Assist Prof in Intelligent Systems, Trinity College Dublin @tcddublinscss @AdaptCentre @tcddublin @TCINeuroscience
Joined September 2015
- Tweets1.2K
- Following282
- Followers1K
- Likes3K
Giovanni Di Liberto retweeted
Excited to share that our paper “Polar probe linearly decodes semantic structures from LLMs” was accepted at COLM 2026! 🎉
Joint work with the amazing: @PierreOrhan, Emmanuel Chemla, @lakretz and @JeanRemiKing
We find a simple geometry for semantic structures in LLMs
👇🧵1/5
Giovanni Di Liberto retweeted
Cool eval. Simply ask an LLM “Land or Water?” and give it a latitude and longitude coordinate as text. Ask 16,200 times, plot as image. The models know. From compressing the internet.
Giovanni Di Liberto retweeted
#backpropagation - for poeple suffering from derivative allergy (hopefully useful for teaching @jeremyphoward)
Giovanni Di Liberto retweeted
I would like to highlight two postdoctoral positions in the project Neuronal Mechanisms of Reading in Children Using OPM-MEG at the University of Oxford:
my.corehr.com/pls/uoxrecruit…
my.corehr.com/pls/uoxrecruit…
Deadline 30 Sep.
Giovanni Di Liberto retweeted
Congratulations @bentang0 and @zspald11 on your fantastic preprint!
An exciting start for the lab into unsupervised learning and foundation models for speech and language decoding.
@Duke_Neurology @DukeBrain @DukeEngineering
Giovanni Di Liberto retweeted
This is very cool. Our yet-to-be-published fMRI data also suggests a tight anatomical connection between vocal and limb control.
The revolutionary brain-computer interfaces that help people with severe paralysis speak, type, move prosthetic limbs, and more are typically specialists, using sensors to pick up activity from select brain areas to perform a single function.
Now, scientists have shown just one neural implant—in this case, a grid of minielectrodes surgically placed on the top part of the cortex—can do double duty, successfully “decoding” brain signals from paralyzed people as they attempt to speak as well as gesture with their upper bodies.
The technology can create a computer-animated image of the user in real time, which can do things such as say hello while waving, or nod while saying yes.
Learn more: scim.ag/4Ac98XJ
ALT Single brain implant lets paralyzed people both talk and gesture via an avatar
Giovanni Di Liberto retweeted
New 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)
Giovanni Di Liberto retweeted
🧠 What music reveals about the human brain ➡️ bit.ly/4zDT00W
This article was also published by the 🇭🇷 Croatian newspaper @24sata_HR : bit.ly/4zpMIl8
#ResearchImpactEU #Science4EU
Giovanni Di Liberto retweeted
New preprint, w/ @KarolinaADrozdz
Understanding language requires internally modelling what's happening where, even when unstated
We built a new task to measure this ability in language models and people, and find it emerges at model scales far smaller than previously thought
Giovanni Di Liberto retweeted
How does a speech plan become fluent movement?
In Nature Human Behaviour, led by @dsuseendar:
Intracranial recordings reveal how hierarchical speech plans become continuous motor sequences across planning, articulation, and monitoring networks.
doi.org/10.1038/s41562-026-0…
1/5
Giovanni Di Liberto retweeted
We’ve spent years showing that people hearing the same story share neural responses in high-level cortex. Now: instrumental music—no words—does too, converging on the story listeners imagine.
New in @NatureComms, led by @ItamarJalon with @LisaMargulis: nature.com/articles/s41467-0…
That's a clever paradigm and definitely a study that I'll cite many times!
New lab preprint!!
Attention or prediction? Characterizing the top-down influence of predictive context on speech encoding
biorxiv.org/content/10.64898…
Brief thread:
Giovanni Di Liberto retweeted
New perspective piece out in @NeuroCellPress with @HassonLab, @zaidzada_, and Adele Goldberg! We try to articulate some of our excitement about LLMs and discuss what kinds of insights they might provide into the neural computations supporting natural language in the human brain.
Giovanni Di Liberto retweeted
Word meanings are constructed dynamically by integrating relatively stable, stored lexical-semantic knowledge with context-specific semantic predictions and idiosyncratic semantic associations that are retrieved from episodic memory
cell.com/action/showPdf?pii=…
Giovanni Di Liberto retweeted
I'm thrilled to be involved in this new study with Randy Buckner's group at Harvard. Clearly demonstrates two hierarchical levels in the language network using precision fMRI. More evidence that the neural architecture of language is hierarchical and structured.
biorxiv.org/content/10.64898…
Giovanni Di Liberto retweeted
Now out in @NatureComms w/ @ev_fedorenko: across many people (>1k) and tasks (>300), the the brain's language network reliably shows up in the functional connectome (correlated activity in fMRI).
1/
ALT Screen capture of abstract of published article. Title: A language network in the individualized functional connectomes of 1199 human brains doing arbitrary tasks Abstract: A century and a half of neuroscience has yielded many divergent theories of the neurobiology of language. Two factors that likely contribute to this situation include (a) conceptual disagreement about language and its component processes, and (b) intrinsic inter-individual variability in the topography of language areas. Recent functional magnetic resonance imaging (fMRI) studies of small numbers of intensively scanned individuals have argued that a language-selective brain network emerges bottom-up from correlations (individualized functional connectomics, iFC) in task-free (e.g., rest) or task-regressed activation timecourses. Here we tested this hypothesis at scale and evaluated its practical utility for task-agnostic language localization: we apply iFC separately to each of 1,957 (fMRI) scanning sessions (1...
Giovanni Di Liberto retweeted
New lab preprint!!
Is that clear? Robust electrophysiological measures of the effects of prior knowledge on degraded speech perception.
biorxiv.org/content/10.64898…
A short thread:
Giovanni Di Liberto retweeted
Four missing keystones in the evolutionary origins of music
Forum by Roberta Bianco (@RobertaBianco0), Robert Zatorre, & Giacomo Novembre (@NovembreGiacomo)
tinyurl.com/mr35733r
Giovanni Di Liberto retweeted
Dimensionality reduction can see structures that do not exist and miss structures that exist.
𝐓𝐡𝐞 𝐬𝐢𝐦𝐩𝐥𝐞𝐬𝐭 𝐞𝐱𝐩𝐥𝐚𝐧𝐚𝐭𝐢𝐨𝐧 𝐢𝐬𝐧’𝐭 𝐚𝐥𝐰𝐚𝐲𝐬 𝐭𝐡𝐞 𝐛𝐞𝐬𝐭 𝐨𝐧𝐞.
Sometimes, no simple way of describing the complexity of the data exists, at least not one that humans can readily interpret.
Then, faithfully representing the data is at odds with finding simple explanations.
“Everything should be made as simple as possible, but no simpler”
A basic intro video with some strategies how to avoid mistakes when using dimensionality reduction:
youtu.be/nIB-WjWzDSc?si=vfKT…
