| Not checkable |
Not checkable |
8 Aug 2023 |
My view has always been talent density beats talent mass.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
In fact, I suspect it's not quite there yet.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
Given you're in a war with the Nazis, I don't really see much choice but to do it if it's possible.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
The nice thing with the next token prediction is it's there for you.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think we should always maintain this attitude that it's really easy to fool yourself.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
Even us who have not been super focused on commercialization and more on safety, the graph goes up and it goes up relatively quickly.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
It's a race between how fast the technology is getting better and how fast it's integrated into the economy.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I'd think of that less in terms of a new architecture and more in terms of a loss function like the data, the environments you're exposing…
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
unitary visions for what it means to live a good life have not worked out well at all.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I guess my perspective would be, even if they don't happen, we're already on this very, very steep trajectory.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
This is unfortunately not one that I'm excited about, but I believe it's happening.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
And it's one reason I'm a bit skeptical of biological analogies.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
No. They would succeed.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
But specific abilities are actually very hard to predict.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
But what it shows is that we can get something very different from and maybe opposite to what we intended.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think the closest thing we have to that is something like mechanistic interpretability.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
We have pretty wide error bars on what's dangerous and what's not so we want to establish a norm of being careful.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I don't think we should be purely phenomenological in like, here are some brain scans of really dangerous models and here are some other…
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I don't know about fast but I'm concerned about this.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think those inventions are possible.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I don't understand it at all.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I don't know if that's a fatal flaw or if that's just the way things have to be.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
One is just that the threshold of skill isn't high enough, comparative advantage.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
That's a little how I think about it.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
The second option is you just find a way.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
Humans see so little data, and they still do fine.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
So you see this phenomenon over and over again where the scaling and the safety are these two snakes that are coiled with each other…
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
We really have very little idea what we're talking about.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think we might end up in a weirder world than we expect.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I'm pretty uncomfortable with: here's the AI's constitution, it's going to run the world.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think the truth is that we still don't know. It's almost entirely an empirical fact.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
For some reason, and it may just have been random, I was obsessed with that particular direction.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
That doesn't imply that Anthropic or any other entity should be the entity that makes decisions about AGI on behalf of humanity.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
The models do display a kind of ordinary creativity.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
Just set those thresholds very well below and then as you proceed upward from there, do kind of more and more rigorous tests and be more…
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
The main thing that would stop it would be if we hit certain safety thresholds and stuff like that.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
There aren't bits for that.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think that was probably an artifact of thinking about RL in the wrong way and underestimating how much the model had learned on its own.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
The honest fact is that we're figuring this out as we go along.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
Part of it is just that physicists learn things really fast.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
Those were the key four ones, which I think are the core of the hypothesis.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
Thanks for having me.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think of it as neuroscience for models.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
It's not that much of a concern with today's passive models.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
They feel like interns in some areas and then they have areas where they spike and are really savants, where they may be better than anyone…
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think they are just on the cusp of being able to put these things together.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
If such a model wanted to wreak havoc and destroy humanity or whatever, we have basically no ability to stop it.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
But honestly, intellectually we have to admit that that's a weird thing that doesn't match up.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I'm skeptical of any kind of precise mathematical or exponential prediction of how it's going to be.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I get nervous when someone says, what are you going to do with a superhuman AI?
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
If we take the rate at which all this is going to happen seriously, which I can't be sure of, then it does make me think that something…
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think it's going to be much less than 30 years.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I actually think fact number one and fact number two are enough to be really worried.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think that's a mix of just the model skill level is not high enough yet.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
That's not my position at all.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
Mechanistic interpretability is the only thing that even in principle is the thing where it's more like an X-ray of the model than…
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
It feels to me like an intern in most areas, but then some specific areas where it's better than that.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I think then you would have to go for some kind of RL.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I thought it had understood a lot of the essence of language but then there's further to go.
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
8 Aug 2023 |
I guess if you had the view that alignment is completely unsolvable, then you'd be like, well, we're dead anyway so I don't want to worry…
Not checkable · 8 Aug 2023 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
DeepSeek's releases don't change this, because they're roughly on the expected cost reduction curve that has always been factored into…
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
is much less interesting from an innovation or engineering perspective than V3.
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
the idea that the technology getting more powerful, having more bang for the buck, is a reason to lift our export controls makes no sense…
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
I don't see DeepSeek themselves as adversaries and the point isn't to target them in particular.
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
Export controls serve a vital purpose: keeping democratic nations at the forefront of AI development.
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
Well-enforced export controls are the only thing that can prevent China from getting millions of chips, and are therefore the most…
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
This shows that the export controls are actually working and adapting: loopholes are being closed
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
It's just that the economic value of training more and more intelligent models is so great that any cost gains are more than eaten up…
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
The goal is to prevent them from gaining military dominance.
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
All of this is to say that DeepSeek-V3 is not a unique breakthrough or something that fundamentally changes the economics of LLM’s; it’s an…
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
Claude is extremely good at coding and at having a well-designed style of interaction with people
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
$1B of economic activity can be hidden, but it's hard to hide $100B or even $10B.
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
I think they make export control policies even more existentially important than they were a week ago
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
The performance of DeepSeek does not mean the export controls failed.
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
another factor of 10 may correspond to the difference between an undergraduate and PhD skill level
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
I do believe many of the claims about their threat to US AI leadership are greatly overstated
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
Everyone should be able to benefit from AI.
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
On these and some additional tasks, there’s just no comparison with DeepSeek.
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
13 Oct 2022 |
But we shouldn't hand the Chinese Communist Party technological advantages when we don't have to.
Not checkable · 13 Oct 2022 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
Biology is probably the area where scientific progress has the greatest potential to directly and unambiguously improve the quality of…
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
AI-powered authoritarianism seems too terrible to contemplate, so democracies need to be able to set the terms by which powerful AI is…
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
there’s an unavoidable constant delay
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
I think this question is more difficult than the others.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
A truly mature and successful implementation of AI has the potential to reduce bias and be fairer for everyone.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
I have not followed the field closely in recent years, but I have a vague sense that computational neuroscientists have still not fully…
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
I am not as confident that AI can address inequality and economic growth as I am that it can invent fundamental technologies
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
We have the opportunity to play some small role in making it real.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
The economic piece actually seems more difficult to me than the meaning piece.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
It is much easier to do experiments on artificial neural networks than on real ones
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
I’m including lots of details mainly because I think a concrete vision does more to advance discussion than a highly hedged and abstract…
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
exploitative or dystopian directions are clearly also possible and have to be prevented.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
it is the first technology capable of making broad, fuzzy judgements in a repeatable and mechanical way.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
I think it is very likely a mistake to believe that tasks you undertake are meaningless simply because an AI could do them better.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
At a guess, the most important thing that is needed might be reliable, non-Goodhart-able biomarkers of human aging, as that will allow fast…
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
Some may not even consider it desirable; it embodies values and political choices that not everyone will agree with.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
I think many will be literally moved to tears by it.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
I think it’s dangerous to view companies as unilaterally shaping the world, and dangerous to view practical technological goals in…
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
Intelligence may be very powerful, but it isn’t magic fairy dust.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
But I think that pessimistic perspective is thinking about AI in the wrong way.
Not checkable · 17 Aug 2015 |
|
| Not checkable |
Not checkable |
17 Aug 2015 |
My current guess at the best way to do this is via an “entente strategy”
Not checkable · 17 Aug 2015 |
|