Sam Rose

I'm now able To push those spirits outside My thought is free And forever lives in me Psychic transfer I've stolen their unique power Nothing can stop the evacuation Nothing can stop the psychic vacuum

Training AI on ocaml could be like generating code and verify against compiler and property based tests, which is a higher quality corpus than scraping and weighting information.

Agents can use compiler as feedback loop, and tools can be configured to make it very strict.

The cost of dev in agent oriented work moves from writing to verifying, which is much better in pure functional languages.

I feel like fp had a resurgence starting around 2021. Then, once ai agents proliferated, unless people already had his preference, then it just got shoved aside somewhat in the industry. Ironically, fp and lang like ocaml make agent based work better

Was this already solved by splitting up services into their own repos (polyrepo)?

If someone says this shit, ask for receipts on how their Agents/models are running flawlessly, and how they achieved it.

to run agents that connect to small local models I should have said. thanks for the response

@LilithSemi I am considering using git.lilithsemi.com/LilithSem… to run small local models. Any advice on it?

I am a lot older than that

My teenage son and his friends are obsessed with buying and using and fixing old tech like this.

yes but yes. Once the AI startups figure this out it'll be the next "big thing".

But will these be the people AI targets first once skynet goes live?

It seemed with some claude exploration that a total system could be created with elixir/phoenix, erldns, caddy orchestrated by elixir and s3. Maybe could be a good open source project idk

fail2ban

I wonder if anything at all is truly "deterministic" I'll guess probably not.

yes will do!

@fmzakari this project is fucking amazing github.com/fzakaria/rewindvm fzakaria.com/2026/10/03/rewi…