bullish on data companies going into training, it’s so obvious that data vendors need to understand their data and learn if it is actually good through training highly doubt that any of these "data labs" will post train anything useful or interesting but I expect that there will be some interesting research that come out in the next few months around data even internally, post training teams spend a good chunk of time trying to understand what data to train on and what capabilities the model is learning, this is work that companies supplying data should know better than us [note: any data vendor that outsources training to someone else is retarded, most of the value is in the process of training itself. secondly, training on tasks that are difficult for GLM/Kimi is quite different from what is happening at the frontier; if you think Ant/OAI don't have alpha relative to open source in how they train Mythos/Astra you are wrong] companies im bullish on to do this: - Preference Model - Fleet - Proximal - Mercor* *Mercor is making a big push here, I expect some interesting things to come out from them, but still skeptical since most of their work has been slop companies im bearish on to do this: - Afterquery - Handshake - Datacurve - Surge - Every other YC rl env co

top noise to signal companies (no particular order) - prime intellect - mercor - afterquery - taste labs - vals - designarena - applied compute (team and product is good, but why tf do they do such ngmi marketing)

this is already happening at Y-Combinator

dms open for interesting things happening will always be anon but don’t send me made up shit or will expose you for being retarded

this guy has larped his way to becoming “the data guy” >charged fees to broker data, sold shitty data and his rep at a lab got ruined >has had false info on data cos (thought mechanize did synthetic lmao) >several friends at different labs met him without exposing they work at a lab, was an asshole until they said what they do all of this and heard that he’s joining a16z as a partner to invest in data, bearish on a16z for not doing proper back channeling

half of the data bros from Fleet and Mercor liked this tweet lmao the hate amongst data players is crazy

data vendors trying to sell data to hillclimb some retarded eval

used to respect Yash until he went full marketing retard on the timeline selling out is not cool

never worked with him but >probably smart but washed, hasn’t been at a frontier lab for 2 years >left the pursuit of AGI during golden period to start some ngmi shit for 2 years >instead of going back to the pursuit, joins fucking mercor >prob paid a bag tho overall this is not what mercor should’ve done, they went clout chasing instead of just hiring solid people to train models on their data

not data related but my friends and I once met this henry lien guy and was the most obnoxious arrogant kid until he learned we were frontier lab researchers, since then there’s a running joke amongst researchers that this kid is a piece of shit data vendor founders give similar vibes btw

the slop that datacurve puts out... wrote an article to explain something that 2-3 sentences could explain, zero information density what was the point of this?

yc funding 10+ data vendors in each batch is the most anti-signal for yc 90% of them don't even make it out of the batch and the other 10% pivot a year later

most insufferable data vendor is taste labs the most larp company cannot believe they got funding > data is shit > founder is retarded > team has no taste

data lab is such an overused term how many of these data labs have ever even ran an ablation much less trained anything interesting other than using tinker don’t forget about how afterquery and mercor pay millions to others to train models for them cause they can’t do it