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

Unpredictable abilities emerging from large AI models

quantamagazine.org

Post reply on HN
Print view View original

251–260 of 326 posts

Re: Unpredictable abilities emerging from large AI models

#251

Post by shawntan »

pedrovhb wrote: post #10

This caught my attention as I found it implausible: > One DeepMind engineer even reported being able to convince ChatGPT that it was a Linux terminal and getting it to run some simple mathematical code to compute the first 10 prime numbers. Remarkably, it could finish the task faster than the same code running on a real Linux machine. Following the link, there's a screenshot to a screenshot [0] of a code-golf solutio…

I was personally quite appalled that so many people believed this was what it was doing, but I suppose since the source was a DeepMind researcher, it lends a lot of credibility.

By the same reasoning, I hope people do realise that there are computations that are just impossible given the finite depth in a Transformer. The only possible way to overcome this in the current paradigm is the Chain-of-Thought related methods.

Re: Unpredictable abilities emerging from large AI models

#252

Post by petilon »

Animats wrote: post #245
guns wrote: post #91

Earlier quoted context omitted.

This is what Stephen Wolfram concludes in a recent article about ChatGPT: > The specific engineering of ChatGPT has made it quite compelling. But ultimately (at least until it can use outside tools) ChatGPT is “merely” pulling out some “coherent thread of text” from the “statistics of conventional wisdom” that it’s accumulated. But it’s amazing how human-like the results are. And as I’ve discussed, this suggests some…

> (at least until it can use outside tools) This is key. ChatGPT/GPT-4 alone are limited to reformulating what they know from their training data. Linked to search engines, databases, and computational tools such as Wolfram Alpha, they acquire much more capability. We're already seeing that with Microsoft Bing. (Update: what happens as large language models learn Excel? Especially since Microsoft is already connectin…

> ChatGPT/GPT-4 alone are limited to reformulating what they know from their training data.

That's not true. They are extrapolating. If they weren't, they wouldn't have the problem known as "hallucination".

Re: Unpredictable abilities emerging from large AI models

#253

Post by fsckboy »

ambicapter wrote: post #240
chias wrote: post #110

Earlier quoted context omitted.

So here are a few screenshots that I personally took after telling it that it was a linux terminal: Correctly produces a convincing output for having decoded the base64-encoding of "ping google.com" and then piping it to bash: https://media.infosec.exchange/infosecmedia/media_attachment... Similar command, but with a garbage domain it hasn't seen before, and a less well-known domain. It produced convincing output in…

Lol why does ChatGPT hit ^C? Do you think it’s getting bored of waiting to respond to you and decides that’s enough time spent on an answer? Edit: the base64 decoding examples are terrifying. I have no idea how that works. Edit2: actually I can sort of see how that works, b64 encoding doesn’t have any obfuscation to it, I could see how an ML model can build a pretty good approximation over time after seeing all the e…

it's probably hitting 'caret' 'C' except when it's using Emacs when it types C-C

Re: Unpredictable abilities emerging from large AI models

#254

Post by shawntan »

chias wrote: post #110
pedrovhb wrote: post #10

This caught my attention as I found it implausible: > One DeepMind engineer even reported being able to convince ChatGPT that it was a Linux terminal and getting it to run some simple mathematical code to compute the first 10 prime numbers. Remarkably, it could finish the task faster than the same code running on a real Linux machine. Following the link, there's a screenshot to a screenshot [0] of a code-golf solutio…

So here are a few screenshots that I personally took after telling it that it was a linux terminal: Correctly produces a convincing output for having decoded the base64-encoding of "ping google.com" and then piping it to bash: https://media.infosec.exchange/infosecmedia/media_attachment... Similar command, but with a garbage domain it hasn't seen before, and a less well-known domain. It produced convincing output in…

https://twitter.com/tanshawn/status/1599442780384014346

Re: Unpredictable abilities emerging from large AI models

#255

Post by Karrot_Kream »

UncleOxidant wrote: post #248
drdec wrote: post #201

Earlier quoted context omitted.

> Certainly the rise of LLMs and generative AI isn't going to be all bad, but I've also got a feeling that not as many software developers will be needed soon since those who can leverage LLMs will experience a pretty decent productivity boost. OTOH, software development will become significantly cheaper. That means the business case for using software in more places just tilted in favor of throwing some software at…

> You'll see more and more businesses using it, ones you didn't expect, like local mom and pop businesses. And in a lot of cases they're not going to need to hire a software developer to do develop it.

Yeah definitely, but we already saw this coming before LLMs right? Low code solutions, site builders, I mean MS Frontpage is as old as the Web itself. Just like you needed top MechEs to build heat exchangers in the old days, nowadays a small team makes a design and technicians handle most of the actual work of assembly and maintenance. Likewise you'll probably just have low code engineers who glue existing parts together to make things work at your local cafe or restaurant.

One of the reasons the work at Big Tech is so fun is because they actually need to engineer a lot of things. Smaller tech companies or businesses that utilize tech can just glue a few libraries together to make it work.

Re: Unpredictable abilities emerging from large AI models

#256

Post by olalonde »

lottin wrote: post #238
olalonde wrote: post #195

Earlier quoted context omitted.

Indeed, it's fascinating to witness a sizable segment of the HN community and distinguished intellectuals like Chomsky displaying absolute denial. I've started bookmarking the comments so I can look back at them in 5 years and have a good laugh. Some from a few months back are already aging badly[0]. [0] https://news.ycombinator.com/item?id=34197033

I didn't bookmark them, but I recall comments from ten years ago predicting that 50% of the workforce was going to be replaced by AI in the next ten years. Fast-forward to "the future" and the reality is AI has left without a job a grand total of 0 people. The grandiose promises we're hearing now concerning AI are nothing new. They were laughable then, as they're now.

Those comments about employment are indeed laughable and I bookmark them as well. We're at near full employment despite the invention of mechanized farming, electricity, the printing press, cars, the computer, the Internet, etc. I was referring to people who are in denial about the rapid progress of AI, and the impact it is going to have in coming years.

[0] Until we reach AGI, but I don't dare attempt to predict when this is going to happen.

Re: Unpredictable abilities emerging from large AI models

#257

Post by selestify »

hgsgm wrote: post #227
Calavar wrote: post #223

Earlier quoted context omitted.

I see a lot of irony in this. When Chomsky proposed generative grammars, his theory of universal language acquisition, and so on, they were radical ideas that upturned the central cannon of linguistics. Time has been on his side - entire schools and subfields of linguistics went extinct as more evidence emerged that Chomsky was fundamentally right. Basically every computer language and data format in existence is par…

What's even more bizarre is the ChatGPT is proving him right , that a neural net can build logical grammar, and he is denying it!

What did he say about logical grammars and neural nets in the past? Sorry, not very familiar with him.

Re: Unpredictable abilities emerging from large AI models

#258

Post by saiojd »

datpiff wrote: post #93
danielbln wrote: post #21

Earlier quoted context omitted.

Can you elaborate on fundamental limits?

> According to NVidia, LLM sizes have been increasing 10X per year for the last few years. Clearly this cannot continue, as the training costs will exceed all the compute capacity in existence. The other limit is training data, eventually you run out of cheap sources.

I would not be so sure about compute capacity? Neural network architectures are still in their infancy, it is very likely that more efficient approaches exist.

Re: Unpredictable abilities emerging from large AI models

#259

Post by AgentME »

lxgr wrote: post #242
sebzim4500 wrote: post #36

Earlier quoted context omitted.

The number of people saying it now is many orders of magnitude more than the number of people saying it 10 years ago. Not saying that means it will happen, but it isn't the same situation.

> The number of people saying it Popular sentiment is a pretty meaningless metric for predicting the future.

Then we shouldn't use the 10 year-old (lesser) popular sentiment as a reason to discount AI either.

Re: Unpredictable abilities emerging from large AI models

#260

Post by galleywest200 »

Earlier today I was messing with the GPT-4 API -- I passed it a list of ingredients and asked it to return a JSON string of non-vegetarian ingredients, separating name and amount...And it did it just fine. This was an "awakening" moment for me as I realized just how _many_ things I can use this technology for with far less effort than I would have before.
Post reply on HN