Sílvia Casacuberta

300 posts

Sílvia Casacuberta banner
Sílvia Casacuberta

Sílvia Casacuberta

@SiCaPu

CS PhD student & @KnightHennessy scholar @Stanford. Prev.: @UniofOxford with @rhodes_trust, @Harvard '23, @MPI_IS, @CSatETH, @IBM. Theory CS for Trustworthy AI

Katılım Nisan 2013
2.2K Takip Edilen811 Takipçiler
Sílvia Casacuberta retweetledi
Boaz Barak
Boaz Barak@boazbaraktcs·
New award: Trevisan Award for Expository Work. Love the idea of awarding work that explains and illuminates, and it is so fitting way to honor Luca, who was a master explainer. windowsontheory.org/2026/02/05/tre…
English
1
9
56
3.7K
Sílvia Casacuberta retweetledi
Simons Institute for the Theory of Computing
Simons Institute for the Theory of Computing@SimonsInstitute·
The successes of generative AI and LLMs involve deep internal representations of the world. In his Theoretically Speaking lecture, Jon Kleinberg explored how these work, and how they relate to the representations of the world that we build as humans. bit.ly/4qE98dO
Simons Institute for the Theory of Computing tweet media
English
1
4
24
1.9K
Judy Shen
Judy Shen@judyhshen·
📣Our study on how AI impacts coding skill formation is now out! (w. @AlexTamkin) AI Assistance is NOT a shortcut to skill formation. Using AI to help you learn to code only reduces mastery if you’re delegating everything.
Anthropic@AnthropicAI

AI can make work faster, but a fear is that relying on it may make it harder to learn new skills on the job. We ran an experiment with software engineers to learn more. Coding with AI led to a decrease in mastery—but this depended on how people used it. anthropic.com/research/AI-as…

English
8
6
102
12.3K
Sílvia Casacuberta retweetledi
Quanta Magazine
Quanta Magazine@QuantaMagazine·
To thwart malicious prompts, large language models use filters that are smaller than the models themselves. Recently, the Turing Award-winning cryptographer Shafi Goldwasser and researcher Sarah Ball showed that this size disparity makes LLMs insecure. quantamagazine.org/cryptographers…
Quanta Magazine tweet mediaQuanta Magazine tweet media
English
0
28
165
24K
Sílvia Casacuberta retweetledi
Quanta Magazine
Quanta Magazine@QuantaMagazine·
The researchers Lijie Chen (left), Igor Oliveira (top-right) and Jitau Li recently used “reverse mathematics” to connect a theorem called the pigeonhole principle to a whole class of complexity problems. quantamagazine.org/reverse-mathem…
Quanta Magazine tweet mediaQuanta Magazine tweet mediaQuanta Magazine tweet media
English
13
95
805
66.7K
Sílvia Casacuberta
Sílvia Casacuberta@SiCaPu·
If you are interested in learning with abstentions (in the predictive or generative setting), calibration, uncertainty quantification, conformal prediction, or any other related topics, I would love to chat! I’ll be in Exhibit Hall C,D,E #3013 this Thursday 11 am - 2 pm PST.
English
0
0
3
148
Sílvia Casacuberta
Sílvia Casacuberta@SiCaPu·
Looking forward to attending #NeurIPS in San Diego this week! I will be presenting our spotlight paper (joint with Varun Kanade) on building predictors that abstain optimally and fairly.
Sílvia Casacuberta tweet media
English
1
0
10
222
Sílvia Casacuberta retweetledi
Preetum Nakkiran
Preetum Nakkiran@PreetumNakkiran·
LLMs are notorious for "hallucinating": producing confident-sounding answers that are entirely wrong. But with the right definitions, we can extract a semantic notion of "confidence" from LLMs, and this confidence turns out to be calibrated out-of-the-box in many settings (!)
Preetum Nakkiran tweet media
English
23
84
585
51.2K
Sílvia Casacuberta retweetledi
Sarah Cen
Sarah Cen@cen_sarah·
In the AI ecosystem, who supplies the data? the compute? the models? We just released a new tool on the AI Supply Chain. Our dataset reveals how AI models, data, compute, capital, and even talent change hands. Here’s why you should care 👇
Sarah Cen tweet media
English
15
39
154
67K
Sílvia Casacuberta retweetledi
Nikhil Chandak
Nikhil Chandak@nikhilchandak29·
Thinking of doing a PhD/research visit? Consider applying to/visiting @ELLISInst_Tue @MPI_IS. We have PIs doing cool work, great funding, and most importantly, one of the best academic compute: 50+ GB200s, 250+ H100s, and many more A100 80GBs. Come join us!
Nikhil Chandak tweet mediaNikhil Chandak tweet media
English
16
31
420
37.2K
Sílvia Casacuberta retweetledi
Jelani Nelson
Jelani Nelson@minilek·
See more about the upcoming Blackwell documentary, and donate to help support it, at gameofgeniusfilm.com For those who don't know, Blackwell also happens to be the first black full professor in the entire @UofCalifornia system (1955, UC Berkeley Statistics).
Jelani Nelson@minilek

A pleasure to be invited by Lisa Blackwell, granddaughter of Prof. David Blackwell, to say a few words about her grandfather at an event this afternoon to promote an upcoming documentary about his life. To quote my colleague Mike Jordan, Blackwell's 'approachability' theory is "the basis of the entire online learning paradigm", and "it took many years for people in the online learning community to realize that David was basically the founder of their field". For example, (Abernethy, Bartlett, Hazan '2011) showed that Blackwell's 'Approachability Theorem' is in fact equivalent to regret minimization for Online Linear Optimization; basically, Blackwell was just 50 years ahead of his time.

English
0
4
39
4.5K
Sílvia Casacuberta
Sílvia Casacuberta@SiCaPu·
When we first started thinking about clones for our final project in @ArielProcaccia’s fantastic course on optimized democracy, we couldn’t have anticipated the journey that it would take us on! What a gift to be able to do research with friends
English
0
0
6
361
Sílvia Casacuberta
Sílvia Casacuberta@SiCaPu·
If you are attending #EC2025 this week, today you can see @eminberker present our work on social choice functions that are robust against strategic nomination of candidates!
Vincent Conitzer@conitzer

(2/2) From Independence of Clones to Composition Consistency: A Hierarchy of Barriers to Strategic Nomination W 1:30pm session Led by Emin Berker&relevant to evaluating AI, where it is easy to create very similar variants of a given AI system ("clones"). arxiv.org/abs/2502.16973

English
1
1
18
1.2K

Keşfet

@judyhshen @AlexTamkin @ELLISInst_Tue @MPI_IS @UofCalifornia @walesalaudeen96 @joavanschoren @EurIPSConf