Tim Vieira

machine learning, reinforcement learning, programming languages, handstands (he/him)

[Tim Vieira @xtimv]

When you say, "This is a reinforcement learning problem," you should say it with the same excitement as "This is NP-hard."

[Tim Vieira @xtimv]

@krismicinski @StarGazerMiao Please do!

[David Chiang @davidweichiang]

@ReviewAcl Why does every author have to fill out the registration form if they already have an OpenReview profile?

[Rico Angell @rico_angell]

It’s deployment time!

You’ve done the pre-deployment evals. You THINK your model is safe, so you ship it 🚀

🚨 After deployment, reports of misbehavior start trickling in

What happened?? How could you have caught it?? 🤔

@icmlconf 2026 Spotlight!

🧵

[Mathieu Blondel @mblondel_ml]

Our paper "Autoregressive Language Models are Secretly Energy-Based Models: Insights into the Lookahead Capabilities of Next-Token Prediction" was accepted for publication at #ICML2026 https://t.co/DUTSj4Sz7u

[Yahya Emara @yemara22]

We made LLM inference a lot faster.

1/n

Speculative decoding throws away draft tokens when they don't match. What if you re-weighted your samples instead using importance sampling?

We introduce SMC-SD, sequential Monte Carlo speculative decoding, an approximate sampler that speeds up inference by 2.36× over SOTA speculative decoding, while producing a controllable Pareto frontier of speed and approximation accuracy.

ArXiv: https://t.co/XWY0jKvbyj

minimal repo release: https://t.co/lI53Qu16Mo

Joint work with Mauricio Barba da Costa (@mauricio_b25181), Chi-Chih Chang (@CCCCC1009CCC ), Cameron Freer, Tim Vieira (@xtimv), Ryan Cotterell, and Mohamed Abdelfattah (@mohsaied).

[Tim Vieira @xtimv]

Claude Code is really bad at f-strings. I can't tell you how many times a day I see some version of this error message a day. Please fix!

[Tim Vieira @xtimv]

Some more vibe-coding fun - every math major's favorite party trick: the wobbly table theorem as an interactive 3D visualization.
https://t.co/P30394hpFe

[Tim Vieira @xtimv]

@elmelis That's a fantastic title. I will list it in my CV under awards next to "Lord of the Semirings"

[Tim Vieira @xtimv]

Ok, AI-assisted coding is pretty rad. Here's something I built to explore the stuff I've written (blog posts and publications). It's still a bit of work in progress, but it's pretty fun, especially the "semantic" tab + sliders.

I'd love some feedback!
https://t.co/8pAIRCaeWP

[Tim Vieira @xtimv]

@elmelis Yeah... It's something I'm known for I guess...

[Tim Vieira @xtimv]

@PromptSlinger @nomic_ai Source code: https://t.co/Lvku612xf3

The whole thing is built from a centralized YAML file. So it should be pretty modular in case you want to try it on different content. I have been meaning to try out for exploring the 15 years of PDFs that have accumulated on my hard drive

[Tim Vieira @xtimv]

@PromptSlinger I generate the embeddings locally from PDF or markdown sources using an embedded model @nomic_ai, which supports large docs

[Tim Vieira @xtimv]

@srush_nlp @mnoukhov Oh interesting, I'll have to check that out. Any chance the inference setup will ever be open-sourced?

[Tim Vieira @xtimv]

@srush_nlp I am fine to be in the "didn't provide gains" camp. The implementation is simple, we give code, it's short, and it's online. So that comment seems misplaced.

[Tim Vieira @xtimv]

@srush_nlp @mnoukhov (LOL - did not mean to imply that all papers are careful or that all blog posts are not careful. I do have my issues with this specific blog post. And I have issues when Claude Code's WebFetch commands land 99% of the time on blog posts rather than books and research papers.)

[Tim Vieira @xtimv]

@mnoukhov @srush_nlp I liked the blog too... until it cost us a few months of time 🤷‍♂️

[Tim Vieira @xtimv]

@mnoukhov @srush_nlp k3 with the calibration parameter works fine, but without it's getting in a roller coaster without a harness.

On the other hand, RB just completely kicks ass & has the nonnegativity property that Schulman wanted.

In your settings is there something that keeps k3 from exploding?

[Tim Vieira @xtimv]

@mnoukhov @srush_nlp And, of course, 🤗 gives it to you right away.

[Tim Vieira @xtimv]

@mnoukhov @srush_nlp It's easy enough to patch vLLM to give it to you, but I haven't figured out how to do it for SGLang yet.