Prime Intellect

It scored 64% on their cyber index, 11 percentage points ahead of DeepSeek V4.1 Flash, and only 5 percentage points behind the trusted version of GPT-6 Sol available through their partner program, Daybreak Blue.

Eli Bukowski from Prime Intellect explained why this was a positive sign.

He wrote, "...one point that was under-discussed in my opinion is that it's the first open lab that is pacing the frontier of open-weight cyber capabilities, which makes it much easier and creates incentives for other labs to make the right decision and do the same when they reach the same level of capabilities." The remaining question is how to set the right level.

Underestimated and open models will become useless for defensive cybersecurity, leaving the world relying on US closed labs to spread defensive capabilities quickly, which is not realistic in my opinion, although I obviously wish them to succeed in this mission.

Time for my grim fact.

Clef's host, a company called Prime Intellect, was quietly cutting the state down to 16,384 tokens and billing for exactly that many.

So the model answered from the first part and never saw the rest, and the people running it only found out later.

Anyway, there's a casserole in the oven, and I haven't done your laundry, sweetheart, so tell me when you're coming to visit.

What were the three AI companies?

Uh, one is a company called Prime Intellect.

Oh, yeah, yeah.

Prime Intellect was, yeah, built around distributed compute initially and does a bunch of stuff for post-training-

Mm-hmm

... um, open source models now.

Mm.

It's like, you know, Vincent Weiser from, you know, Prime Intellect or Runa Quist from, you know, AI underwriting company- I just met him

... or Luke Drago.

He's great.

Yeah.

So, um, I'm interested in, like, what are s- some of the ways that, uh, like, founders of today, what do you find inspiring of, like, certain founders that...

Like, we have Prime Intellect.

We have, like, uh, Luke Drago of Workshop Labs.

His claim is that the models are already geniuses, and we have been managing them all wrong.

[digital sound] Alex Zhang is a second-year PhD student at MIT, and his paper on recursive language models has already been built into Prime Intellect's Agent, into Harvey's legal AI, and into the harnesses people threw at the Arc AGI3 benchmark.

In April, he and his advisor, Omar Katab, wrote an essay whose whole summary fits in one line: AI models are already good enough for the next leap in capabilities.

They call it the mismanaged geniuses hypothesis.

Like there is still an efficiency aspect of, of all of these things that is super, super important.

[dramatic music] Let's slow down because this is where it gets strange.

Zhang helped build Prime Agent with the people at Prime Intellect, a harness whose only tool is a Python notebook, and one of its pieces, the continual harness, came from his fellow PhD student, Seth, who built it to get language models to play games.

So what is the harness allowed to touch? And continual harness is also, by the way, very simple.

The first reason is you start to have a lot of teams that are doing that for us.

Our friends at Prime Intellect, their main job is doing that.

Trajectories, they're basing.

A lot of teams are starting to help companies doing that.

Let's see what else happened today.

Prime Intellect, a company that rents out AI computing power and now serves AI models directly, launched a service called Prime Inference this month.

It already handles almost a trillion tokens a day.

The chunks of text a model reads and writes, mostly for reinforcement learning and coding agents.

Has he tried eating bleach? yeah it's fucking crazy dude and dude what's this graphic about sexless adults

oh yeah I mean this guy here was dunking on me he's some he actually is the founder of some company or maybe he works with some company called Prime Intellect which dope name for a business but he was just like kind of parodying me saying I'm not joking when I say this Tinder hinge and dating in general is over porn is now very very good you just don't know it yet because all porn circulating today is from two to three generations ago Pornhub has solved infinite personalized gooning So he was trying to dog on me, right? But I read the post and I was like, well, isn't that exactly what's actually happening right now? It's not solving personalized gooting, but isn't it actually negatively impacting the number of people that are choosing to have sex?

And then I went and I found

a bunch of stats and