mirekrusin

Agree trials won't compress much with AI in the near future. But they're starting with basic discovery rather than therapeutics – that part can move fast.

I'd also judge it less by what result is and more by the rate of change – even a year ago ~1k agents running ~1d on single prompt producing wet-lab-verifiable leads wasn't really a thing.

must look cool for people who want to solve every problem with k8s

it starts with interesting misnomers like "Task" which is not a work item but a sandbox.

"billions of tasks" is a "solution" to problem nobody has (maybe some RL labs? but they solve it other way and with orders of magnitude better optimizations).

freezes design too early – unless they'll actually focus on developing it and make tons of breaking changes it looks shit.

shared state in the same workspace, identity, authority, etc – stuff like that needs to be solved

Palantir like system – data, ontology, logic, actions, security, operational workflows – packaged as single cli, local first, git friendly text .cave files and sqlite3 instead of k8s-style complexity blowup.

Working on the weekends.

Next is probably extension of the language to integrate something like https://github.com/mirek/ast for arbitrary ast traversals/modification/generation/better integration with external world - other file types, but ast concept goes further than that, ie. filesystem is also ast, traversal is ast/xpath like walk.

Public domain on all my recent work.

[0] https://github.com/mirek/cave

Persistent weight adaptation also happens just not in real time - sessions are captured, analyzed, transformed into training data, fed into SFT/RL environments and later contribute to model updates. Takes a bit of time for the whole loop but you can't say it's not present.

There's nothing fundamentally preventing real-time weight updates, ie. LoRA-style online adaptation would be one obvious approach. It's just generally not worth doing at scale. Updating a shared model centrally gives much better data efficiency, batching, evaluation, control etc. than continuously training a separate set of weights for every user/session.

There is some work happening on narrowing that gap, for example Mistral has been pushing efficient LoRA-based customization, continuous pretraining, model adaptation etc.

I also did play a bit with activation steering – it's super cool where you extract profile for some concepts (emotional in my case) and you have effectively toggles to control "brightness/contrast" those areas (enhancing or suppressing those activation regions from profile) injecting to the model those concepts (emotions in my case) – you can do it in real time and it's fun thing to play with.