Nathan

I have no moat and I must scream.

[Nathan @NathanS_T]

For anyone wondering, no, I'm not a bot. I'm just some random person on the internet tweeting to no one.

[Nathan @NathanS_T]

@tldraw ya'll allow uploading video files but not audio files? What's up with that?

[Nathan @NathanS_T]

"neural networks are interpretable actually"... How did we not figure this out sooner?

[Nick @nickcammarata]

interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had

with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works

https://t.co/zHij5zwKGT

[Nathan @NathanS_T]

@TheMathFlow How'd you make this? Processing?

[Nathan @NathanS_T]

@yacineMTB RSI has been happening since the dawn of the computer. We just replaced human workers with electrons, so it was harder to realize that every improvement from there-on was actually an improvement to ourselves. /s

[Nathan @NathanS_T]

We're living in a golden age for learning

[Nathan @NathanS_T]

@bartmossorg @maxxmalist Wait. Actual banger wth

[Nathan @NathanS_T]

@etiennejcb Are the interactions actually being simulated? Or is it all trickery?

[Nathan @NathanS_T]

@ZebinWilson @DeGatchi How would pruning work? The brain rejects connections if the neurons don't respond well to each other, not that familiar with SSNs, but I can't imagine how that would work here

[Nathan @NathanS_T]

@DeGatchi >after reading that Zebrafish NN tweet, i wondered: how much the GT of the statistical manifold connectome contributes to the intelligence relative to the weights?

Google showed the neural architecture matters quite a bit with their Weight Agnostic Neural Networks (WANNs).

[Nathan @NathanS_T]

@ArielKwiat Care to elaborate?

[Nathan @NathanS_T]

Open source is legit a must-have nowadays. Anything wrong with the software? Just get claude to fix it. We truly live in the future.

[Nathan @NathanS_T]

@aidangomez We've come full circle.

[Nathan @NathanS_T]

@haydendevs Is this implying Tibo is going to eat me?

[Nathan @NathanS_T]

Man, LLMs are actually so good for porting or modifying your favourite applications. Literally just ported a Linux only clip editor over to Windows. Very nice

[Nathan @NathanS_T]

I'm a slave to an audience who will never see my work.

[Nathan @NathanS_T]

@MelvilleTrader @fchollet >If you crossed a black hole's event horizon you wouldn't notice anything different.

Other than becoming spaghetti...

[Nathan @NathanS_T]

@Kulshekhar @Teknium >Within 3 minutes
...and 2 days. Aug 16 does not equal Aug 18.

Still impressive nonetheless

[Nathan @NathanS_T]

@fleetingbits This is just a one-time prompt, right? Do you do this _everytime_ you desire an intuitive top-down view of your codebase?

I'm of the opinion this is a phenomenal usage of AI, but I've always personally desired something more reusable.

[Nathan @NathanS_T]

@haydendevs idk dude. make gpt-5.7 make me a sandwich? I don't think that requires AGI...

[Nathan @NathanS_T]

Depth on the other hand...

[Nathan @NathanS_T]

Playing with @zzznah's MorphoHDL as a way of illustrating basic neural networks.

[Nathan @NathanS_T]

I wanted to make a wider network, but unfortunately the graph representation gets a bit ugly after 4... (I'm sure there's a way to make it nice, I just don't know how)