Anil Kumar

I hacked ChatGPT and Google's AI – and it only took 20 minutes I found a way to make AI tell you lies – and I'm not the only one.

I got a tip that all over the world, people are using a dead-simple hack to manipulate AI behavior. It turns out changing what AI tells other people can be as easy as writing a blog post *on your own website*

I didn’t believe it, so I decided to test it myself www.bbc.com/future/artic...

Hmm ok, well not to throw a spanner in, but I'd bet pause comes into the overton window this year, even if it's not possible. Dario and Demis literally just said it at Davos.

I have a diagram I'll share with you that demonstrates why...

Welcome to WHERE MACHINES THINK Exploring and understanding the mathematical spaces that enable artificial (and maybe natural) intelligence. Essays and analyses at the intersection of machine learning, neuroscience and physics

I'm starting a Substack newsletter, WHERE MACHINES THINK (just imagine scare quotes around the word think, to maintain appropriate skepticism). The welcome post is here: wheremachinesthink.substack.com/p/welcome-to...

The Mythology Of Conscious AI | NOEMA Why consciousness is more likely a property of life than of computation and why creating conscious, or even conscious-seeming AI, is a bad idea.

The Mythology Of Conscious AI www.noemamag.com/the-mytholog...

1/ We found that deep sequence models memorize atomic facts "geometrically" -- not as an associative lookup table as often imagined.

This opens up practical questions on reasoning/memory/discovery, and also poses a theoretical "memorization puzzle."

That you try to capture a conscious state for its own sake, because you think that matters. The true artist in my view is somebody who captures conscious states and that's the only reason why they eat. So you eat to make art. And another person makes art to eat.

And the arts is slightly different. It's a mutation that is arguably not completely adaptive. It's one where people fall in love with the loss function. Where you think that your mental representation is the intrinsically important thing.

And when the contribution is negative then the superorganism kills it in order to be fitter in the competition against other superorganisms. And it's totally brutal. I don't like fascism because it's going to kill a lot of minds I like.

Fascism is a mode of organization of society in which the individual is a cell in the superorganism and the value of the individual is exactly the contribution to the superorganism.

And in a way we are yeast. Everything we do, all the complexity that we create, all the structures we build, is to erect some surfaces on which to out compete other kinds of yeast. And if you realize this you can try to get behind this and I think the solution to this is fascism.

The thing that is able to digest anything and turn it into structure to sustain and perpetuate itself, for long as the local puddle of negentropy is available.

Evolution is about getting eaten by monsters. Don't go into the desert and perish there, because it's going to be a waste. If you're lucky the monsters that eat you are your own children. And eventually the search for evolution will, if evolution reaches its global optimum, it will be the perfect ..

The arts are the cuckoo child of life. Because the meaning of life is to eat. You know, life is evolution and evolution is about eating. It's pretty gross if you think about it.

“I find that most people serve practical needs. They have an understanding of the difference between meaning and relevance. And at some level my mind is more interested in meaning than in relevance. That is similar to the mind of an artist. The arts are not life. They are not serving life.

Hey, thanks for the Alfred snippets. I don't know if it is our fault. We did not choose these incentive structures.

omg. I just set it up a few days ago. Also what about my neovim config?

YouTube video by Neuromatch Academy Working Memory Involvement in Higher Cognition: Insights from fMRI Modeling

Our recent presentation from the Impact Scholars Program at @neuromatch.bsky.social - Neuromatch Academy!
www.youtube.com/watch?v=F5_z...

Statistical and deep learning models acted as “sensors” for task demand and WM activity, predicting how much WM is involved in emotion and language domains.

Join Our Mission | Lila Sciences Lila Sciences unveils the first-ever scientific superintelligence platform, securing $200M in funding to revolutionize discovery in life, chemical, and…

We are introducing the world's first Scientific Superintelligence platform for life, chemical and materials sciences. Join our Mission www.lila.ai/news/join-ou...

🚨 Heads up, Pythonistas 🚨

Something exciting is coming and you won’t want to miss it...PyCon UK 2025 Call for Proposals is opening soon! 🚀🐍

If you've got an idea for a talk, a workshop, or a Young Coders' session, we want to hear from you! 📝✨

👀 Stay tuned 👀

#PyConUK2025 #python #development #CFP

🚨 New Preprint!!

LLMs trained on next-word prediction (NWP) show high alignment with brain recordings. But what drives this alignment—linguistic structure or world knowledge? And how does this alignment evolve during training? Our new paper explores these questions. 👇🧵

Two Kinds Of AI Polar Opposites

Two Kinds Of AI
open.substack.com/pub/zerothpr...

YouTube video by Wolfram Institute Seminar | Joscha Bach | Can we understand consciousness using the paradigms of AI?

www.youtube.com/watch?v=FKu7...

Temu imports more than 10000 tons of fast fashion every day.

Every living system must adapt, or die.

In my first blog post, I show how this fundamental principle can be mathematized:

✅ Brains adapt, and adaptation is about KL divergence minimization.

Let's unpack the main insights 🧵[1/n]
Link: mysterioustune.com/2025/01/13/w...

🧠🤖🧠📈 #AI

I hope you’re well.

Sounds scary

There’s a single formula that makes all of your diffusion models possible: Tweedie's

Say 𝐱 is a noisy version of 𝐮 with 𝐞 ∼ 𝒩(𝟎, σ² 𝐈)

𝐱 = 𝐮 + 𝐞

MMSE estimate of 𝐮 is 𝔼[𝐮|𝐱] & would seem to require P(𝐮|𝐱). Yet Tweedie says P(𝐱) is all you need

1/3