Alex Lew

This work will be presented at CogSci next week with an awesome talk by Lio Wong! (do please stop by the talk to chat if at @cogsci_soc !)

Very exciting and (IMO) compelling account of the open-endedness of human cognition, simultaneously explaining how: (i) we're locally quite rational (probabilistically coherent, etc.) (ii) we can nonetheless "solve" the frame problem and reason about a v. wide range of situations

How do people reason while still staying coherent – as if they have an internal ‘world model’ for situations they’ve never encountered? A new paper on open-world cognition (preview at the world models workshop at #ICML2025!)

Gabe, this is not at all what I had in mind. This is probably way better :D I think it's probably the most exciting inference-time strategy I've seen in a (very) long time. (I suspect it's probably too restrictiveto defer control entirely to a script, but this can be adjusted.)

Want to use AWRS SMC? Check out the GenLM control library: github.com/genlm/genlm-co… GenLM supports not only grammars, but arbitrary programmable constraints from type systems to simulators. If you can write a Python function, you can control your language model!

Many LM applications may be formulated as targeting some (Boolean) constraint. Generate a… - Python program that passes a test suite - PDDL plan that satisfies a goal - CoT trajectory that yields a positive reward The list goes on… How can we efficiently satisfy these? 🧵👇

Excited to present this work at VerifAI @iclr_conf today! Stop by Garnet 218-219 at 3:15pm for the talk 🙏

I myself used an early version of GenLM control to translate from natural language into complicated formal expressions for Bayesian theory-of-mind reasoning -- and it's so nice to be able to just specify a grammar and get syntactically valid samples. x.com/LanceYing42/st…

Very excited for this to be shared more widely as one of the first users!! Definitely check out the talk if you're at ICLR. One of many awesome follow-up papers that build on @alexanderklew's work on Sequential Monte Carlo steering of LLMs :)

#ICLR2025 Oral How can we control LMs using diverse signals such as static analyses, test cases, and simulations? In our paper “Syntactic and Semantic Control of Large Language Models via Sequential Monte Carlo” we: Cast controlled generation as an inference problem, with the LM…

Tackling complex problems with LMs requires search/planning, but how should test-time compute be structured? Introducing Self-Steering, a new meta-reasoning framework where LMs coordinate their own inference procedures by writing code!

the #TPM ⚡Tractable Probabilistic Modeling ⚡Workshop is back at @UncertaintyInAI #UAI2025! Submit your works on: - fast and #reliable inference - #circuits and #tensor #networks - normalizing #flows - scaling #NeSy #AI 🕓 deadline: 23/05/25 👉 …able-probabilistic-modeling.github.io/tpm2025/

New preprint on controlled generation from LMs! I'll be presenting at NENLP tomorrow 12:50-2:00pm Longer thread coming soon :)

New preprint is live! Tweet thread coming 🚧🔜 📅 Excited to present this work in-person: - 4/11: Poster at New England NLP (NENLP) 2025, Yale University (tomorrow!) - 4/27: Oral talk at VerifAI@ICLR 2025, Singapore