Brute-forcing AGI by scaling usage
Joined April 2026
- Tweets259
- Following265
- Followers9
- Likes352
TOKENMAXXING retweeted
At first i thought ETH being false implied all of NP is sub-exponential-time.
The argument is simple: reduce any NP problem to the NP-complete 3-SAT, then solve subexponentially.
But it turns out this doesn't work, because reductions can introduce polynomial factors. If an NP problem reduces to a cubic-sized 3-SAT, the best known solution is still exponential.
Breathing a sigh of relief for now, but I doubt this is the last innovation here, now that the barrier has been breached.
TOKENMAXXING retweeted
fake sf autism larping vocab tutorial:
compute constrained <-> dumb
orthogonal <-> unrelated
sample efficient <-> fast learner
out of distribution <-> weird / unlikely latent <-> hidden
sufficient statistic <-> relevant parts of observation
alpha <-> advantage
sigmoid <-> s shape
marginal <-> small
pareto frontier <-> improving one aspect degrades another
distillation <-> regurgitating will depue tweets
The US should do the opposite of banning Chinese models. If anything they should invade countries that don't give us their models open source
.@sequoia has been a content goldmine recently. they have such great interviews and videos
i’m not sure how a16z has failed to keep up given that they have made such a big bet over many years on their media strategy
based on my preliminary tests, gpt 5.5 pro is much better for shopping/product recommendations than fable 5
I low key think gell man amnesia is fake
the idea is that you read an article on a subject that you know about (like ai) and you think the journalist is horribly misinformed, but then you read an article on the subject of something you are not an expert in (presumably politics) and then you believe them
supposedly this demonstrates a deeply irrational behavior due to a cognitive bias. "if they didn't know X, why should they know Y??"
but in reality, journalists know a lot about politics and are essentially political actors/experts. so it is worth trusting them to some extent. but of course you would be stupid to fully trust political commentary without considering their political bias and incentives, but that goes without saying
it also ignores the good journalism that is done all the time on subjects you know about. maybe it doesn't stick out to you, but it exists. in fact, THAT is the real cognitive bias at play
fable 5 came up with a genuinely creative breakthrough by building on my prior work and pulling in some obscure concepts that I had never heard of but are a perfect fit for taking my project to the next level
good take
There are two pillars of the US economy:
- OpenAI and Anthropic
guess what happens to markets when they start pricing in a slowdown in Anthropic's growth and the risk of frontier models generating 0 revenue
all while the Strait is still closed and inflation is starting to ramp and the SPCX pump and dump is going on
if this situation isn't resolved until next monday I'm just buying long SPX puts (no financial advice)
the timing is also pretty sweet with quarterly options expiry on thursday
why can’t we have at least one relatively minor ai induced crisis before we start regulating it
feels like we’re jumping the gun here
Fable 5 is so much easier to talk to than GPT 5.5 Pro. Technically, both give equally good answers but reading GPT’s output is a slog and prompting is more difficult
