allseer (@liquidgaviscon) header
allseer (@liquidgaviscon) profile picture

market crash march 2027 weak AGI 2028 AGI 2030 ASI 2033

San Francisco, CA
Joined September 2022
Photo by allseer (@liquidgaviscon)Photo by allseer (@liquidgaviscon)Photo by allseer (@liquidgaviscon)Photo by allseer (@liquidgaviscon)Photo by allseer (@liquidgaviscon)Photo by allseer (@liquidgaviscon)
Intelligence requires the careful chiseling of the brain to fit your data intelligence = log data this cannot happen overnight or quickly. the complexity of the eye, or performance of human danger levels comes from millions of years of evolution interacting with the world as one’s own entity. not as something you can quickly RL
17
The human brain has been chiseled for millions of years. We cannot simply replicate this in the blink of an eye.
3
A Ferrari is 10X faster than you on a racetrack, but you're infinitely faster than it crossing a jungle. The problem with the fast takeoff crowd in AI is that they think the world is a racetrack, when in fact it's more like a jungle.
129
274
38
3,140
71,994
Of course this is coming from a company that is ABOUT TO IPO IN 2 DAYS AND WANTS IT TO GO WELL
The METR graph, updated:
Readers added context they thought people might want to know
This chart is not the official METR graph. It was created by Benjamin Todd via extrapolation from Anthropic's internal AECI scores, as METR is no longer updating its public time horizon measurements. x.com/ben_j_todd/sta… benjamintodd.substack.com/p/ai-trend-ext… metr.org/time-horizons/ metr.org/horizon-chart-… metr.org/blog/2026-1-29…
6
Fear of RSI is timed suspiciously close to Anthropic IPO. A stock will always peak when the world has irrational mania. It’s not an accident.
19
OHH All these RSI fanatics just want their bag They don’t actually believe in RSI deep down
1
6
Because they lack a theory of AGI, the doomers just rely on extrapolation of current trends. All the same boring talk/pod/lecture. Yes, plotting a line has some value, but is very shallow and risky form of prediction (a lazy approach that reminds me of Thiel’s concerns with “muddling through”). Give us a theory, and at least put some skin in the game.
Episode dropping tomorrow with @paulfchristiano. He thinks there’s a 1 in 3 chance of extreme superintelligence within a year (absent any slowing down). And he'll explain what labs and governments should do right now given that possibility.
10
2
2
25
5,491
What if AI lets smart people who normally wouldn’t make games make them, and we just get really good games? Like the inverse animator effect, where animators used to be smart but aren’t anymore.
a new arena and two new characters
1
34
When will we get to the point where there is no bowing out shape? That would be the human brain
今のフロンティアの事前学習って、ここまで安定してるんだな…… (Qwen4)
1
35
As long as we use pretraining, LLMs will only asymptote, not exceed, human intelligence.
9
Jensen: “I think intelligence is a commodity. Intelligence is a word we’ve elevated to very high form. The word we should really elevate is humanity, character, compassion, generosity. Don’t let this commoditization of intelligence cause you anxiety. U should be inspired by that”
3
25
10
255
48,954
The mania about how Anthropic will produce superintelligence within a year indicates we are at the tippy top of their stock value. It will soon come crashing down.
14
Intelligence = the co-evolution of data and algorithms RSI only accounts for the latter, not the former.
7
Hot take on why RSI will not lead to superintelligence: 99% of intelligence requires interacting with the real world, either socially or physically. Without using real money, having a real body, or living with its own identity, AI will create fake games that resemble the real world but are totally reward hacked. Just think about human evolution. Each small learning (socialization, learning to pick up sticks dexterously) requires hundreds of thousands of hours of real world learnings before making it into our DNA, even if done as efficiently as possible. And each new evolution happens in sequence, one generation at a time. This will dramatically slow down superintelligence. RSI will produce toy models of intelligence that are useless in the real world.
17
RL produces a hollow shell with zero substance.
9