Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

ChatGPT for Robotics

microsoft.com

Post reply on HN
Print view View original

41–50 of 100 posts

Re: ChatGPT for Robotics

#41

Post by bamboozled »

actusual wrote: post #40
bamboozled wrote: post #37

So ChatGPT can’t do basic maths or can I get it to write functional code most of the time, but we’re going to have it control expensive pieces of hardware ? I don’t get it ? I wish I did…

I think "get it to write functional code most of the time" severely discounts the value of the code produced by ChatGPT. Knowing zero Swift (but being an SDE), I was able to build an audio plugin for Logic Pro X in a day, by leveraging ChatGPT. I'd previously tried this twice, and gave up because of the learning curve and my lack of free time. It's the most insane 0% to 80% tool I've ever seen.

That’s cool.

I’m personally looking forward to when I can use a ChatGPT yo build my own ChatGPT and make tons of money out of it too !

It’s pretty interesting because soon absolutely No business will be safe. All tech, most apps, everything can just be stolen as fast as ChatGPT can write it. I’d go as far to say as ChatGPT itself is vulnerable to this ? Am I wrong ?

I’ve been working on a startup and I’m actually really evaluating if it’s worth the time now. I mean it could be stolen pretty fast. Not sure how to reconcile it all yet.

Make my an app which is functionally the same as Netflix, then write the deployment code so it runs on AWS. Ha.

Microsoft and Open AI will may benefit a lot from ChatGPT, but it might eat itself too.

Re: ChatGPT for Robotics

#43

Post by JamesSwift »

_boffin_ wrote: post #19
JamesSwift wrote: post #12

Any insights into how this is able to generate the Microsoft logo SVG near the end of the article? I feel like this would be a nearly impossible task unless there was another layer of knowledge on top of the language model. I'm leaning toward the example being gamed in some way (e.g. retrying until it happened to work). EDIT: also notable is that the first ChatGPT answer decribes the color as 'orange' but the SVG ans…

I think they stuck that on there

I understand the rendering is an out-of-band addition. But the question is, if this is a statistical language model (i.e. a monte carlo word generator), then how is it able to go from the description to the SVG successfully without understanding things more logically.

My best guess is that theres an actual example in the training corpus that its pulling from. Or that this is being gamed in some way.

Re: ChatGPT for Robotics

#44

Post by elforce002 »

bamboozled wrote: post #37

So ChatGPT can’t do basic maths or can I get it to write functional code most of the time, but we’re going to have it control expensive pieces of hardware ? I don’t get it ? I wish I did…

I think they're trying to capitalize the hype around it before Trough of Disillusionment (hype cycle).

Re: ChatGPT for Robotics

#45

Post by MichaelRazum »

overspeed wrote: post #10

The microwave example is very interesting. The prompt only asked to "warm up my lunch" without specifying how. SayCan[1] generated step-wise high level instructions using LLMs for robotic tasks. This takes it a step further by converting high level instructions to low level actions almost entirely autonomously. [1] https://say-can.github.io/

Looks great, but why are most of the robotics videos 10x slower then real time?

Re: ChatGPT for Robotics

#46

Post by MichaelRazum »

polygamous_bat wrote: post #26

Relevant work (mine!) from last year: https://mahis.life/clip-fields/ We showed how to use Vision-language models like CLIP in conjunction to language models and object detectors to perform open-lanhuage navigation in a household (so "warm up my lunch" goes to the microwave.)

Looks cool. Wher did you get the robot? Is it possible to buy the hardware?

Re: ChatGPT for Robotics

#47

Post by JamesSwift »

zootreeves wrote: post #32
Jensson wrote: post #21

Earlier quoted context omitted.

By hardcoding tasks, like old assistants. If they don't have proof that it isn't hardcoded then it is almost surely hardcoded. For example Google Duplex, I remember many being very impressed by this: https://www.youtube.com/watch?v=D5VN56jQMWM That entire conversation was just hardcoded with some if-else statements depending on what the other part said. The tech isn't as impressive when you know that, and anyone who…

It doesn't quite manage the google logo https://imgur.com/69t6bfK

lol yeah I rendered that and its not even close https://imgur.com/eFhMfYG

Definitely makes me think something is up. I wonder if they primed the model with some extra microsoft content to get better demos.

Re: ChatGPT for Robotics

#48

Post by polygamous_bat »

MichaelRazum wrote: post #46
polygamous_bat wrote: post #26

Relevant work (mine!) from last year: https://mahis.life/clip-fields/ We showed how to use Vision-language models like CLIP in conjunction to language models and object detectors to perform open-lanhuage navigation in a household (so "warm up my lunch" goes to the microwave.)

Looks cool. Wher did you get the robot? Is it possible to buy the hardware?

It's a hello robot Stretch [1], which costs ~$20K.

[1] https://hello-robot.com/product

Re: ChatGPT for Robotics

#50

Post by ChatGTP »

elforce002 wrote: post #44
bamboozled wrote: post #37

So ChatGPT can’t do basic maths or can I get it to write functional code most of the time, but we’re going to have it control expensive pieces of hardware ? I don’t get it ? I wish I did…

I think they're trying to capitalize the hype around it before Trough of Disillusionment (hype cycle).

I’m honestly starting to think the same thing. Obviously it’s cool tech and it feels like we’re in the verge of the singularity at times, then at other times it all feels like bullshit. Some very smart people I work with said they’ve gone off using it because it’s really not that useful and the novelty has warn off.

Ultimately, OpenAI can’t keep control of this tech alone, forever. Fundamentally they must know this. If a LLM can make anything, it can replace OpenAI and Microsoft too. It’s a double edge sword.

Probably wise to milk the hell out of it.

Post reply on HN