Associate Professor @hseas. Information theorist, but only asymptotically. Brasileiro/American.
Cambridge, MA
Joined July 2009
- Tweets59
- Following95
- Followers374
- Likes77
Flavio Calmon retweeted
๐ฃ Excited to announce our oral presentation at #ICLR!
LLMs capture rich semantic structure, as evidenced by their strong performance across a wide range of language and reasoning tasks.
But Sparse Autoencoders (SAEs), a popular interpretability tool, mostly learn local, noisy, token-level features when applied to LLMs (e.g., hundreds of features for the word โtheโ).
So why arenโt SAEs finding that rich semantic structure?
๐ Because they ignore the sequential nature of language.
We introduce Temporal SAEs to bridge this gap.
arxiv.org/abs/2511.05541
๐งต [1/N]
Flavio Calmon retweeted
Can GenAI agents๐ค manage a supply chain? Lessons from the classical beer game๐บ (1/n) @davidsimchilevi @FlavioCalmon @AndreCalmon
Flavio Calmon retweeted
โผ๏ธ๐New paper alert with @ushabhalla_: Leveraging the Sequential Nature of Language for Interpretability (openreview.net/pdf?id=hgPf1kโฆ)! 1/n
Flavio Calmon retweeted
How can we improve LLMs without any additional training? ๐ค
The standard playbook is using Best-of-N: generate N responses โก๏ธ use a reward model to score them โก๏ธ pick the best ๐
More responses = better results... right?
Well, not exactly. You might be reward hacking!
Instead, you should hedge! ๐ฏ
Flavio Calmon retweeted
Can we use coding-theory, heavy-tailed distributions, and optimal-transport to create ๐๐ฒ๐ฟ๐ผ-๐ฑ๐ถ๐๐๐ผ๐ฟ๐๐ถ๐ผ๐ป, ๐ฒ๐ฎ๐๐ ๐๐ผ ๐๐๐ฒ, ๐๐ฎ๐๐ฒ๐ฟ๐บ๐ฎ๐ฟ๐ธ๐ ๐ณ๐ผ๐ฟ ๐๐๐ ๐? We show they can โ and the result is pretty exciting! ๐ ๐งต (1/n)
Flavio Calmon retweeted
Without its international students, Harvard is not Harvard. hrvd.me/IntStudents25t
Flavio Calmon retweeted
Happy to share we received best paper at NENLP workshop at Yale ๐ฅณ๐ฅณ!
tldr: Current alignment methods give excessive discretion to annotators in defining what good behavior means. This means we don't know what we are aligning to โผ๏ธ
We formalize discretion in alignment and propose mechanisms for data curators & model developers to monitor for it.
Paper link below โฌ๏ธ
Flavio Calmon retweeted
9/n Full paper here: ๐ arxiv.org/abs/2502.10441.
Huge thanks to my amazing team of co-authors:
@maartenbuyl, @hskhalaf, @claudiomverdun, @caiocvm, and @FlavioCalmon.
Work done @hseas!
Flavio Calmon retweeted
AI is built to โbe helpfulโ or โavoid harmโ, but which principles should it prioritize and when?
We call this alignment discretion. As Asimov's stories show: balancing principles for AI behavior is tricky.
In fact, we find that AI has its own set of priorities
(comic @xkcd)๐
Flavio Calmon retweeted
The standard practice in differential privacy of targeting ฮต at small ฮด is extremely lossy for interpreting the level of privacy protection. In practice (e.g., for DP-SGD), we can do much better!
We show how in the #NeurIPS2024 paper:
arxiv.org/abs/2407.02191
Short summary๐
Flavio Calmon retweeted
Imagine an all-powerful AI with any ideology you don't agree with! Super proud of this work, where we show that every LLM reflects a different ideological worldview, which should worry everyone.
Flavio Calmon retweeted
Finally, I am pleased to announce
๐ชขInterpreting CLIP with Sparse Linear Concept Embeddings (SpLiCE)๐ชข
Joint work with Usha Bhalla, as well as @Suuraj, @FlavioCalmon, and @hima_lakkaraju, which was just accepted to NeurIPS 2024! Check out the paper here:
arxiv.org/abs/2402.10376
Flavio Calmon retweeted
Part 2 of my 2024 publication tweets! Please welcome Multi-group Proportional Representation, a novel metric for measuring representation in image generation and retrieval. This work was recently accepted at @NeurIPSConf 2024. (1/n)
Flavio Calmon retweeted
First up, how do various aspects of trustworthy machine learning interact? Can we expect a production ML system to satisfy all regulatory requirements of fairness, privacy, and interpretability simultaneously when past research generally focuses on one component at a time? (1/n)
Mario Diaz Torres, a brilliant researcher and mathematician, passed away suddenly on August 31st. @MDMarioDiaz was a rising star in the LatAm math community and was doing exceptional work in information theory, differential privacy, and related areas. bit.ly/4d9XGPB
Mario was incredibly passionate about math, information theory, and statistics. He was homeschooling his son so he could โteach him math in a principled and advanced manner.โ Now his family really needs our support. Please consider donating here: gofund.me/07df91a5
Mario was a friend, close collaborator, and the first post-doc I hired at Harvard. This is a devastating loss to our community. Please consider reading one of Mario's papers this week. You can also learn more about his research here: bit.ly/3XCNxW2
This week, I spoke on the panel โAI, Rights, and Democracyโ at the Brazilian Supreme Court. Thank you @STF_oficial for the invitation. It was an incredible experience! See my talk (in pt-br) here: youtu.be/mNkZ_Aw2tFs&t=4002s
Flavio Calmon retweeted
Back home from FAccT - I thankful for the work our community is doing & the values it stands for. Serving it has been a labor of love for me & I am beyond grateful to have done so this year along my truly wonderful program co-chairs & human beings @mikarv @RDBinns @FlavioCalmon