'I'm a Doomer': OpenAI's New Interim CEO Wants to Slow AI Progress Down | Hacker News Reader

'I'm a Doomer': OpenAI's New Interim CEO Wants to Slow AI Progress Down

vice.com

29 pointsisaacfrond56 comments

No, AI is not going to kill us anytime soon. Yes, LLMs are not even close to AGI and the research on them does not lead to research on understanding cognition (read Chomsky on this), which is required for AGI. Yes, DO ban all AI research and investment until society as a whole, by consensus, grants the privilege of continuing. OpenAI saga already has shown Sillicon Valley tech culture is in no fit shape to be involved in such manners in any form. The culture that gave us Musk buying Twitter and torpedoing its value by half the next day and the shitcoin scammer SBF does not have a good track record.

>the research on them does not lead to research on understanding cognition (read Chomsky on this), which is required for AGI.

Many phenomena have been discovered unintentionally and/or without understanding them. Deliberately developing an agi would probably be easier with a better understanding of cognition, but it's almost certainly not required.

With all due respect, Chomsky has no intuition for this kind of technology. Was sad to see him reject the technology, felt oddly like sour grapes when really LLMs have validated his work on universal grammar.

> No, AI is not going to kill us anytime soon.

> Yes, DO ban all AI research and investment

So do you think it's a threat or not?

LLMs are not even close to AGI but they fake it enough that we can be pretty sure humans will allow them to take arbitrary decisions without further control.

Will LLM kill us all and rule the world, no. Will LLM kill some of us because someone decided that they are cheaper at taking vital decisions ? I’m pretty sure it’ll happen.

Your comment is like a random collection of sentence fragments that add up to nothing. You want to ban Silicon Valley from being involved in software projects regarding large amounts of compute? Who's more suitable exactly? The Amish?

You can't ban compute. Here: 5 + 5 = 10. Might as well ban writing, or drawing.

The cat's out of the bag. Now that we all know AI works, every country in the world will have theirs. No one has a monopoly on math, no one can regulate it. Especially the US and especially Silicon Valley. That's what they do. Silicon Valley brought you the personal computers, the Internet & web, the smartphones and now AI.

The military brought me the computers, the internet and the smartphones. Sillicon Valley merely commercialized the government funded research. And yes, this commercialization has brought many benefits. But that tap has long went dry. Sillicon Valley innovation now is Juicero, iPhones that change from square to rounded corners every two years and enshitification; not ground breaking innovation.

And as for the compute bit of your comment, yes you can. You can outlaw financing of these ventures. You can prosecute the researchers for doing the research. We have these mechanisms. There's ethics boards.

General Steve Jobs. Fearsome fellow.

I wonder how deeply the board and the rotating CEOs actually understand the technology and practicality of achieving AGI, or whether they are falling over themselves due to buying into the LLM hype.

Killing us in a nuclear holocaust is for now a very far fetched hypothesis. I surely hope Nuclear C&C networks are properly air-gaped from the internet, and that actually launching the nukes require humans in the loop (yes, humans that could be socially engineered by an uber smart AGI, but then we are reaching).

The immediate danger from UI is on the economic and political realm. Reinforcing wealth-concentration, monopolies, all powerful surveillance and propaganda.

I care more about those immediate dangers, that are related to how humans that control the AI can use it for their benefit versus the rest of us, than to the distant theoretical scenario of an AI going crazy and deciding to go all skynet on US.

I fear the AI owners, they are going to fuck us way before the AI has any idea about fucking humans.

The next few years are going to be interesting for lawyers. I imagine all kinds of lawsuits including, but not limited to, stealing of trade secrets are going to started by OpenAI foundation against Sam's new employers and all the retaliatory counter-suing.

This hire makes is clear why MS is taking the IP and breaking ranks w/ OpenAI. He wants to slow down AI to development to 1/10th of its current pace? You can't run a company like that. He could run a library research program but that's about it.

MS doesn't want to slow down AI; they want to own it. Many people realize that AGI isn't a thing and LLMs will never get there. Computers lack volition, but corporations have in it spades.

Why I feel that China will not slow down AI progress?

They won’t but mainland China hasn’t shown a very good track record of keeping up, much less inventing new methods that gain any sort of traction.

They do seem to produce plenty of good researchers, who sometimes proceed to do good research with proper funding from western institutions.

I would be more worried about Europe.

> I would be more worried about Europe.

Not really a concern. The AI phobia here in Europe is pretty bad, to the point where governments are shooting themselves in the foot by legislating progress on advancing tech out.

I really don't think you're right there, many new models come from companies like Tencent and WeChat. Chinese investment in AI is increasing massively.

A huge percentage of AI researchers are Chinese, if PRC can convince them to work there instead of the US they won't have any trouble.

Not only are a rather decent number of AI developers and researchers who are Chinese, perhaps, and likely, some of the best in the world. but not only this but they are also the only manufacturing center for the GPU/CPUs needed to run the things. Tight collaboration between AI researchers and the manufactures of the hardware? No other country (absent their direct neighbors) have any chance at growing manufacturing facilities fast enough to catch up, especially with that combination.

> Not only are a rather decent number of AI developers and researchers who are Chinese, perhaps, and likely, some of the best in the world

I didn't dispute that. In fact, I brought it up.

> only manufacturing center for the GPU/CPUs needed to run the things

they don't have the designs and sanctions by American companies are surprisingly already in place and historically effective.

Finally, you are presupposing that "slowing down" in America means moving slower than China. Not only is that not necessarily the case, I don't think it's even really possible. This research can be done on readily available gaming GPU's if needed. Do you really think the US will intentionally cripple these markets in this fashion? More likely, companies will lobby for the opposite.

Any slowing down done by the US will amount to virtue signaling.

It is infuriating how the West keeps sabotaging itself. Both China and Yandex/Russia are moving ahead while the West is self-sabotaging with all the "responsible AI" bullshit.

Does China have similar level, viable models?

Yes. Yi-34b is competitive with Llama2-70b which is often competitive with gpt 3.5. GPT-4 is on its own level atm but the point is that China has already created a good model that effectively beats Meta's current best.

If they try to keep the same pace of LLM advancement to preserve a high valuation, they might fail - LLM training costs skyrocket and it will get harder and harder to get the data. The CEO announcing a slow-down could likely be a strategy not to spook investors.

Reply on news.ycombinator.com