Can robotics overcome its data scarcity problem?

Поделиться
HTML-код
  • Опубликовано: 22 янв 2025

Комментарии • 90

  • @DrWaku
    @DrWaku  11 месяцев назад +17

    If you or someone you know is an expert in an AI topic, I'd love to chat sometime for video ideas! See you in discord.

    • @actuallynph
      @actuallynph 11 месяцев назад +3

      What about a video on fine tuning of LLMs

    • @PG13park
      @PG13park 11 месяцев назад

      I am not an AI expert, but I am a student working on an AI documentary and I would love to have the opportunity to interview you. I will ping you in the discord.

    • @ronm3804
      @ronm3804 10 месяцев назад

      you should make a video about the potential crash of LLM's due to their regurgitated training data and decrease in accuracy relevance.

    • @roylyon6631
      @roylyon6631 10 месяцев назад

      Thanks for your enlightening discourse. I had two ideas. one: Could DNA programing be superior for AGI and singularity given the enormous capacity for data and low energy requirements. How can we get it to move quicker? two: Might a global trade association or membership requirement in an oversight entity help with monitoring ethical usage? Those AI providers who would elect not to participate could be monitored more covertly. Not big brother but " big daddy" the philosopher king of all AI. The one all other AI providers bow down to voluntarily. The ultimate security AI--?

  • @alanchen5400
    @alanchen5400 11 месяцев назад +7

    Holy, this video just helped me understand the robotics field a ton more. This is probably one of the best videos I've seen on robotics!

  • @Samucacamilo
    @Samucacamilo 9 месяцев назад +1

    This may be one of the best robotics videos I've ever seen.

    • @DrWaku
      @DrWaku  9 месяцев назад

      Thank you very much for your kind comment :)

  • @DaveShap
    @DaveShap 11 месяцев назад +39

    Only about 20% through the video right now, but I hope you cover the first mover advantage. Like, if Tesla is the first to get 10000 robots out, they'll have better data from real robots.

    • @DrWaku
      @DrWaku  11 месяцев назад +23

      Hmm, I didn't talk too much about that. More about the bootstrapping problem. Video ideas for future haha

    • @Walter5850
      @Walter5850 11 месяцев назад +15

      Why would you comment at only 20% of the video?

    • @wburris2007
      @wburris2007 11 месяцев назад +3

      Tesla is also using video data and models from self driving cars as a starting point for their robots.

    • @zrakonthekrakon494
      @zrakonthekrakon494 11 месяцев назад +8

      You two should collab on a video where you have a discussion about future projections

    • @kairi4640
      @kairi4640 11 месяцев назад +1

      ​@@zrakonthekrakon494ikr? Literally thought same thing. Would be cool to see ai discussion RUclipsrs collab together and talk in a video.

  • @danielchoritz1903
    @danielchoritz1903 11 месяцев назад +3

    I wonder if there is somewhere the first "battle angel"- A robot learning martial arts. You could make a tournament and let each participant share the responding data, while the winner gets paid his new model as a seeded participant for the next.
    This could be the next big sport, or better, this will become one or the most watched sports-event in the next 5 years.
    I hope somebody gets this idea on the run. This would be more than just fun to watch.

  • @fanfanfanf
    @fanfanfanf 11 месяцев назад +7

    Hi @drwaku
    I’m wondering how AGI and ASI would accelerate the development of human aging science. When do you think we can expect therapies for aging and will AGI alone be able to make it happen, or will ASI be needed? Will AGI itself be able to cause powerful breakthroughs in this matter literally every single day, since invented?

    • @hydrohasspoken6227
      @hydrohasspoken6227 11 месяцев назад

      No AGI will happen within the next 200 years. Look, billiona are being invested to solve much simpler problems yet we can't solve them easily. But let me try to be optimistic: AGI is massively infinitely complex.
      By the way, there are only 3 groups of people talking about AGI:
      -CEOs(for that sweet investment flow)
      -Content creators(for obvious reasons)
      -Common folk(for the excitement).
      No 9-5 job AI specialist is expecting AGI within the next 200 years.

    • @snowcones8292
      @snowcones8292 11 месяцев назад +1

      @@hydrohasspoken6227lmao boomer

    • @edwardmacnab354
      @edwardmacnab354 11 месяцев назад

      they are using AI in protein folding leading to advancements in gene therapy

    • @edwardmacnab354
      @edwardmacnab354 11 месяцев назад

      @@snowcones8292 it applies to us all / Wait till you see how quickly it is all gone Grasshopper

  • @mattcruz5266
    @mattcruz5266 11 месяцев назад +1

    Hey Dr Waku, could you tell me why I have not heard about evolutionary biology being used to develop foundation models? For example the human eye started as light sensing cells that could only tell when it was light and dark then the surface the cells were on bent allowing them to tell direction. It seems like this would be something easy to develop.

    • @Hashtag-Hashtagcucu
      @Hashtag-Hashtagcucu 8 месяцев назад

      Eye has neurons so they process the light and only transmit to the brain trough the optical nerve what is relevant for the brain

  • @cacogenicist
    @cacogenicist 11 месяцев назад

    Of course we _do_ need better feedback in the fine manipulators for robots. We need sort of peripheral nervous systems, and just much more sophisticated hardware with tactile sensors.

  • @rickc.ferreira
    @rickc.ferreira 10 месяцев назад +2

    This was a really good video. Thank you

    • @DrWaku
      @DrWaku  10 месяцев назад

      Thanks for your comment!

  • @Hashtag-Hashtagcucu
    @Hashtag-Hashtagcucu 8 месяцев назад

    @DrWaku great thematic!! I discover your channel couple days ago and I like very much all your videos!❤ …24 years ago, talking to a linguist friend about automatic machine translation strategies, he explained to me why it is so difficult to understand natural language semantically without having a physical body. For example, imagine the life experience of an extraterrestrial fish with a human-like brain where the concept of weight, up and down and with no hands is essentially different than what many would interpret in our terrestrial language metaphors. And our language is full of them( see embodied cognition of Prof. George Lakoff) That's why it's so important to train LLM with embodied data. I suspect all those expensive HUMANOID robots are to training LLM to understand human natural language and body language , not to do our housekeeping work!

  • @roylyon6631
    @roylyon6631 10 месяцев назад

    Might a compulsory membership requirement in an AI trade association help with ethical and security considerations in a proliferating set of AI providers?

  • @Dan-t7d
    @Dan-t7d 11 месяцев назад +6

    Fantastic 🎉

    • @DrWaku
      @DrWaku  11 месяцев назад +1

      Thank you!!

  • @andregustavo2086
    @andregustavo2086 10 месяцев назад

    Please .make a video about back propagation 🙏🙏🙏

  • @paramsb
    @paramsb 11 месяцев назад

    Great video as always :)

  • @williamal91
    @williamal91 11 месяцев назад +1

    Good morning Doc, from the UK best wishes

    • @DrWaku
      @DrWaku  11 месяцев назад

      Cheers! Thanks for watching as usual

  • @urniurl
    @urniurl 11 месяцев назад

    A useful overview, cheers!

  • @andreasmoyseos5980
    @andreasmoyseos5980 11 месяцев назад

    Thank you brother, excellent stuff as always.

  • @LongDefiant
    @LongDefiant 11 месяцев назад

    How do you get machine learning to understand creativity?
    For instance, if you want them to understand movement you move a lot and translate that into data.
    If you want them to understand language then you give them a lot of language and you turn that into data.
    But how can you teach an AI to be creative? Are we just banking on random emergent capabilities? Like move 37? How long will it take for AIs to random into creativity or insight?

    • @carlosamado7606
      @carlosamado7606 10 месяцев назад

      creativity is very ambiguous even for humans. the mere act of making some art requires a taste, technique and goal. then you achieve an aesthetic that for you makes sense. the thing is you might create something not many resonate with but will be seen as creative in 100 years. is not even easy for humans to be creative. everything is a remix.

  • @MrJaggy123
    @MrJaggy123 11 месяцев назад

    "Laura" fine tuning? Typo, or pun? 🤔

  • @DesignDesigns
    @DesignDesigns 11 месяцев назад

    Great video....Robot is the future, no doubt....

  • @noBoxKhan
    @noBoxKhan 11 месяцев назад +10

    Wake up mom! New Dr. Waku video just dropped!

    • @DrWaku
      @DrWaku  11 месяцев назад +3

      XD the benefit of having a slightly unpredictable posting time?

    • @AGINCY
      @AGINCY 11 месяцев назад +4

      Doctor who?

    • @kairi4640
      @kairi4640 11 месяцев назад +2

      So close to Wakfu.

  • @viralsheddingzombie5324
    @viralsheddingzombie5324 11 месяцев назад

    I was born without a cerebellum. I wonder if they considered using T-rex style robot arms?

  • @crawkn
    @crawkn 11 месяцев назад +1

    Robotics models need to have the equivalent of mirror neurons, and learn how parts of their machine bodies correspond with those of the humans they visually observe performing tasks. There is plenty of visual training data, they just need to learn mimicry.

  • @noBoxKhan
    @noBoxKhan 11 месяцев назад +1

    Wouldn’t synthetic visual data be extremely useful in this case? Imagine we could generate video that is similar to the visual inputs the robot could use train its movements.

    • @vitalyl1327
      @vitalyl1327 11 месяцев назад +3

      This is exactly what we do. More so, data does not even need to be photorealistic, skills learned in simplified worlds are transferable to reality and more elaborate simulations.

  • @tommags2449
    @tommags2449 10 месяцев назад

    Love you Dr Waku

  • @chrissscottt
    @chrissscottt 11 месяцев назад

    Interesting thanks.

  • @SIRENSCINEMATIC-SCOTLAND
    @SIRENSCINEMATIC-SCOTLAND 11 месяцев назад

    Thanks for this video I work in robotics and SDG :)

  • @SasskiaLudin
    @SasskiaLudin 11 месяцев назад +1

    The time is ripe for Piagetian robots.

  • @petemoss3160
    @petemoss3160 11 месяцев назад

    VR teleoperation & AR observation data ...

  • @G1364-g5u
    @G1364-g5u 10 месяцев назад

    found your channel, then started binging

  • @zrakonthekrakon494
    @zrakonthekrakon494 11 месяцев назад +1

    You are like the AI explained for the supply chain and industry dimension of AI instead of the technical details

  • @Dan-t7d
    @Dan-t7d 11 месяцев назад +3

  • @QuantumIllusionStudio
    @QuantumIllusionStudio 11 месяцев назад

    Good vid

  • @oowaz
    @oowaz 11 месяцев назад +1

    80%? that's crazy

    • @DrWaku
      @DrWaku  11 месяцев назад

      I know right... Hard to get that kind of monopoly in any context, let alone high margin high tech production.

  • @831Miranda
    @831Miranda 11 месяцев назад +1

    Seems like enlisting the help of children might be a good way to gather data from the real world ? Like outfitting a whole kindergarten with sensorial gloves, etc 😂 But on a more serious note, I'm very concerned that SAFETY (ie not being evil, having extreme ethics for the good of mankind) is absolutely critical for robots. LLMs and AGI and SGI will need the same. Hasn't anyone come up with more laws than Asimov's? We need to teach AI about 'man's inhumanity to men' and how * that * is the greatest problem for the planet....

  • @JoelMorton
    @JoelMorton 11 месяцев назад +1

    Everyone support this guy so he can afford to buy a space heater for for his apartment.

    • @DrWaku
      @DrWaku  11 месяцев назад

      Don't forget the electric bill to power the space heater. Those things are expensive at 20 below.

  • @econdude3811
    @econdude3811 11 месяцев назад

    Rather than program robots to allow fine motor movements, why not give them the capability of learning, just like humans do when they're little. I have no idea how to do something like that but intuitively it seems better

  • @Sci-Que
    @Sci-Que 11 месяцев назад

    I don't know enough about robotics or AI to have an opinion on how they will get there. They will get there soon though. I am sure of that, simply from seeing how far they evolved in 2023.

    • @hydrohasspoken6227
      @hydrohasspoken6227 11 месяцев назад +1

      Being "sure" doesn't mean much.
      Remember back then when the whole world hoped to travel in supersonic speeds around the Globe?
      Capabilities don't grow exponentially forever. They eventually hit a plateau. Will be no different with AI. Ask self driving technology

    • @Sci-Que
      @Sci-Que 11 месяцев назад

      @@hydrohasspoken6227 Well my friend, Look around you. I am very sure of this. All of our technology came from trial and error until it was reality.

    • @Kurdish20226
      @Kurdish20226 11 месяцев назад

      @@hydrohasspoken6227self driving tech is what scares me. It has largely slowed down for the past 5-6 years. Hope that doesnt happen to AGI.

    • @hydrohasspoken6227
      @hydrohasspoken6227 11 месяцев назад

      @@Kurdish20226 , it will, because nothing grows exponentially forever.

    • @hydrohasspoken6227
      @hydrohasspoken6227 11 месяцев назад

      @@Kurdish20226 , do not expect AGI within the next 200 years(I am being very optimistic here).

  • @ScientistBaffled
    @ScientistBaffled 11 месяцев назад

    Maybe in combination with a technology like X Y O. decentralized geospatial yum! Yum!

  • @Ari_diwan
    @Ari_diwan 11 месяцев назад +2

    What can the normies like me do to improve chances of success in this new world we are entering

    • @Kurdish20226
      @Kurdish20226 11 месяцев назад +2

      If you were to ask me, not much other than just follow the the progress of the tech and listen to experts. We dont know if well stagnate soon for some time to come regarding this tech or be 50 years in the future in the next 5-10 years.

  • @dondecaire6534
    @dondecaire6534 11 месяцев назад

    AI and robotics working in tandem to feed each other data in a simulated environment and then in a real world environment will help BOTH progress much faster than we can even comprehend. That is why simulated data and environments like Sora are so important to the robotics industry it allows them to test those results in the real world and then feed those results back to AI where it can be improved and then fed back to robotics improved, it's perpetual advancement.

  • @flareonspotify
    @flareonspotify 11 месяцев назад

    thats what naural link is for

  • @raduromanesti6408
    @raduromanesti6408 11 месяцев назад +1

    🇷🇴❤️✌️🔥

  • @michaelnurse9089
    @michaelnurse9089 11 месяцев назад

    Tesla bot has near human hands. I wonder if it can train simply from watching a video of a human do something...

  • @TheAuraEngineer
    @TheAuraEngineer 11 месяцев назад

    Solution: pay people to wear body trackers all day 💀

  • @thomasruhm1677
    @thomasruhm1677 10 месяцев назад

    You should say: Any errors are his. I just st start listening, by the way.

  • @tommags2449
    @tommags2449 10 месяцев назад

    A yes or no question.
    Will the world be a better place if billions of people loose there jobs and have no place to turn?
    Chat GPT
    No

  • @freecode.ai-
    @freecode.ai- 11 месяцев назад

    Dr. Please title your next video "SHOCKED" plain and simple. The content could be about anything but I promise the view count for that video will be 1000x any other video on your channel and you deserve it!