Dan Ley

researcher at @guidelabsai

[Dan Ley @dan_w_ley]

@Jaeyeon_Kim_0 Great work !

[Guide Labs @guidelabsai]

It’s official: the first large-scale inherently interpretable language model is here.

Steerling-8B from @guidelabsai is the first and largest model that can trace every token it generates back to:
→  Input Context
→ Training data
→ Human-understandable concepts

In other words, we've successfully trained Steerling-8B to trace its outputs and explain what has impacted that decision for more reliable manipulation. This isn’t post-hoc explainability. Interpretability is built directly into the model.

🔓Steerling-8B can self-monitor for memorized content and suppress it at inference time without retraining. That makes interpretability a first-class design principle, not an afterthought.

This is a major step toward models we can actually understand, debug, and trust.

Over the coming days, we’ll be sharing investigations into what Steerling-8B’s interpretability enables in practice. Stay tuned as we dive deeper into our research & how we are building LLMs we can trust.

🚨 Try it LIVE and help improve it:

Guide Labs: https://t.co/EyEMFz2p9O 

GitHub: https://t.co/PIVwJgleFP

Hugging Face: https://t.co/0apB117l4o

Huge thank you to @TimFernholz and @TechCrunch for featuring this breakthrough. https://t.co/DIZpq5XqGS

#Steerling8B #GuideLabs #AI #MachineLearning

[Dan Ley @dan_w_ley]

We taught LLMs to learn “typical examples” from their training data, and to reveal which ones they use at each timestep, in real time, without slowing the LLM down or hurting performance. Thanks to Guide Labs for hosting! 🚀

[Guide Labs @guidelabsai]

We trained PRISM, a family of interpretable language models that trace their predictions to training data in a single forward pass.

When a language model predicts the next token, which training samples is it relying on? PRISM answers this by design.

[Guide Labs @guidelabsai]

🚀 We've developed a language model that's interpretable by design. 8 billion parameters, no performance sacrifice.

[Guide Labs @guidelabsai]

Excited to announce our $9M USD seed round led by @Initialized and the first large-scale interpretable LLM: an 8 billion parameter model capable of explaining its outputs through mechanisms humans can actually understand.

[Dan Ley @dan_w_ley]

@miniapeur Funny enough- we share the same birthday

[Dan Ley @dan_w_ley]

don’t sleep on haize

[Haize Labs @haizelabs]

proud to have played a part in red-teaming the o1 series pre-launch to ensure their robustness and reliability

super exciting to see @OpenAI keep on releasing great work!

[Haize Labs @haizelabs]

proud to have played a part in red-teaming the o1 series pre-launch to ensure their robustness and reliability

super exciting to see @OpenAI keep on releasing great work!

[Dan Ley @dan_w_ley]

@TaylorBond5 For now

[Dan Ley @dan_w_ley]

@TaylorBond5 @brfootball @FabrizioRomano Get him in

[Dan Ley @dan_w_ley]

@OpenAI put an option in ur chatbot to remove the latest response if it sucks or doesn't understand you

worth doing so users dont have to start over or fear bad replies muddling future queries

[Dan Ley @dan_w_ley]

@ThePSF if a module doesn't have a method but one of its submodules does consider pointing it out in the AttributeError ... common situation and we can avoid spending time googling the docs

[Dan Ley @dan_w_ley]

@NoContextBrits Harry Kane inspired by Mesut Ozil

[Dan Ley @dan_w_ley]

@leonardtang_ Nothing wrong with this

[Dan Ley @dan_w_ley]

@uddupa true... there's caring about the thing and there's caring about all the stuff distracting you from caring about the thing

[Dan Ley @dan_w_ley]

default towards not caring

[Dan Ley @dan_w_ley]

I can now watch LLMs get finetuned live on an 85 inch tv

[Dan Ley @dan_w_ley]

@pa_invisible @CantonasMind exactly... he's already half way to the floor before even touching AWB

[Dan Ley @dan_w_ley]

@GaryMarcus Well duh

[Dan Ley @dan_w_ley]

kung fu panda 4 might be the greatest film of our generation