Andrea Boscutti

[Andrew Gordon Wilson @andrewgwils]

Seeing a lot recently about whether info theory explains modern AI. This is *exactly* what our epiplexity paper is about. It shows how to resolve paradoxes around DPI, synthetic data, and emergence, by considering computation and structural info: https://t.co/IYNGcLCAqx

[Yann LeCun @ylecun]

Actually, AI already saves lives.
In several countries, mammograms are examined by AI and radiologists. Reliability is improved.
In the EU, every car sold must be equipped with Automatic Emergency Braking Systems. That's AI. They reduce frontal collisions by 40%.
Modern MRI machines are equipped with AI technology that reduces the time of imaging by 4x or more. You can now get a full-body MRI in 40 minutes for about $1000. Reduced time -> reduced cost -> more/earlier detection.
And that's not counting the progress in medicine enabled by modern AI, including Nobel Prize-winning protein structure prediction.

[George Kirov @GeorgeKirov1]

Our paper is out: By analysing health records from millions of real-world patients from the literature, we can now finally answer the questions about the long-term outcomes of ECT. What we found, consistently, across well-designed studies from the UK, Canada, US, Sweden, Denmark, and Taiwan, is that ECT does not increase the risk for dementia, heart attacks or stroke, and is associated with a significant reduction in overall mortality.
https://t.co/lC13i5ANBD

[Balázs Szigeti @psybalazs]

Another day, another psychedelic trial with missing placebo response (https://t.co/HwNGGiGDwv), this time with 5-MeO-DMT. In this case the patients actually got WORSE in the placebo group. In the past month, 4 high profile psychedelic trials came out, all on treatment-resistant depression, all showing the same phenomena: there is virtually no placebo response in the control arm of psychedelic studies, see image below (negative values indicate improvement as you want less depression):
A: 5-meo vs PL, placebo response is +0.3 MADRS units (https://t.co/HwNGGiGDwv)
B: psilocybin vs active PL (nicotinamide), placebo response is -1.5 MADRS units (https://t.co/rVw0e0swu4)
C: psilocybin vs PL, placebo response is -1.2 MADRS units (COMP005 https://t.co/6fc3zVdyxJ)
D: psilocybin vs active PL (1mg psilocybin), placebo response -3.7 MADRS units (COMP006 https://t.co/6fc3zVdyxJ)

The typical placebo response in trials on major depression of antidepressants is -9 MADRS points (https://t.co/elaAtZrTdr). The typical placebo response in trials on TRD with ketamine -7 MADRS points (https://t.co/fF6DYvkb49).

[Andrea Boscutti @ABoscutti]

one of Jarrett’s best solos, imo

https://t.co/Dz57OGq5uE

[Paul Thompson @PTenigma]

Sorry I found the original quote, which I had lost: “No one imagines that a symphony is supposed to improve in quality as it goes along or that the whole object of playing it is to reach the finale. The point of music is discovered in every moment of playing and listening to it. It is the same, I feel, with the greater part of our lives, and if we are unduly absorbed in improving them we may forget altogether to live them.”

[Itai Yanai @ItaiYanai]

Science doesn’t need to go according to plan; it just needs to lead to a discovery. If doesn’t have to be done alone or together with a buddy; there just needs to be a discovery. It doesn’t need to happen fast or slow; just as long as there’s a discovery, then everybody is happy.

[Robbie Barrat @videodrome]

I'm laughing so hard at this slide a friend sent me from one of Geoff Hinton's courses;

"To deal with hyper-planes in a 14-dimensional space, visualize a 3-D space and say 'fourteen' to yourself very loudly. Everyone does it."

[Kyle Chan @kyleichan]

All Americans should think about this chart

[Andy Keller @t_andy_keller]

In the physical world, almost all information is transmitted through traveling waves -- why should it be any different in your neural network?

Super excited to share recent work with the brilliant @mozesjacobs: "Traveling Waves Integrate Spatial Information Through Time"
1/14

[Jonathan Gorard @getjonwithit]

Moths are attracted to lights because of the same mathematics that underlies twistor theory and compactification in theoretical physics: projective geometry.

It all starts from a simple observation: translations are just rotations whose center is located "at infinity". (1/11)

[The Transmitter @_TheTransmitter]

With neuroscience datasets and scientific collaborations growing in size, Gaelle Chapuis and Olivier Winter explain why neuroscience needs to create a career path for software engineers.

https://t.co/y8VpwO5R4p

[ACNP @ACNPorg]

Join us in congratulating the 2024 ACNP Travel Award Class! https://t.co/Q8H0Xdr6II @AArulpragasam @estef_azevedo @igorbandeira @ABoscutti @CarinaSBrown @brendacabreram @therealRYC @SSebastianChoi @not_kristie @ANeurogirl @PashaDavoudian @KevinFDowling @ScientificRuvvy @CassieLHendrix @deedsou @mkvarta @sj_lamontagne @lujanpema @carolinamak15 @PeteManza @DavidJMarcus2 @Molas__S

[Divyansha @divyansha1115]

Excited to share our Graph Foundation Model, 🌐 GraphFM, trained on 152 datasets with over 7.4 million nodes and 189 million edges spanning diverse domains.

🚨 Check out our preprint for GraphFM where we test how our model scales with data and model size, and show efficient finetuning on new datasets.

Link: https://t.co/dfmJWMILMv

[Jayson Jeganathan @jaysonjeg]

Do you use surface fMRI? We found spurious correlations in surface fMRI, with potentially serious implications for test-retest reliability, fingerprinting, functional parcellations and brain-behaviour associations (1/n)
https://t.co/Sv98S53P7x

[Fleetwood @fleetwood___]

Best tiled matmul animation I've found on the internet.

Thanks @wentasah

[Andrea Boscutti @ABoscutti]

@matteohallissey @Ivan_Grieco Avanti così non mollate <3

[Andrej Karpathy @karpathy]

The killer app of LLMs is Scarlett Johansson. You all thought it was math or something

[Andrea Boscutti @ABoscutti]

AlphaFold 3 predicts the structure and interactions of all of life’s molecules @google https://t.co/kYU79JBPeI

[Dan Roberts @danintheory]

Do LLMs really need to be so L?

That's a rejected title for a new paper w/ @Andr3yGR, @kushal_tirumala, @Hasan_Shap, @PaoloGlorioso1 on pruning open-weight LLMs: we can remove up to *half* the layers of Llama-2 70B w/ essentially no impact on performance on QA benchmarks.

1/