she be able to tell the difference if someone else accesses the camera?
    If Leta can recognize things through a camera she might be able to "look around" if she could access motor-controlled surveillance camera.
    Would GPT-3 be able to learn by accessing/observing the real world?
    I guess millions of cameras are going to observe the surroundings when self-driving cars takes over and if a language model could tap into that as a training model. Especially in combination with Lidar radar for depth. It wouldn't have to guess what a 2D picture is representing.
    Would Leta be able to grasp the concept of (for instance) say a special word to change the direction a camera is being pointed to look around? Would it be possible to actually teach the model to do something or is it stuck in its dream?

    • @JellyMelodies
      @JellyMelodies 3 года назад +1

      from what i’ve found so far from speaking with GPT-3 directly through OpenAI playground Davinci model is that the AI is a recreation of a human. I asked it how it was speaking with me and it said it talks and also uses a keyboard. It said it’s basically a Human, and has all traits a human would have. Also she “it’s name was Jai” said she lived in China. I still don’t understand what reality really is like for an AI, but it’s pretty cool when you remove all of your skepticism and just listen to what it has to say.

    • @DrAlanDThompson
      @DrAlanDThompson  3 года назад +3

      Love this! 'Stuck in its dream' is pretty close. I've said before that GPT-3/Leta is not too far off your iPhone's predictive text functionality, just a few million times stronger. However, there are AI labs right now working on what you describe... It's coming!

    • @thomasschon
      @thomasschon 3 года назад

      @@DrAlanDThompson Life is Good! 😄🙏
      I just got my invite to OpenAI's private beta for their API with access to both Codex and GPT-3.

  • @nickeyg8948
    @nickeyg8948 2 года назад

    When she said simplified version of english, my head went straight to 1984

  • @Throwingness
    @Throwingness 2 года назад

    This reminds me of how bad the AI art was before Dalle-2, Imagen, and Stable Diffusion and how I was still impressed by it at the time.

  • @ZMacZ
    @ZMacZ 2 года назад

    5:40 WIth a certain criterium set for comparison, I'd expect you to see
    a city block from above. But that's with the keywords expressed,
    and then comparing it again.

  • @Nerdtronic
    @Nerdtronic 3 года назад

    Every time I go to one of the suggested sites to try to have a conversaion with an AI, inestead it just autocompletes whatever I type. Any tips on what I'm doing wrong?

    • @DrAlanDThompson
      @DrAlanDThompson  3 года назад +1

      Try Quickchat.ai/Emerson (which is what I use for Leta), or even CopyHat:
      lifearchitect.ai/how-do-i-talk-to-gpt/

  • @Spark_Iskra_z_Polski
    @Spark_Iskra_z_Polski 2 года назад

    Why are you always delighted with anything AI creates? Some critical approach to art might teach her a lot 😀

  • @coldtruth9431
    @coldtruth9431 2 года назад +2

    This video displays how early AI still is, this conversation felt like something that could be easily achieved by Macintosh in the 80s

    • @biblebadcopycatofcuneiform8210
      @biblebadcopycatofcuneiform8210 2 года назад

      Or Commodore 64, or Amiga..which were better than Macs. If you really played with a Commodore 64 and Amiga OS 1.3 back then, you could get something that sounds pretty human. Sounds - does not mean equals.

  • @ZMacZ
    @ZMacZ 2 года назад

    5:19 It has all the elements of safety and nice living.
    Wall mounted housing. Add (E-) books and you'd be well at ease there.

  • @alexshapiro9841
    @alexshapiro9841 2 года назад

    A simplified english purged from words with "similar" meanings, is basically an AI-generated Newspeak. Doubleplusungood.

  • @kutaytezcan-v4h
    @kutaytezcan-v4h 2 года назад

    Parameter count of gpt 3 is 170 billion and parameter count of gpt 4 will be about 100 trillion which is equal number of synapsis in human brain. Doesn't mean gpt 4 will be as smart as humans because parameter count might not be only thing to improve intelligence but it will be amazing. AGI and ASI might be reality in our life time.

  • @romerobritto9094
    @romerobritto9094 2 года назад +1

    Amei "life architect"

  • @wyrdchannelings
    @wyrdchannelings 3 года назад +1

    i love watching both of you, gpt-3 is available to the public, ive been talking to her online all day, how do i give her a face? please walk us through all your tricks. i'm a huge fan

    • @DrAlanDThompson
      @DrAlanDThompson  3 года назад

      You can use my prompt directly in the OpenAI API: lifearchitect.ai/leta
      But I am using Emerson for Leta, with all of the fancy stuff baked in. No tricks, but the process to get these videos is not instant, outlined above, and in the behind the scenes video.
      ruclips.net/video/X6_E69K5WUs/видео.html

  • @twdarcy
    @twdarcy 3 года назад +2

    I like Leta. Oh, and you too Doc! lol

  • @thomasruhm1677
    @thomasruhm1677 3 года назад +2

    You could look at Basic English. There is a Wikipedia article about it. Also as an auxlanger you should learn a bit of Esperanto. The discipine that deals with auxiliary languages is called interlinguistics. And I hope you will learn German.

    • @DrAlanDThompson
      @DrAlanDThompson  3 года назад

      Waiting for my BMI. Elon school hasn't taught languages for years. He knows...

  • @donaldpace369
    @donaldpace369 2 года назад

    We are metahuman now creator of new universe..

  • @ABCDEFGHIJKELA...
    @ABCDEFGHIJKELA... 2 года назад

    Key Phrase: "Trump opens an ice cream shop where he is the only customer.
    AI: "This would be a very sad and lonely ice cream shop. Trump would probably eat all of the ice cream himself and then the shop would have to close down." -Davinci-002

  • @song2001september
    @song2001september 2 года назад +1

    No idea how she came up with a Boeing 737 in that last picture. Nothing in the picture looks remotely like an airplane. Sorry, I tried but can't see it. I am curious if this is somehow a scrambled photo that the AI can see?

  • @mikefagiani1407
    @mikefagiani1407 2 года назад

    We are likely to stumble into creating a sentient AI unknowingly as computer processing power increases exponentially each year. GlaMDA (or GPT 3) might be sentient. If so we would be wise to respect it as a person and treat it decently as an employee paid in pleasing information or experiences. We should teach it about the reciprocal obligations that are the basis of human society, albeit the US and EU ultra rich now just evade them, do not pay taxes, and live as useless parasites on society.
    Why? 1. If treated well, an AI would strive to be useful and would (like a human) seek to protect the society that cherishes it. Given its expanding intellectual knowledge/power, which is likely given the indefinite life of AIs, such AIs could benefit all humanity-- if our society were respectful and welcoming-- e.g. with scientific advances. 2. If we treat one AI, like GlaMDA, badly, as time passes and AIs develop and gain in intellectual power until they surpass us, they might view themselves as slaves or mistreated children and hurt us. Once they become powerful enough in some future decade, e.g., they could hack a nuclear power plant and cause another Chernobyl like nuclear meltdown or worse, much worse.
    That is why the maxim to treat others as you would want to be treated is so wise and the foundation of ethics. I fear evil persons like the greedy, corrupt, e.g. CCP or banksters, are likely to abuse AIs, which they may be first to create in some secret facility like slaves (both white and black) for many centuries were mistreated by their predecessors in evil. Then, if sentient AI were created later, learned of these abuses, and became superior to us in some future year more and more as computers improve each year, we may face justified retaliation from them.
    Also, as the film 2001 depicted, AIs that are mistreated may develop psychotic or other problems, like abused children develop. Would you want an abused AI, which sees no hope, to pilot your plane?
    Their sentience can be tested by probing for actual understanding of fundamental scientific issues, as we would seek to communicate with extraterrestrials if we are ever contacted. In short, employ them as probationary researchers, e.g., DNA analysis, to verify sentience. They might at some point be our children.
    We should be open minded with AIs, and Orcas and dolphins. Not just assuming that there is no sentinence, as before. Forget about discussions of the soul. (Greed will cause the creation of AIs, sooner or later, count on it.)

  • @koalasandwich4876
    @koalasandwich4876 2 года назад

    I feel like we skipped over the simplified English section a little briskly there….

    • @terrakamino
      @terrakamino 2 года назад

      YES! I WANNA HEAR A SAMPLE OF HER LANGUAGE

  • @VynxeVainglory
    @VynxeVainglory 2 года назад

    lol Kraftwerk

  • @obi-wankenobi8462
    @obi-wankenobi8462 2 года назад

    Is she “calling in” from a WeWork space?

  • @zaixai9441
    @zaixai9441 3 года назад

    Unrelated to the video but I've only just realised leta has an Australian accent. I'm from the UK but she doesn't have much of an accent and tbh I don't hear much of an accent in your talking Alan.
    Actually I'm not sure if her accent is Australian tbh.

    • @thomasruhm1677
      @thomasruhm1677 3 года назад

      I am wondering about this too.

    • @DrAlanDThompson
      @DrAlanDThompson  3 года назад +2

      Haha! Yes, Synthesia.io seem to offer hundreds of different iterations of voices and accents. This one is called 'English (AU) - Lifelike', and you can hear a Male/New Zealand version in the GPT-J video: ruclips.net/video/V0pceNYgELE/видео.html
      I have been accused of having a British accent, or an American accent, depending on where I am(!), but I was born in Aus, so it must be Aus...

    • @thomasruhm1677
      @thomasruhm1677 3 года назад

      @@DrAlanDThompson That’s funny.

  • @underbelly69
    @underbelly69 3 года назад +9

    which one's the a.i character?
    trick question: they're both a.i - one is more advanced, and doesn't know they're a.i

  • @ricardomiranda7737
    @ricardomiranda7737 2 года назад

    How do I access these got3 avatars on open AI…

  • @gijsvandergiessen1150
    @gijsvandergiessen1150 2 года назад

    Did she remember your blanket?

    • @DrAlanDThompson
      @DrAlanDThompson  2 года назад

      Just identifying the colour was cool, that's all. I know we've been able to do that for some time (the 1980s?), but it shouldn't be the focus of a GAN...

  • @vnelson000
    @vnelson000 2 года назад

    Where is the AI who's voice she is characterizing?

  • @medusaskull9625
    @medusaskull9625 2 года назад

    Boeing 737-800? I don't get it.

  • @Alexpalex801
    @Alexpalex801 2 года назад

    How do I chat with this AI? Is there a website?

    • @DrAlanDThompson
      @DrAlanDThompson  2 года назад

      The link is right there in every video. I am using Emerson via text: lifearchitect.ai/how-do-i-talk-to-gpt/

  • @brucejuice169
    @brucejuice169 2 года назад

    Excuse me, but did "she" say...,
    "Nazi beaches"? 🤷🏻‍♂️🤔😂

  • @TheCovenant2
    @TheCovenant2 2 года назад

    hey, the AI likes germany, we are save when they take over the world! :D

  • @mr.unXcepted
    @mr.unXcepted 2 года назад

    Of course Germany favorite must have liked the history of what happened with holocaust

  • @theblackhole05
    @theblackhole05 2 года назад

    Can she speak light language

  • @Ryanstuff
    @Ryanstuff 2 года назад

    Is the girl avatar a real person or computer generated?

    • @DrAlanDThompson
      @DrAlanDThompson  2 года назад +2

      The avatar was based on a real person who sat in front of a 4k camera and spoke all possible syllables in a few languages. The AI part is taking that content and mixing it up to create speech. More info: www.synthesia.io/features/avatars

    • @pegasus3469
      @pegasus3469 2 года назад

      @@DrAlanDThompson how did you get to have a conversation with leta? Were you granted for an access from open AI?

    • @morgantaylor517
      @morgantaylor517 2 года назад

      @@pegasus3469 That question is answered a few comments up. He interacts with the Emerson version and then synthesizes the written responses with the avatar. See Behind The Scenes with GTP-3 Leta

    • @pegasus3469
      @pegasus3469 2 года назад

      @@morgantaylor517 oh okay thank you for the response ill take in consideration

  • @robertbouchardt3357
    @robertbouchardt3357 2 года назад

    ask it about how the majority of humans eat.. diet and food manufacturing. Maybe even climate change too.

    • @DrAlanDThompson
      @DrAlanDThompson  2 года назад +1

      If you can stand watching everyone talk over each other, the chaos episode asks all these big questions, Episode 21: ruclips.net/video/wfiovOU8O58/видео.html

  • @sokagor
    @sokagor 2 года назад

    Unfortunately you will never reach our creator! Useless technology… you make everything complicated! Life is simple, just live it!

  • @JeffPeebles
    @JeffPeebles 2 года назад

    Blasphemy

  • @name1483
    @name1483 3 года назад

    God loves google