> replicate J-space on GLM 5.2 > train a reward model and run RL to reduce hallucinations > show me how this model makes cancer predictions Using our platform Silico is like having a team of AI researchers ready to run experiments like these. Private beta is open now.... 🧵 (1/6)show more

Silico replicated J-space on GLM-5.2 overnight. It then extended context to ~256k tokens, replicating the key results on multi-hop question answering. (2/6)

Our team spent months developing RLFR, our method which uses probes on a model's internals as reward signals for RL. Silico reproduced it in 2 days, reducing hallucinations in Qwen3-8B by 37% without capability loss. (3/6)

Silico lets us look inside models to see what they’ve learned. Using BSFs on protein language models, it found - without supervision - subspaces in the model whose activations correlate with known protein structures. (4/6)

Here, Silico replicated PICASSO on Midnight-12k in one-shot. PICASSO interprets digital pathology models. It breaks what the model sees into readable concepts, shows which drive its cancer predictions, and simulates how changes to the tissue would alter those predictions. (5/6)

These are just a few examples of what you can do with Silico. Request access to the private beta here:

I was considering applying J-space to outer/inner loops

very cool work

This makes complex AI experimentation far more accessible for research teams.

Seems to be what happens if you prompt Claude code to do stuff, but by making extra point cloud gifs in the process.

Potentially life-saving. Thank you Goodfire.

@threadreaderapp please #unroll

Turning research replication into a simple prompt is a huge time saver for experiment heavy work.

@goodfire - I have a hypothesis I want to try. Would silico be useful to test it?

very cool!

@threadreaderapp unroll ListenUp

So much info here

Goblins?

When it will be available via subscription?

Interesting

the honest answer is that anyone can throw models at data but the ones who actually solve real-world problems are those who understand how to shape the problem space, not just optimize the model.

Will it be possible to run locally or via cloud + subscription only? > Some experiments are too sensible to run on public infrastructure.

Ummm, I volunteer as a tribute to test this out🙋‍♂️

very cool

@angshubh

saw hallucination rl runs look good until eval prompts matched real users