
founder @hud_evals + @datavendor_ai, prev cs+physics @columbia, @ycombinator. inside the sim2real gap
SF
Joined September 2022
- Tweets1.4K
- Following1.4K
- Followers2.9K
- Likes6K
everyone is talking about raw data brokering like its for absolute scum and that making rl environments is for enlightened builders.
a year ago making rl envs was for scum and making a gpt wrapper was enlightened.
do what is low status.
when people ask me what data is getting bought right now, the answer is literally: everything.
liquidity / volume / scale are def a question, but a think at this point i've seen even the most ridiculous stuff get bought.
Jay retweeted
We're partnering with @DreamscaleLabs to evaluate frontier robotics models in the cloud.
We explore running VLAs in parallel simulation, serving heavy World Action Models from the cloud, and real-time inference on physical robots.
hud.ai/research/evaluating-r…
Certain SaaS can actually go for way higher than $2m
This is probably the golden age of selling engineering context.
Data brokers are now buying up SaaS data sets for anywhere from $100K to $2M
SaaS are struggling now and some extra money to fill the gap by declining revenues is nice
That data is then re-sold to AI training companies and AI frontier labs for the exact purpose of integrating whatever that SaaS did into the AI frontier lab's app
Pretty obvious to see where this is going
watching these new data brokers talk about what type of data they buy is sort of like watching a virgin lie about having sex.
they talk about the most vague, illiquid supply like gas station data and then point you to a broken vibe coded form.
dvst
I tried this for making slides and it beat codex.
Excited to see this finally launch.
We are releasing Ultramarine 1 Flash, a frontier-class model built for knowledge work. It is open-source and achieves similar performance to Fable 5, while being 60x cheaper.
It was post-trained by a team of 5 for ~$300k all-in. Available on API now, with weights releasing in 2 weeks.
if you want to pivot to selling data: @datavendor_ai
xitter.cf/ColinGardiner/status/2…
The data broker/lab/reseller/aggregator/marketplace pivot is real.
I have so many receipts in my inbox of people pitching one thing, and now, a few months later, they've launched as a data provider of some sort to AI labs.
Becoming a crowded trade and in theory should push prices down on the average.
Jay retweeted
Lots of folks focus too much on models and algos, when data is often the bottleneck!
To understand what a good RL task looks like, check out our blog :)
When building RL tasks and evals, it's difficult to distinguish between actual model capability failures and bad RL task design.
We're sharing some our task design principles so you can write better tasks before ever evaluating a model on them.
hud.ai/research/what-is-a-go…
Congrats @Meta for an amazing model release!
xitter.cf/finkd/status/210968930…
autonomous labs are in its first inning. super bullish on this team.
xitter.cf/navvye/status/21085996…
We built an 8000 sqft. high-throughput wet-lab in 30 days, to solve molecular discovery - starting with pesticides.
Most teams in drug-discovery don’t receive target-organism verification until clinical trials. Pesticides allow us to rapidly validate our model in the target organism directly. With the help of our high-throughput lab, we were able to generate compounds that killed a $13 billion pest, the Fall Armyworm.
We believe that the bitter lesson is coming for biology.
Some designs for @datavendor_ai -- in preparation for upcoming release.
Most “data brokers” are actually “data brokies”.
Many are telling on themselves based on the data they say is valuable.
dvst
Jay retweeted
i've only grown in my conviction that as we reach narrowly superintelligent models, scarce data is of absurd and increasing value

















