Science would be so much better if we did review (of grants and papers) constructively and collaboratively, instead of only using them to produce binary accept/reject decisions. To do that, we have to separate review processes from decisions. One idea for grants 👇

Had great fun discussing some recent work and the value of toy models in neuroscience last night! Our next workshop will be: * 16.04 * 16-20:30 * At the Crick Sign up (for free) and come along! www.eventbrite.co.uk/e/understand...

📣 We've extended the application deadline for #Neuromatch and #Climatematch Academy Students and Teaching Assistants by one week! 🚨 📅 New deadline: March 22, midnight in your local time. ➡️ Apply now: portal.neuromatchacademy.org/sign-in Tag someone who almost missed out! #TA #SummerSchool

Hot take: I prefer if subfields of machine learning not use the key-query terminology (as a mental state) in every possible chance. I think, it biases thinking towards a lower-D subspace and limits our ability to draw more general conclusions and analogies. For example, see contrastive learning.

On first thought, I have this feeling that Chaos and truly random events (like radioactive decay) will be provide predictive challenges to any artificial super intelligence.🤔

Is this the beginning? www.theguardian.com/technology/2...

I find this plot very telling about the current state of LLMs...maybe more parameters or engineering overheads will help? Source: arcprize.org/arc-agi/2/

I my opinion, I think embedding models need a little bit bigger share of the praise than is due compared to transformers. Whenever there is talk about LLMs most of the argument is warped around transformers, but I think without good embedding infrastructure, this would have been more difficult.

What is true? what is hype? What is an ad for AI? www.bbc.com/news/article...

I asked Chatgpt (the free version) about "an idea for a scientific paper I can work". Here is the first option from its answer 😆:

This might be a good test case as of why LLMs are not the answer: I tried using Scispace to do a literature review on the "role of visual experience in the development of visual feature maps". At one point it presented two theoretical frameworks for this topic: ........

"Stakeholders construct AI differently ... in ways that are useful to them ... and these differences have significant social and educational implications.” codeactsineducation.wordpress.com/2026/01/16/c...

Maybe a not so smart take, but there are four types (or subtypes) of ON-OFF directionally selective retinal ganglion cell in mice, upward, downward, backward and forward (Sans and Mashland 2015). Yet, why didn't nature use an efficient 2-cell binary code for these directions? Nature makes me 🤪.

How many independent verifications are ok for a result (experimental hypothesis) to be adopted into other work? When I read a new interesting paper, I am always excited and worried. Excited about the prospects of the results and their potential, worried about the certainty/confidence of its claims.

Submissions (short!) due for SNUFA spiking neural networks conference in <2 weeks! 🤖🧠🧪 forms.cloud.microsoft/e/XkZLavhaJe More info at snufa.net/2025/ Note that we normally get around 700 participants and recordings go on YouTube and get 100s-1000s views. Please repost.

Seems plausible to me. www.youtube.com/watch?v=_L1J...

The following article is now in press at Psychological Review. Interested to hear what people think! "The successes and failures of Artificial Neural Networks (ANNs) highlight the importance of innate linguistic priors for human language acquisition". osf.io/preprints/ps... via

I am just thinking about the benefits of investing "too" much on new Data centers in the age where breakthroughs might be just around the corner. I mean, who knows, maybe a new architecture that is more AI friendly (with the appropriate model of course) can emerge in the near future.

Is brain-wide distributed code at odds with temporal coding? 🤔 I don't know, this thought has been buzzing my brain lately.

If symbolic reasoning is the peak of the human cognitive evolution, then an AGI should not only reach this level, but evolve beyond and push through to the next level in the cognitive hierarchy.

Academic authors, here's a peek into the black box of journal publishing from an journal editor if you can bear it:

Was my pleasure working on this project, a great avenue to discuss interesting ideas and make new connections.

I find the following sentence interesting: " Instead, the work of Hubel and Wiesel and the Nobel committee's recognition directed brain research along an overly narrow path for over 50 years." onlinelibrary.wiley.com/doi/epdf/10....

I think a crucial project to undertake is to visually layout the human cortex in a friendly way (like a map with connections) at different levels of spatiotemporal details, while highlighting and summarizing relevant theories and hypotheses for the zoomed-on area. This is one of my dream projects.

It is thought that V1 complex cells are phase invariant because they integrate the activity of many simple cells. This is such a simple and easy mechanism to comprehend yet after 1000s of models and papers the visual cortex is still an enigma. I find this amusing.

Hot take of the year: I think for humanity to move forward, we need to amend our egoistic evolutionary heritage with a splash of altruistism. Maybe egoism is one of the ingredients for fast tracking intelligence, but maybe also, it is its own ceiling.

Why is it "What and Where" not "What or Where"? 🤔

A quote from the paper by Spillmann et. al. 2015: "Future research on extraclassical RF properties will have to include studies on processes of perceptual learning and memory. The temporal firing characteristics of neurons are critical in these processes (citations)....

A great paper that elaborates the interdependence between different types of weight classes and the task at hand. It also highlights the functional utility of each weight class.

Is there a proof that the visual cortex has equivariance transformations? It might be that equivariance of CNNs is an artificial property that is not available in the visual cortex? Maybe, equivariance is the property of kernels/filters stacked vertically and not arranged horizontally as in the VC?

I wrote a thing for @thetransmitter.bsky.social. The attack on scientific infrastructure happening in the US shows that relying on any one country is not a good option for science. We need to start supporting and building international, decentralised infrastructure for science.

"Redundancy is the part of our sensory experience that distinguishes it from noise" H.B. Barlow, from "Unsupervised learning 1989"

True 🤣. Source: Sarcasm on Facebook

Did intelligence arise from observing iid data or from data that is correlated in space & time? 🤔

One cannot simply retire from passion! www.youtube.com/watch?v=5lwE...

If only life was like movies. I just wanted to share this ending (spoiler warning, watch at your own risk!): www.youtube.com/watch?v=CkxX...

Science rant: since I can't find a solid account (maybe I am not looking hard enough) for a curvature receptive field that builds upon the V1 simple cells (edge and spatial freq. etc..), and locally continuous (nearby curvatures are mapped to nearby neurons). I asked Chatgpt. I like the art.

I am starting to think that the Fermi Paradox is not a Paradox at all. May there is a hidden universal law that civilizations can't overcome a certain criticality in its development.

Is it fine to support these kind of startups or research? www.youtube.com/watch?v=G90M...

It amazes me the huge difference in technology between different parts of the world. The technology stratification is vast and its rate of advancement is huge. In 200 years, we went from animal to robot labour. Thinking about that there are people out there without the modern day nuance is ....?

Hot take: I think the highest priority for computational neuroscience and machine learning is to reduce the redundancies and exploit the symmetries found in the various approaches employed to study biological, and specifically, brain computations.

I prefer using open LLMs than hush hush ones. I believe AI development should be a property of humanity not a single entity, whether company or a person. www.youtube.com/watch?v=btvB...

Interesting perspective: www.youtube.com/watch?v=BNJE...

Wow! I speculated a while back that OpenAI might be scanning API logs to get the hold-out questions on this dataset but now it seems the whole thing was secretly funded by OpenAI who had privileged access to the data. Apparently I wasn't being cynical enough.

Let's assume that Dolphins, instead of humans, are the most sophisticated and intelligent beings on earth. A very genius Dolphin made an LDLM (Large Dolphin Language Model). Can this model invent Human language or evolve into a system that can? I hope this question ignites some discussions.

Here is why I think that it is unlikely that LLMs will lead to AGI. Starting with the movie Automata (www.imdb.com/title/tt1971..., don't mind the 3 stars, I think the movie is underrated). This movie has a scene that hunts me to this day, a rogue AI develops intelligence that far surpasses humans.