@ItsMrLin

Research Scientist @Meta Applied AI, ex-CAS/CDS. Human-in-the-loop optimization, Bayesian methods, and sample-efficient learning from feedback. Stanford CS PhD.

 Views my own.
ItsMrLin.com
Joined January 2013
  • Tweets287
  • Following903
  • Followers595
21Media
“I like this direction, but it’s optimizing the wrong notion of good.” #ICML2026: we are presenting LILO: Bayesian optimization w/ language feedback on Thur (7/9), 10:30–12:15 Hall A #3612 icml.cc/virtual/2026/poster/… Grounded route to language-guided agent improvement. Let's chat.
Katarzyna (Kasia) Kobalczyk@kasia_kobalczykJun 22
Preference optimization often reduces feedback to comparisons or rankings But decision makers often know more than “A >> B” They can explicitly state their preferences and priors in words LILO is a new framework turning such feedback into signal for Bayesian Optimization 🧵
4
320
Excited to present LILO at #ICML2026. One lens: a principled auto-research-style harness built on Bayesian optimization. LLMs surface deep prior knowledge and capture language feedback; BO turns it into uncertainty-aware search/exploration. I’ll be at the poster. Let’s chat!
Katarzyna (Kasia) Kobalczyk@kasia_kobalczykJun 22
Preference optimization often reduces feedback to comparisons or rankings But decision makers often know more than “A >> B” They can explicitly state their preferences and priors in words LILO is a new framework turning such feedback into signal for Bayesian Optimization 🧵
1
5
570
Paper here: arxiv.org/abs/2510.17671 More details on how LILO turns language feedback and prior knowledge into BO-driven, uncertainty-aware optimization.

LILO: Bayesian Optimization with Natural Language Feedback

Many real-world optimization problems are guided by complex, subjective preferences that are difficult to express as explicit closed-form objectives. In response, we introduce Language-in-the-Loop...

arxiv.org
131
Zhiyuan Jerry Lin retweeted
Max Balandat@MaxBalandatJun 22
Check out our work on LILO led by @kasia_kobalczyk and @ItsMrLin. LILO allows users to perform black box optimization with natural language feedback. Hope to see you at the poster at #ICML2026 Paper: arxiv.org/pdf/2510.17671
Katarzyna (Kasia) Kobalczyk@kasia_kobalczykJun 22
Preference optimization often reduces feedback to comparisons or rankings But decision makers often know more than “A >> B” They can explicitly state their preferences and priors in words LILO is a new framework turning such feedback into signal for Bayesian Optimization 🧵
2
9
688
Zhiyuan Jerry Lin retweeted
Chirag Nagpal
@nagpalchirag12 Feb 2024
My kindest sympathies for the Bayesian Optimization folks
Jascha Sohl-Dickstein
@jaschasd12 Feb 2024
Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.
11
40
1
591
122,484
Zhiyuan Jerry Lin retweeted
Max Balandat@MaxBalandat21 Nov 2023
2024 PhD Internship opportunity! Join me and my amazing colleagues on Meta's Adaptive Experimentation team to work on Bayesian optimization, probabilistic modeling / Gaussian Processes, and sample-efficient decision making: metacareers.com/jobs/9056341…
1
28
2
125
34,666
Spotted @infoxiao at @CODEConference this year!
2
1
11
2,537
Glad to see this conference has gotten noticeably bigger and better with more folks from even broader backgrounds 😃
2
272
Zhiyuan Jerry Lin retweeted
Yann LeCun
@ylecun24 Aug 2023
You knew that was coming: Code LLama !!! - Llama-2 tuned for code generation, debugging, etc - base models, Python-specific, and instruction-tuned. - 7B, 13B and 33B params models - Available with the same license as Llama-2 - Blog: ai.meta.com/blog/code-llama-… - Paper: ai.meta.com/research/publica… - Code: github.com/facebookresearch/… - Models: ai.meta.com/resources/models…

Introducing Code Llama, a state-of-the-art large language model for coding

Code Llama, which is built on top of Llama 2, is free for research and commercial use.

ai.meta.com
Gabriel Synnaeve
@syhw24 Aug 2023
Replying to @syhw
We're releasing base models, Python-specialized models, and Instruct(ion following) models, all in sizes 7B, 13B, 34B params. Get the code and weights: github.com/facebookresearch/… Read the research paper: ai.meta.com/research/publica… Read the blog post: ai.meta.com/blog/code-llama-…
39
342
35
1,684
341,339
Finally I can start preying on predatory journals.
andrew gao
@itsandrewgao10 Aug 2023
Go to Google Scholar and look up ‘As an AI language model” -“ChatGPT”’
2
374
It just occurred to me that it's been a while since I used a non-dark-themed IDE. Is this the industry standard now? Miss my old Eclipse days (not really).
2
384
#NeurIPS2023 reviews are out. Once again our proposed methods/algorithms were criticized for being "too simple" and thereof lacks "novelty". I get it that novelty is an important aspect of scientific inquiries but sometimes such comments feels more like lazy reviews.
4
3
4
101
28,482
Isn't simplicity a merit instead of a drawback if there are theoretical and/or empirical evidence backing its validity? What can be done differently in future papers to avoid such criticism? Or perhaps such simple ideas are meant to be kept in a secret treasure chest 🫤
1
23
1,742
Wah
Soumith Chintala
@soumithchintala30 Jul 2023
No More GIL! the Python team has officially accepted the proposal. Congrats @colesbury on his multi-year brilliant effort to remove the GIL, and a heartfelt thanks to the Python Steering Council and Core team for a thoughtful plan to make this a reality. discuss.python.org/t/a-steer…

A Steering Council notice about PEP 703 (Making the Global Interpreter Lock Optional in CPython)

Posting for the whole Steering Council, on the subject of @colesbury’s PEP 703 (Making the Global Interpreter Lock Optional in CPython). Thank you, everyone, for responding to the poll on the no-GIL...

discuss.python.org
317
Zhiyuan Jerry Lin retweeted
Sindhu Ernala@SindhuErnala10 Jul 2023
📢 Latest @Nature @HSScomms publication from our team on the heterogeneous effects of social support on the adoption of FB vaccine profile frames.
Humanities & Social Sciences Communications@HSScomms7 Jul 2023
Published today: The heterogeneous effects of social support on the adoption of Facebook’s vaccine profile frames feature dlvr.it/SrrcSd

The heterogeneous effects of social support on the adoption of Facebook’s vaccine profile frames...

Humanities and Social Sciences Communications - The heterogeneous effects of social support on the adoption of Facebook’s vaccine profile frames feature

nature.com
1
2
19
2,204
**Applauding this commitment to the open source community** 👏 @Meta
Łukasz Langa@llanga8 Jul 2023
Wow. @Meta commits to dedicate three engineer-years to implement the removal of the GIL from #Python and fix upcoming compatibility and performance issues with it. All this dependent on whether the Steering Council accepts PEP 703. discuss.python.org/t/a-fast-…

A fast, free threading Python

We’ve had a chance to discuss this internally with the right people. Our team believes in the value that nogil will provide, and we are committed to working collaboratively to improve Python for...

discuss.python.org
2
351
After a decade since Zuck’s first tweet. Guess all the FOMO kids are on #Threads now, including yours truly @ItsMrLin. Wonder how the whole academic tweets community is gonna be like in a few years…
Mark Zuckerberg
@finkd6 Jul 2023
188
Zhiyuan Jerry Lin retweeted
worms cited@christapeterso2 Apr 2023
Strongest possible advice to early grad students: you need outside hobbies or projects that are meaningful to you. It is dangerous for your only source of value, identity, and friendship to be academia. You absolutely have to find something else to invest in, it’s just not safe
29
401
130
2,652
406,591
PSA: clean your latex source before uploading to ArXiv. There are handy tools such as this one from @GoogleAI github.com/google-research/a…

GitHub - google-research/arxiv-latex-cleaner: arXiv LaTeX Cleaner: Easily clean the LaTeX code of...

arXiv LaTeX Cleaner: Easily clean the LaTeX code of your paper to submit to arXiv - google-research/arxiv-latex-cleaner

github.com
DV@DV255910696507623 Mar 2023
You might know that MSFT has released a 154-page paper (arxiv.org/abs/2303.12712) on #OpenAI #GPT4 , but do you know they also commented out many parts from the original version? 🧵: A thread of hidden information from their latex source code [1/n]
1
447
The end game?
OpenAI
@OpenAI14 Mar 2023
Announcing GPT-4, a large multimodal model, with our best-ever results on capabilities and alignment: openai.com/product/gpt-4
180
Load more