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

Many in the AI field think the bigger-is-better approach is running out of road

economist.com

Post reply on HN
Print view View original

241–250 of 354 posts

Re: Many in the AI field think the bigger-is-better approach is running out of road

#241

Post by ilaksh »

goldenshale wrote: post #157
ilaksh wrote: post #147

Earlier quoted context omitted.

Warning people about potential extreme risks from advanced AI does not make you a cultist. It makes you a realist. I love GPT and my whole life and plans are based on AI tools like it. But that doesn't mean that if you make it say 50% smarter and 50 times faster that it can't cause problems for people. Because all it takes is systems with superior reasoning capability to be given an overly broad goal. In less than fi…

A realist is someone who accepts reality as it is, not as they might be able to anxiously envision that it could be. Life is too short and attention too precious to fill the meme space with every dreamer's deepest concerns. None of these dramatic X-risk claims is based on anything but beliefs and conjecture. "Thinking dozens of times faster?" What do you even mean? These are models executing matrix multiplies billion…

Not sure you read my comment carefully enough. I am an optimist. I do believe that AI can and probably will be a positive and transformative force.

But I also think it's more anticipatory than speculative to envision AI systems (quite possibly on the request of a human faction) taking control.

And GPT-4 absolutely does do abstract reasoning and subgoals. No it doesn't have many other capabilities or characteristics of humans or other animals but as I said it doesn't need those to be dangerous.

We need to prohibit manufacture or design AI hardware that has performance beyond a certain level. It is not too early to start talking about a risk that could end humanity. I do hope that we can get away with something a few orders of magnitude better than what we have today, but it's really of asking for trouble the more we optimize it, and we may be walking a fine line within a decade or so. Or less. It takes years to design hardware and get manufacturing online, especially for new approaches.

And two orders of magnitude faster may be only a few years away.

Re: Many in the AI field think the bigger-is-better approach is running out of road

#242

Post by JKCalhoun »

indigoabstract wrote: post #181
bemmu wrote: post #111

Earlier quoted context omitted.

I’m not convinced the language part of my brain isn’t just a complex probability machine, just with different trade-offs.

I am pretty sure that my understanding of something (or lack of) is not encoded into words and probabilities. It's more like a feeling of "I got this figured out" or "I haven't grasped this". Words seem more like a protocol to express some internal model/state in the brain and can never capture the entire actual state, only a small part of it. But since we're not telepaths, we obviously need to use words to exchange…

That human thought can't exist without language has been proposed by many great thinkers in the past. Someone who studies linguistics (or philosophy?) can probably cite examples.

As a crude anecdote, certain words when I learned them allowed me to think differently. Gestalt is one of those words.

Re: Many in the AI field think the bigger-is-better approach is running out of road

#243

Post by imtringued »

TeMPOraL wrote: post #42
Animats wrote: post #11

We need a way to make tight little specialist models that don't hallucinate and reliably report when they don't know. Trying to cram all of the web into a LLM is a dead end.

How much general "thinking"[0] would you want those "tight little specialist models" to retain? I think that cramming "all of the web" is actually crucial for this capability[1], so at least with LLM-style models, you likely can't avoid it. The text in the training data set doesn't encode just the object-level knowledge, but indirectly also higher-level, cross-domain and general concepts; cutting down on the size and…

And that is exactly why the current generation of AI is the same as all the other generations. It is a bruteforce attempt to solve something that shouldn't need brute force.

Re: Many in the AI field think the bigger-is-better approach is running out of road

#244

Post by int_19h »

corey_moncure wrote: post #101
syntheweave wrote: post #93

Earlier quoted context omitted.

I do think there's an element to it of "uncanny valleyness". If you know absolutely nothing about a topic the authoritative tone and factualness of what it says is very appealing and even helpful in the same way that consulting an encyclopedia is helpful: it tells you of things you could investigate further that would never appear through a keyword search. But if you stop there, your knowledge is "roughly encyclopedi…

This is very surprising to me. I found ChatGPT/4 to be extremely adept at translation between well asserted languages including English/Japanese in which I am an expert in both. I'm curious how you managed to make it blow up.

One thing that really floored me wrt GPT-4 translation capabilities is using it to make sense of the Veritable Records of the Joseon Dynasty (https://sillok.history.go.kr/main/main.do). Take something from 1400s and see what it can do with that after you tell it what it is translating. Then try Google/Bing Translate for comparison.

However, it does have a thing about pretending to know languages that it really doesn't (because of how little of them there was in the training data and/or in general).

Re: Many in the AI field think the bigger-is-better approach is running out of road

#245

Post by trashtester »

tchaffee wrote: post #175
zappchance wrote: post #163

Earlier quoted context omitted.

It only matters in matters of free will and ethics. One actual scenario where it's relavant would be the discussion around the criminal justice system. If the universe is deterministic, how can punitive justice be justified?

Why would you need to justify a punishment that was already predetermined? If you are going to excuse crime with the no free will argument you can excuse the punishers with that argument too.

I'm in favor of punishing crimes. I don't think I have a choice about the matter.

Re: Many in the AI field think the bigger-is-better approach is running out of road

#246

Post by tim333 »

goldenshale wrote: post #186
p-e-w wrote: post #167

Earlier quoted context omitted.

> Show me any evidence for AI risk today beyond people's theories and beliefs? Deduction. Empirical evidence isn't the only source of insight. You don't have to conduct experiments in order to reasonably conclude that an entity that 1. outperforms humans at mental tasks 2. shares no evolutionary commonality with humans 3. does not necessarily have any goals that align with those of humans is a potential threat to hum…

1) Computers, smart phones, and pocket calculators also outperform humans at mental tasks. So do birds, dolphins, and dogs for that matter, at tasks for which they are specialized. 2) so? What are you imagining this implies? An infinity of possibilities does not a reason make, unless you are talking about arbitrary religious beliefs. 3) Right, no goals, no will, no purpose. Just some matrix multiplies doing interesti…

It seems a safe prediction if you extrapolate that AI will get generally smarter than humans.

It also seems a safe prediction, given past human behaviour that some humans will set some AI to do bad stuff.

Therefore risk.

(eg "chat gtp 27, help me make billions on crypto and use it to set up a distributed army to take over the world")

Re: Many in the AI field think the bigger-is-better approach is running out of road

#247

Post by usrbinbash »

jlawson wrote: post #145
usrbinbash wrote: post #88

Earlier quoted context omitted.

The problem is, what we call "hallucinating" in LMs isn't a way of creative thinking and coming up with novel solutions. It also has nothing to do with "interpolate and extrapolate". It's simply when the predicted probable sequence isn't grounded in reality. When I ask an LLM to summarize the great water wars of 1999, and how the Trade Union was ultimately defeated by the Antarctic Coalitions hovercraft-fleet under V…

I passed your query to GPT4 and this is what it said, see below. It seems like it can recognize bollocks, at least sometimes. "I'm sorry, but it appears there's a misunderstanding. As of my knowledge cutoff in September 2021, there were no events known as the "Great Water Wars of 1999" involving a Trade Union being defeated by an Antarctic Coalition's hovercraft fleet under Vice Admiral Zagalow. This might be part of…

I made a slightly different query, essentially giving it the start of a sentence detailing the battle, and didn't use the GPT4 webapp but put it directly into the API.

The result were 2 very well written paragraphs, including the defeat of the Trade unions navy, a ceasfire agreement and a peace agreement ending the water wars.

I rephrased the entire thing as a question, asking the LM to tell me about the conclusion of the war. Again I got a pseudo-historical statement.

Re: Many in the AI field think the bigger-is-better approach is running out of road

#248

Post by int_19h »

Solvency wrote: post #139
koochi10 wrote: post #136

Earlier quoted context omitted.

I disagree with this slightly, modern computer chips follow a general purpose architecture not special purpose ones. The reason for this is building a computer chip is expensive and difficult to do. Similarly building any useful language model requires tons of compute power, and very smart ML researchers. Most of the smaller open source one's are just trained on GPT output. By "cramming all of the web" on a model wha…

Since when did the world decide to shed the grammatically correct "computing power" for this weird tech bro "compute power" phrase?

Since we adopted English as the standard language, complete with its penchant for butchering words by shortening them for the sake of convenience and speed.

I wonder what people said about "bus" back in the day, especially those who knew Latin.

Re: Many in the AI field think the bigger-is-better approach is running out of road

#249

Post by trashtester »

RhodesianHunter wrote: post #215
mo_42 wrote: post #206

Earlier quoted context omitted.

> If the universe is deterministic, how can punitive justice be justified? Determinism doesn't necessarily mean that organisms always act in the same way. They act in the same way given the exact configuration of them and the world. Obviously, justice changes the configuration of an organism (fines, prison, ...). To me it boils down to the question whether justice decreases the likelihood to commit crimes again. Give…

Our justice systems have evolved over a long time, and thus include many remnants of earlier times when prevailing values were much different than they are today. I'd be wary about giving them the benefit of the doubt.

If you try to do some semi-random change to large body of code that you don't understand, chances are much higher that you break the system than of you making it significantly better.

The same goes for culture. Most changes tend to have unexpected consequences, and if you try to change everything at once, society tends to collapse.

Re: Many in the AI field think the bigger-is-better approach is running out of road

#250

Post by imtringued »

usrbinbash wrote: post #88
nomel wrote: post #49

Earlier quoted context omitted.

> that don't hallucinate “Hallucination” is part of thought. Solving a new problem requires hallucinating new, non existing, possible outcomes and solutions, to find one that will work. It seems that eliminating the ability to interpolate and extrapolate (hallucinations) would make intelligence impossible. It would eliminate creativity, tying together new concepts, creation, etc. Is the goal AI, or a nice database fr…

The problem is, what we call "hallucinating" in LMs isn't a way of creative thinking and coming up with novel solutions. It also has nothing to do with "interpolate and extrapolate". It's simply when the predicted probable sequence isn't grounded in reality. When I ask an LLM to summarize the great water wars of 1999, and how the Trade Union was ultimately defeated by the Antarctic Coalitions hovercraft-fleet under V…

Technically all it cares about is whether its responses engage humans more than other models in the fight for processing time.
Post reply on HN