Luke Chaj

person sim on an organic substrate

[Luke Chaj @luke_chaj]

If AI forecasters are already superhuman, does it mean markets become truly efficient and there are no more Warren Buffetts, or does everyone become one?

At least capital allocation will not be an issue, and money will find every bottleneck.

Capital allocation singularity.

[Dan Schwarz @dschwarz26]

AI forecasting is now approximately superhuman. Today, FutureSearch is exiting our public beta and launching to everyone.

FutureSearch is the original AI forecasting company, started in August 2023. We’re currently #1 of 194 in the most competitive AI forecasting tournament, and we score above the #3 and #2 human forecasters in the premier mixed human-bot tournaments. We’re beating the crowd on Kalshi with a pure forecasting strategy, all our forecasts and trades there are public.

Thousands of people used the beta and ran >10k high-effort forecasts. Ask it anything about the future! We now support decision forecasts too: “If I do X, will I achieve this outcome?”

This video shows the part we’re proudest of: world modeling. Forecasts draw on a persistent latent representation of the future, and we’ve shown it improves accuracy. The more your forecast on a domain you care about, the higher accuracy you should expect.

It’s free to try. https://t.co/e0JKSLzxVZ

[Luke Chaj @luke_chaj]

Markets are pricing it in, but only partially.

If consumers believed in mass job disruptions, they would hold onto their money much more.

Does this mean that the human AI augmentation narrative is holding up?
(I believe it's false to be clear)

The future will be cool. The meantime, not so much.

[Jerry Tworek @MillionInt]

It is mindblowing to me how few people realize that their lives and everything they know will change drastically in the near future.

At this point, it should be pretty clear.

[Luke Chaj @luke_chaj]

$300B ARR for all LLM and video model serving is doable by EOY

[Nick Dorsey @Midnight_Captl]

I heard from a trusted source that Anthropic’s ARR as of mid July was $80B

At the pace they’re on there’s a chance they crack $100B ARR by the end of August

[davidad 🎇 @davidad]

if your definition of “AGI” is “better at most tasks than per-task expert humans” (back in the day, we used to call this “ASI”), that is coming next quarter

[Luke Chaj @luke_chaj]

Math sees the biggest progress because it's easiest to verify. Programming is more cpu heavy.

For bio, we need organ/organism on chip. Entire wetware centers of them.

[Noam Brown @polynoamial]

An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science.

We believe it will be a major step for scientific reasoning. https://t.co/iP6cyheZ7i

[Luke Chaj @luke_chaj]

@BlueWhaleFlys @hive_echo @robertskmiles how do you define "1 brain step"?

[Luke Chaj @luke_chaj]

@BlueWhaleFlys @hive_echo @robertskmiles transformer doesn't have limits in the sense that there are things it cannot do. https://t.co/YzJDoFn3RK

[Luke Chaj @luke_chaj]

I believe that the original discussion was about the limits of "understanding." I assumed that we cannot know, because we are part of the system that we are modeling, and our internal model is a compression by definition. You are speculating about the limits of transformers, for some reason.

[Luke Chaj @luke_chaj]

@BlueWhaleFlys @hive_echo @robertskmiles I am not sure what is your argument here. If you are trying to say that brains are different than transformers there is no doubt about that obviously.

[Luke Chaj @luke_chaj]

@hive_echo I suspect they have problems with the parallelization of RL rollouts, so it makes them push in all dimensions. It may change when AI discovers something radically different, maybe some skill based model merging technique.

[Luke Chaj @luke_chaj]

@hive_echo When using this constrained approach, hundreds of hours tasks would be doable this year. Even making it a little bit dumber looks like a good tradeoff.

[Luke Chaj @luke_chaj]

@hive_echo They will probably develop the next model across many dimensions, but I would really like to see increases in agency and long time horizons without making it smarter, especially when factoring in the latest hacking sprees.

[Luke Chaj @luke_chaj]

@hive_echo I assume you mean: https://t.co/jbnZuq55sP

Probably millenium prize level problems will be approachable in 1-2 years.

[Luke Chaj @luke_chaj]

@BlueWhaleFlys @hive_echo @robertskmiles I do not believe this line of reasoning makes any sense. It's not important if it's an LLM, a brain, or a graph based database system. The process used to create hierarchical pattern detection is also irrelevant.

[Luke Chaj @luke_chaj]

@woke8yearold we are ancestors

[Luke Chaj @luke_chaj]

@hive_echo solved 😎

"A set of axioms of complexity N cannot yield a theorem of complexity substantially greater than N"
https://t.co/jyYSUJPsNb

[Luke Chaj @luke_chaj]

@hive_echo I think I know. we just cannot know it, it's 100% formalized already, it must be

[Luke Chaj @luke_chaj]

@hive_echo something struck me when I was rotating this way in my mind but it will take time to unfold it

[Luke Chaj @luke_chaj]

"My view is that the “smoothing of the ball idea or intelligence not being indefinitely scalable idea “ is wrong"

It is wrong, but I assume you mean the part in which it relates to the universe that we are in, it doesn't look like it can be measured without empirical approach. It's more about compressibility of our universe than about any process that compress it when being inside it.