Pinned Tweet
I will get a PhD one day. No matter what or what anyone says to me, I will get that PhD.
16
2,439
Reza Rezvan retweeted
A bit of advertising: I recently completed revisions for a tutorial paper titled "Geometric Functional Inequalities for Markov Processes: An Invitation via the Langevin Diffusion", which can now be found online at arxiv.org/abs/2611.07785.
Reza Rezvan retweeted
you either die getting schmidhubered or live long enough to become a schmidhuber yourself
I'm sorry that I'm doing this, but this is well known for a while.
It was one of the key points of the NeurIPS 2022 paper "Revisiting Neural Scaling Laws in Language and Vision" by... also Googlers! arxiv.org/abs/2209.06640
Lilian has a vibe-coded widget at the end of her scaling laws blog where as soon as you move a slider, this is immediately obvious too. lilianweng.github.io/posts/2…
Andreas (also Googler until recently) also mentions the ambiguity in passing in his blogpost blackhc.net/blog/2026/margin…
Reza Rezvan retweeted
This chart shows that - thus far - 0% of the AI research inside Anthropic happens autonomously.
This is not a sign of imminent runaway ASI, as I made clear here. rameznaam.com/p/ai-rsi-isnt-…
"The Smartest People in AI Believe This Is the World’s Most Important Chart" wsj.com/tech/ai/ai-anthropic…
"It took a whole village of Claudes to produce the data in this chart"
Reza Rezvan retweeted
Its true - this stubbornness is a core part of the culture and a key survival mechanism at both societal and individual levels. Consider that every country invaded by Arabs in the 7th century now speaks Arabic - except Iran. And despite multiple devastating invasions over centuries, Iran repeatedly rebuilt its distinct administrative structures and preserved its core culture.
P.S. The 800yr war also devastated Iran and enabled the Arab invasion, which they were powerless to defend.
Reza Rezvan retweeted
Quick thread on AI and productivity:
Lots of people say this chart shows AI is now showing up in the productivity data.
But I made this chart to argue the opposite: a trend that pre-dates AI and only shows up in the US doesn’t look like AI.
Here’s how I’m thinking about the data and underlying dynamics:
Reza Rezvan retweeted
Very important study to make note of if you're a senior mathematician, especially if you're an AI-accelerated researcher. What's the point of existence of a department populated by old rotting corpses with no one left to hold out a candle for the future?
"Early career researcher responses to AI in math" (by Anna Brandenberger, Hanna Mularczyk, Dora Woodruff): arxiv.org/abs/2610.40584
Abstract: "(…) However, the personal stakes are highest for early career researchers. (…)"
Reza Rezvan retweeted
I know many PhD students are questioning what their paths mean as AI accelerates scientific discovery. That anxiety deserves to be taken seriously.
I wrote some reflections in a recent blog post. My students found them helpful, so I’m sharing them more widely.
I hope it offers a little light as we navigate this strange time. Please share it with someone who might need it.
pages.cs.wisc.edu/~sharonli/…
Reza Rezvan retweeted
Digital minimalism
vitalik.eth.limo/general/202…
Reza Rezvan retweeted
On efficient algorithms, the Exponential Time Hypothesis, and losing the plot.
argmin.net/p/galactus-the-de…
Reza Rezvan retweeted
i think as long as you `import torch/jax`, you can never qualify as frontier.
because frontier naturally implies you have the ability to get there by yourself.
it's just logical and commonsense. there is no argument around this.
Reza Rezvan retweeted
An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4
Reza Rezvan retweeted
There is an (unpatchable?) supply-chain zero-day in PyTorch autograd that allows a regularization term to use the model's weights as RAM and SGD as its CPU clock to run any computation during training.
Reza Rezvan retweeted
1/9 Current optimizers are static during training. What if the optimizer could think and react while the network trains?
We let an agent watch the training dynamics and rewrite the optimizer's source code as training goes: a reactive optimizer.
In every language and image setting we tried that has no tuned recipe, it matches or beats the best recipe one can one-shot, and it guards against loss spikes and overfitting by itself. In the near future, optimization decisions may belong to agents, not humans.
📝 xingjianbai.com/reactive-opt…
💻 Try it on your task: github.com/xingjian-bai/opal
🧵
Reza Rezvan retweeted
Implicit Transfer Operators (proceedings.neurips.cc/paper…) gets the industry treatment and a significant scale up in the new BioEmu2.
Public preview of BioEmu2! @MSFTResearch AI for Science’s new foundation model for biomolecular structure, dynamics and function. Open weights, MIT license.
github.com/microsoft/bioemu2
Movie: BioEmu2-simulated binding of intrinsically disordered peptide PUMA to MCL-1 receptor. 🧵⬇️