b-squared @bbopen

There's so much alpha in synthetic data generation and model fine tuning though for those with the domain knowledge who know what they're doing. Or even if you don't, use these coding agents to guide you rather than creating api wrappers or modeless software. Even if you're AGI-pilled it is safe to say if you build a good specialized model instead of software that is core to your app, you stand a chance to make it.

1 0 2 199 0

View Details

b-squared @bbopen

11 months ago

Ken, I’d say nature isnt always the optimal but offers “at least one” prescription for learning. I listened to this pod on an overseas flight yesterday and started cooking with @grok because Rich’s squirrel example didn’t sit right with me (parental imprinting). Rather, I started thinking about how an octopus is solitary and learns independently from birth. Fascinating creatures (see the doc My Octopus Teacher). No parent demonstrations, just play, a learned sense of how the world works, and careful use of imagination. In the wild it looks, reaches, tastes, and retreats if something feels off. I suppose in isolation and in unfamiliar terrain we humans have similar behaviors. Code: paste.rs/ZQh8s.py What I ended up with is a concept of evolving exploration habits so the agent tries lots of different ways to poke the world. Part of it inspired by your Novelty Search. What's different from prior art is collective pressure evolution is used to shape the data the learner sees, the model governs which “imaginations” of the agent count so imagination helps rather than hallucinates.

2 0 2 152 0

View Details

b-squared @bbopen

11 months ago

@ludwigABAP He had a paid ad read for Google saying he prepped for the interview by using Gemini to read about RL

1 0 2 475 0

View Details

b-squared @bbopen

11 months ago

@ashleevance @CAForever @corememory I live in Solano County, work in AI, and support Jan & CF's vision. I'm looking forward to moving and starting a business there. I'm trying to share this enthusiasm with my neighbors, many of them have kids who've had to move away from CA for opportunities that CF would provide.

1 0 3 140 0

View Details

b-squared @bbopen

11 months ago

@scottstts @omercheema Based on what? Coding?

1 0 0 45 0

View Details

b-squared @bbopen

11 months ago

@altcap @Aidanbeltskeys @BG2Pod @bgurley I’ll take the other side. I learned some things about OpenAI in enterprise. We hear and experience a lot of the consumer side, but enterprise is the place where there’s the largest opportunity but noisiest signal.

1 0 2 57 0

View Details

b-squared @bbopen

11 months ago

@melissa 36 month old got the Dimensions PKA the other day. She likes them. I think we're going to burn through this one quickly though. Nano banana isn't good at creating these things either!

1 0 1 136 0

View Details

b-squared @bbopen

11 months ago

Claude code and GPT-5 Codex CLI and whatever’s next are not here to make garden variety slop wrappers. they’re here to get us up to speed on everything from evolutionary generated synthetic data to prod ml ops. ngmi otherwise!