Tenobrus calls most 'sky-high-target' neolabs self-deception or con artistry; Nathan Lambert agrees they're moonshots
Tenobrus argues nearly all 'sky-high-target' neolabs except Thinking Machines and SSI are self-deception or con artistry, giving them 1-2% odds, saying founders ignore concrete facts about frontier model trajectories; Nathan Lambert calls the businesses total moonshots. Tenobrus says the founding window has closed and there aren't 10 years to wait. Mostly a debate about lab economics and timelines, not a safety finding.
All posts · 12, most important first
- my model of flapping airplanes, and ~all "sky-high-target" neolabs besides thinky and ssi, is that they are some combination of extremely successful self deception and con artistry. there are many multi-bil valuation companies out there that have like some hope of doing some successful enterprise sales and consulting, but near zero chance of achieving their stated mission or goals. they seem to exist mostly as a wrapper for very ambitious people to feel like they're able to have more autonomy than they would at a frontier lab and for investors to feel like they maybe didn't completely and utterly fuck up by not getting early allocation to anthropic.Reposing now that the fundraise is official. I want real answers this time! https://t.co/FJbkxSp2Mm open on X29 likes · 0 reposts
- I have many friends at neolabs who I love dearly but I really think the businesses are total moonshots, some of my friends understand this2 likes · 0 reposts
- or to be more concrete i don't think it's some very abstract "timeline disagreement" where we have different models of how the scifi future will go i think there's some very concrete facts about frontier models today and their trajectories over the last six months and ignoring them is self deception or con artistry0 likes · 0 reposts
- sufficient capital and talent density to have nonzero chance of impact . like still 1-2%ish but1 likes · 0 reposts
- before ppl misunderstand me there are many orgs who are sometimes classified as neolabs and maybe in many ways are but in fact have pretty realistic goals and clearly validated products and revenue streams. i don't mean them. im talking about companies who's value prop lies in some as of yet unrevealed potential research discovery or abstractly defined alternate approach to model training0 likes · 0 reposts
- i think neolabs made much more sense about a year and a half ago, they had many more potential worlds where they could work out. and i think that window has largely closed2 likes · 0 reposts
- there aren't 10 years left to wait6 likes · 0 reposts
- i don't think that quite checks out ? i'm not saying they were universally self deceiving at inception, it's pretty reasonable to start moonshots! just at the point of like raising at multi bil valuations in 20260 likes · 0 reposts
- i have significantly less than 3 year agi timelines yes and that is indeed a large factor3 likes · 0 reposts
- and again i think there's a big difference from "company who makes models" and "sky high target neolab" in the sense i mean1 likes · 0 reposts
- no, when anthropic was founded we were mils away from RSI0 likes · 0 reposts
- i think there are ooms of efficiency out there and that frontier models will be the ones that discover it, and i think open source models have def had billions of dollars of impact on labs bottom lines, but billions remains a tiny irrelevant percentage, and said value doesn't accrue to creators of open source models0 likes · 0 reposts
@tenobrus commentary · 
@natolambert research ·