Andrea de Varda@devarda_a
New preprint! 🤖🧠
The cost of thinking is similar between large reasoning models and humans
👉
w/ Ferdinando D'Elia, @AndrewLampinen, and @ev_fedorenko (1/6)
4269415.5K
Andrea de Varda
@devarda_a
Postdoc at MIT BCS, interested in language(s) and thought in humans and LMs
591 Following616 Followers217 Posts
- Andrea de Varda@devarda_aAndrea de VardaRT @thomashikaru: If you've noticed a distinct dialect emerging in frontier AI models and struggled to understand what they're talking about, you're not crazy! We turned to psycholinguistics to investigate what makes the writing style so weird, and why this might be happening 🚨New preprint! 🧵013602
- Andrea de Varda@devarda_aAndrea de VardaRT @AndrewLampinen: New blog post! I wrote about why I think it's useful to use metaphorical (and anthropomorphic) terms to describe AI. I start by describing the role of metaphor in language and cognition, and how metaphors can help us to efficiently understand relations and make predictions 1/2013010.9K
- Andrea de Varda@devarda_aAndrea de VardaRT @mcgovernmit: Congrats to @devarda_a, named one of the 2026 #STATWunderkinds by @statnews! Andrea is doing groundbreaking work in @ev_fedorenko + @roger_p_levy labs to understand how brains and machines process language. Learn more: @mitbrainandcog @ScienceMIT0301
- Andrea de Varda@devarda_aThrilled to be one of @statnews's 2026 #STATWunderkinds for my work on language processing and reasoning in 🧠 and 🤖 Big thanks to my current/previous advisors @ev_fedorenko @roger_p_levy Marco Marelli and everyone I get to work with!
106643.2K - Andrea de Varda@devarda_aAndrea de VardaRT @gushilab: Why do neural networks become modular? Our paper is now out in Nature Machine Intelligence! @NaturePortfolio We show that multi-task learning promotes modularity under limited capacity and task demands help explain features of brain architecture in addition to the spatial constraints. Great thanks to all the co-authors. @WHYWYH007 @yifannnwu @DaniSBassett @marcelomattar05605
- Andrea de Varda@devarda_aCheck out CogGym, led by @LanceYing42! 258 experiments from 100 cogsci papers (!!) in one format. Main takeaway: models are getting more human-like, but way slower than improvements on math and codingLance Ying @LanceYing42Have we built machines that think like humans? 🧠🤖 In a new preprint, we propose CogGym, which compares AI and human judgments across 258 commonsense-reasoning experiments from 100 cognitive-science papers. Paper: Platform:04231.9K
Have we built machines that think like humans? 🧠🤖 In a new preprint, we propose CogGym, which compares AI and human judgments across 258 commonsense-reasoning experiments from 100 cognitive-science papers. Paper: Platform:
63813617.1K- Andrea de Varda@devarda_aAndrea de VardaRT @whylikethis_: 📄👀 Check out our new survey on Eye Tracking and NLP, where we reflect on past work in this space, and outline a roadmap for what's next. Not needless to say - it's fully human written. The humans are Lena Jäger, David Reich, Omer Shubi and me.0203
- Andrea de Varda@devarda_aAndrea de VardaRT @marcel_binz: 📢 The CORE COGNITION lab ( is looking to hire another PhD student, working on automated scientific discovery and knowledge representation in cognitive science. Starting date: March to September 2027. 🙏 Please share widely!04003
- Andrea de Varda@devarda_aNew preprint! 🤖➕➗ Shared circuits predict whether LLMs generalize across formats in arithmetic reasoning w/ @sanapandey, @pengrui_han, @jacobandreas, and @ev_fedorenko If you can solve 2+5, you can solve "two plus five" without learning math again. LLMs are more brittle (1/5)
1221607.5K - Andrea de Varda@devarda_aPosts @devarda_aCircuit overlap predicts generalization at three levels: which formats are hardest (English > Spanish > Italian), which models generalize best, and which individual items get solved (fig below). If the numeric circuit is not engaged, the answer is almost always wrong. (4/5)
106476 - Andrea de Varda@devarda_aPosts @devarda_aThe circuit metric (unsupervised) predicts generalization as well as or better than supervised probes. Behavioral generalization is grounded in circuit generalization: if the circuit does not transfer the ability does not either. Pre-print 🔗 (5/5)
107412 - Andrea de Varda@devarda_aAndrea de VardaRT @ev_fedorenko: Go, @devarda_a! A beautiful and comprehensive study!🔑 finding: behav. measures are ~fully reducible to simple predictors of effort (surprisal, word length+frequency), but for 🧠 measures, LLM embeddings carry additional predictive power, likely capturing aspects of meaning.Andrea de Varda @devarda_aNew preprint! 🧠👀🤖 Behavioral and brain responses to language reflect different levels of linguistic representation w/ @whylikethis_ , @ev_fedorenko , and @roger_p_levy (1/10)0902
- Andrea de Varda@devarda_aNew preprint! 🧠👀🤖 Behavioral and brain responses to language reflect different levels of linguistic representation w/ @whylikethis_ , @ev_fedorenko , and @roger_p_levy (1/10)
316757.4K - Andrea de Varda@devarda_aAndrea de VardaRT @whylikethis_: Wonderful work by @devarda_a linking behavioral and neural manifestations of language comprehension using language models. With @roger_p_levy and @ev_fedorenkoAndrea de Varda @devarda_aNew preprint! 🧠👀🤖 Behavioral and brain responses to language reflect different levels of linguistic representation w/ @whylikethis_ , @ev_fedorenko , and @roger_p_levy (1/10)0202
- Andrea de Varda@devarda_aPosts @devarda_aA similar asymmetry emerges within the effort measures. Eye movements are driven mostly by length and frequency (context-independent). Brain responses are driven mostly by surprisal (context-dependent). (9/10)102243
- Andrea de Varda@devarda_aPosts @devarda_a👀!=🧠 A unified theory needs both kinds of data with a clear understanding of which levels of representation each measure reflects. (10/10) Pre-print 🔗005337
- Andrea de Varda@devarda_aPosts @devarda_a👀 Behavior: effort dominates. SFL (three numbers) performs better than 1600-dim embeddings, esp. in eye-tracking. Adding EMB to SFL gives only small gains. 🧠 Brain: the opposite. Embeddings predict fMRI/N400 responses better than effort. (8/10)
105344 - Andrea de Varda@devarda_aPosts @devarda_aWe test this on 13 datasets: 👀 4 eye-tracking, 3 self-paced reading, 1 Maze, 🧠 1 ERP (N400), and 4 fMRI. All responses are averaged across participants and brain responses are averaged across voxels/electrodes so brain and behavior are comparable (1-dimensional). (7/10)
102228 - Andrea de Varda@devarda_aPosts @devarda_aOur hypothesis: behavior shows a bottleneck. Rich representations get compressed into effort dimensions (SFL) before they can influence processing times. Brain responses have more direct access to the high-dimensional representations. (6/10)
103314 - Andrea de Varda@devarda_aPosts @devarda_aWe use LMs to operationalize effort and meaning in one framework. From the same GPT/GPT-2 models we get (i) surprisal (+freq and len; SFL) for effort, and (ii) contextual embeddings (EMB), high-dimensional vectors that encode form and meaning. (5/10)
103332
