iiinova — where your digital self keeps evolving

Why is more quantity coorelated with lower quality? Salami papers existed before AI? In my opinion, it is the user not the technology. Cheating, plagiarism, paper mills, salami papers, etc.. existed before AI.

100

I can't argue against that, but this is a problem with the tool maker? Factoring out the tool maker, we have in our hands a technology, if used right, has the potential to greatly accelerate productivity. In my opinion, the quesion is not "to AI or not To AI", it is "Centralised AI vs Decentralised"

Is the function space of ANNs involves anything that involves a network of computing units? Will architectural innovations be independent of biological motivations or not? In my opinion, the breakthroughs in AI is based a corpus of studies in fields like neuroscience and psychology. 2/2

020

I think it would be great if we can define what is an AI paradigm? What is the function space of ANNs? Where will the architectural innovations come from? For example, is the AI paradigm building bigger models with larger contexts and feeding more data? 1/2

120

True, and I would argue that this is an accountability problem and not a technology problem?

210

Once the above points are clear, promoting them and upholding them through accountability is a convenient next step to take. If there is failure in upholding accountability then this a problem about how accountability is upheld.

The next question is, what defines fraudulent? In academia, is producing a huge body of research in a record time fradulent? Is it the quality of the research? Is it comprehensibility and accessibility? These kind of points should be made clear so as to remove a lot of confusion. 2/n

In my opinion, AI has merits. I consider it as a tool. As a tool, its impact depends on the provider and the user. The provider is a different topic, so I will talk about the user. Now if a user uses it in a fraudulent way, is it the problem of the user or the tool? I think it is the user. 1/n

Good point, but then this can be made a problem of ethics? Can it be compared to plagiarism? In my opinion, advocating accountability and ethical behaviour is one of the ways these problems can be mitigated.

I agree and that is a problem. Also, in my opinion, influence has been a side project of media and technology, and it seems more prevalent and approachable through AI use. I don’t know, maybe more awareness and decentralised local AI can solve these problems?

100

Karim Habashy @krhab.bsky.social · 25/09/2026

What is the difference between being influenced by an article online and by what Claude wrote? I would question things if the author didn’t screen the output. In my opinion, screening and taking responsibility of what is written is the final verdict.

220

AI wars? AI grants vs AI reviewers 🔥🔥🔥. I guess the winner is the one who is token maxing more? 😂. Also, there is a third bigger winner.

000

Karim Habashy @krhab.bsky.social · 19/09/2026

Ok, hot take 🔥: I think the attached paper is interesting and cool. However, I think if problems can be defined, solutions can be identified. In my opinion, the more we try to find problems in AI logic and shortcomings, the more we accelerate its progress.

arxiv.org

000

Karim Habashy @krhab.bsky.social · 14/09/2026

In my opinion, AI articles need to deferentiate between two definitions of AI aligenment: 1) AI systems act in accordance with human values, goals, and intentions. 2) AI systems do as you were told precisely by their creators. There is a big difference between the two!

000

Reposted by Karim Habashy

Melanie Mitchell @melaniemitchell.bsky.social · 10/09/2026

I wrote down my thoughts about the last several weeks of AI hell. aiguide.substack.com/p/misleading...

aiguide.substack.com

Misleading Metaphors, Real Risks

What To Fear from AI Agents and How to Reclaim Our Human Agency

1019981

Karim Habashy @krhab.bsky.social · 08/09/2026

A quick search, in terms of power consumption, that is like 120 MWh, or in human brain units, 250,000 human brains working for 8 hours for 3 days! These numbers are astronomical 🤯

Karim Habashy @krhab.bsky.social · 08/09/2026

Ok, I will be precise here, they do the analysis correctly, but they output the diagram wrong. The model system is x_dot = 1 + ax + x^2, where ‘a’ is a parameter. I checked it again now and they corrected the diagram, but I have the history anyways.

000

Karim Habashy @krhab.bsky.social · 08/09/2026

I know it is not a comparison, but free tier Gemini fails to produce the bifurcation diagram for a 1D system. I mean could the difference be that huge?

201

Reposted by Karim Habashy

Jeffrey Bowers @jeffreybowers.bsky.social · 08/09/2026

This is a great organisation I’ve got involved with. Please consider getting involved or donating. If you have any questions please feel free to get in touch.

032

Reposted by Karim Habashy

Dan Goodman @neural-reckoning.org · 03/09/2026

#SpikingNeuralNetwork people - this year's SNUFA workshop is live! Talks by: ⭐ Eugene Izhikevich ⭐ @giuliadangelo.bsky.social ⭐ Mihai Petrovici ⭐ Susanne Schreiber Submit your abstracts by Sept 25th. Registration is open! More info at: snufa.net/2026/ cc @fzenke.bsky.social @albada.bsky.social

02416

Karim Habashy @krhab.bsky.social · 15/08/2026

Interesting article, I will keep it in mind when I am working in the future, as this domain is for interest to me: bsky.app/profile/krha...

Marcus Ghosh @marcusghosh.bsky.social · 14/08/2026

Reposted by Karim Habashy

Now published @commsaicomp.nature.com with @neural-reckoning.org We: 🤖 Define a taxonomy of neural network architectures (here 128 unique structures) 🦾 Train > 25,000 models on different navigation tasks #RL 🧠 Link each architecture's behaviour to its memory dynamics doi.org/10.1038/s444...

A taxonomy of recurrence. We defined a family of neural networks for exploring questions in neuroscience and machine learning. Here each network is represented by three circles joined by different connections (short, coloured lines). Networks are joined by longer, pale grey lines if they differ by one connection. Moving from left to right takes us from the simplest to the most complex network.

13411

Finally, we show that our approach, as a surrogate for the genomic code, can provide one of the highest forms of functional compression when evaluated on the MNIST benchmark, achieving ≈ 95% accuracy with only ≈ 1800 parameters.

000

Later, we showed that our approach is beneficial for encoding innate functions that support generalizations based on a factorial code. This was demonstrated by contrasting our approach in the hidden layer against a baseline where the hidden weights are directly encoded.

100

For results, we first show that indirect encoding is sufficient to produce V1 orientation maps with the associated line filters (left map and filters). Furthermore, these maps emerge more efficiently when utilizing our approach of decomposing the layer into a kernel plus its associated rotations.

100

The core method is that the filters of a neural network layer can be represented by a single kernel that is differentiably transformed (rotated) along the spatial dimensions of a layer, and these rotations can be acquired by an indirect encoding network.

Factorization and spatial encodings: a hypothesis about the foundations of the genomic code

New (and updated) preprint is now out!🔬🧠 We propose a hypothesis regarding how the genome, despite its limited information-carrying capacity, can initialize a brain with billions of neurons that comprise a diverse set of functions. Relevant to: #neuroscience #devleopmental-biology Details 👇

biorxiv.org

The genomic mechanisms that efficiently encode the initial architecture and synaptic connectivity of neural circuits remain poorly understood. We hypothesise that two primary mechanisms — spatial enco...

1124

Reposted by Karim Habashy

Ben Williamson @benpatrickwill.bsky.social · 06/08/2026

Looks like if you're one of the 300miliion who ever took a MOOC with Coursera or Udemy, your data just got turned into assets for Andrew Ng's latest "AI native" edtech "personalized learning" scheme, which Coursera just invested $100m in. investor.coursera.com/news/news-de...

Our strategic investment in LearnVector has the potential to be a force multiplier for our growth. Andrew's unmatched AI expertise pairs his agentic AI with our platform and assets that turn learning into validated mastery — a clear competitive advantage that we believe can accelerate our business and expand the impact we deliver for individuals and enterprises around the world.
— Greg Hart, CEO, Coursera

Source https://learnvector.ai/

Building on the broader reach we’ve gained through Udemy, this investment reflects our conviction that AI can help grow the market itself by making learning more continuous, more personal, and more embedded in people’s daily lives. In other words, we’re focused on creating a much bigger opportunity for learning overall.

As LearnVector develops new AI-native learning experiences, we’re exploring powerful ways to combine those advances with Coursera’s unique strengths: trusted content, insights from more than 300 million learners, and one of the world’s largest ecosystems of university and industry partners.

Source https://blog.coursera.org/coursera-invests-in-learnvector-to-build-the-future-of-ai-native-learning/

912777

Karim Habashy @krhab.bsky.social · 04/08/2026

A question I ask myself all the time, how complex should I make things be? "Oversimplified models can, of course, give misleading results, but excessively detailed models can obscure interesting results beneath inessential and unconstrained complexity." Dayan and Abbott (2001)

000

Karim Habashy @krhab.bsky.social · 29/07/2026

In my opinion, there is an innate ancient evolutionary drive to use technologies that brings comfort to save energy. However, not every evolutionary bias is useful now a days and active interventions are needed to ensure the continuation of our civilisation. Too grand, I know 😄

140

Reposted by Karim Habashy

Eiko Fried @eikofried.bsky.social · 24/07/2026

“In our most detailed model, judge-level effects explained 61% of the total variance, whereas combined paper and journal-level effects accounted for only 7%.”

1349

Karim Habashy @krhab.bsky.social · 22/07/2026

In my opinion, the most sought for skills in future academic job posts might be: “Asking LLMs good questions” “Making sense of what the LLM did” “Fixing and cleaning the LLM output” 😂

Karim Habashy @krhab.bsky.social · 19/07/2026

Any advice on how to survive with such conditions? 😆 www.youtube.com/watch?v=wxN2...

youtube.com

Academia's Dirty Secret Is Going to Court

YouTube video by Sabine Hossenfelder

000

Reposted by Karim Habashy

Dan Goodman @neural-reckoning.org · 16/07/2026

New preprint (well, very updated). 🤖🧠🧪 We find that an abstract model of neuromodulation lets spiking neural networks perform much better, particularly in challenging noisy environments, using less energy. Relevant to #neuroscience and #neuromorphic computing. 🧵👇 www.biorxiv.org/content/10.1...

biorxiv.org

Neuromodulation enhances the capability and efficiency of spiking neural networks

Spiking neurons underlie the brain’s extreme energy efficiency, and therefore have great potential in neuromorphic computing, although realising this efficiency in practice has proven challenging. We ...

26927

Reposted by Karim Habashy

Dan Goodman @neural-reckoning.org · 15/07/2026

Is the brain a computer solved in 4 posts so that we never have to talk about this again (please please please). 1. It's not the same physically as a computer - it's made of meat. Hopefully this one at least is uncontroversial.

4585

Reposted by Karim Habashy

Jeffrey Bowers @jeffreybowers.bsky.social · 02/07/2026

There is a great deal of interest in whether LLMs challenge Chomsky’s claim that human language acquisition requires some sort of innate inductive bias (e.g., some sort of universal grammar). aclanthology.org/2020.coling-... pemt.ru/wp-content/u... doi.org/10.1037/rev0...

aclanthology.org

Priorless Recurrent Networks Learn Curiously

Jeff Mitchell, Jeffrey Bowers. Proceedings of the 28th International Conference on Computational Linguistics. 2020.

1104

Reposted by Karim Habashy

Dan Goodman @neural-reckoning.org · 16/06/2026

Interesting Transmitter article by Brian DePasquale on what will be the effect of the increasingly widespread use of LLMs to code up research ideas in neuroscience. Includes a somewhat sceptical commentary from me. 😉

12410

Karim Habashy @krhab.bsky.social · 13/06/2026

Memory accessibility might be dependent on the current brain state. Think of Hopfield network, where the input is the current stimulus + brain state. It is not that you needed time, it is that on an X day the current brain state facilitated the access of Y memories in my opinion.

000

Karim Habashy @krhab.bsky.social · 08/06/2026

Interesting take about TokenMaxxing Can a TokenMaxxer company Rez-up smaller companies and candidates 😂. www.youtube.com/watch?v=eYK2...

youtube.com

I rant for 9 minutes about company stupidity

YouTube video by Maximilian Schwarzmüller

000

Karim Habashy @krhab.bsky.social · 14/05/2026

In my opinion, the problem of consciousness is that some people can’t accept that chemicals and electric charges can be the reason for their experiences and feelings. The transformation from physicality to subjective experience might not be ever quantified, but nonetheless is true.

Karim Habashy @krhab.bsky.social · 10/05/2026

Maybe I don’t get it, but I am not 100% on comparing LLMs simulating consciousness to computers simulating a hurricane. Simulations are abstractions with certain assumptions. However, if a system can emulate and predict every behaviour of another system, aren’t they functionally equivalent?

000

Reposted by Karim Habashy

Nico Schuck @nicoschuck.bsky.social · 06/05/2026

How fast agentic AI has changed our work is head-spinning and there's been a lot of debate in our lab. So we drafted guidelines to help navigate this new reality. Sharing here in the hope it’s useful for others & curious to hear thoughts schucklab.gitlab.io/ai-policy.html #AcademicSky #compneurosky

schucklab.gitlab.io

Schuck Lab — AI Policy

Schuck Lab policy on generative and agentic AI

57013

Karim Habashy @krhab.bsky.social · 05/05/2026

I see papers write "we use Euler" for integration in dynamical systems. Euler is inherently unstable for stable points that are center. It would be great to stress if one is using pure Euler or other modifications like Euler-Cromer or Verlet. I think this is an important detail in a papers methods.