🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

The Shams I thought was trending vs. the Shams actually trending. It happens every time :(

Handwriting leads to widespread brain connectivity - typing does not.

What if an LLM could update its own weights? Meet SEAL🦭: a framework where LLMs generate their own training data (self-edits) to update their weights in response to new inputs. Self-editing is learned via RL, using the updated model’s downstream performance as reward.

New on the Anthropic Engineering blog: how we built Claude’s research capabilities using multiple agents working in parallel. We share what worked, what didn't, and the engineering challenges along the way. anthropic.com/engineering/bu…

Well, I’ll be damned!

Workaccount2 on Hacker News just coined the term "context rot" to describe the thing where the quality of an LLM conversation drops as the context fills up with accumulated distractions and dead ends news.ycombinator.com/item?id=443087…

Sad news that Helen De Cruz passed away. We are grateful that they agreed to be part of our editorial board as they embodied the ideal philosopher: brilliant, kind, creative, endlessly curious, passionate about public philosophy, and striving for change dailynous.com/2025/06/20/hel…

RIP Helen De Cruz. youtu.be/g5GdajPPG0s?si…

Mathematics of an Artificial Neural Network:

It’s not just the Apple paper. Reasoning outside the distribution has been the core problem for neural networks for 30 years. Here’s yet another study that is a case in point. It has been my central goalpost for nearly 30 years, and I will keep emphasizing it until the

Excellent opening keynote yesterday at PST by Sabina Leonelli. Thought-provoking messaging against AI hype in science. A bit lacking on the epistemology that acknowledges successful ML use in the hard sciences, but a needed corrective to sloppy general industry narratives.

LLM’s sycophancy seems to me to be embedded in the data: we’ve normalized abusing words like “great”, we even call our interlocutors words and deeds as “awesome” and “amazing” in both corporate and casual life even when they’re anything but that.

5 weeks, 4 countries, 10 cities, 6 talks, 2 panels. Week one/first talk at the Society for the Philosophy of Technology, done!

As usual, an impressively thorough, philosophically cautious, yet thought-provoking talk by Emily Sullivan as keynote and recipient of IACAP’s Herbert Simon Award.

Next week I’ll give the META lecture at Politecnico di Milano on Serendipty, AI, and automated science. meta.polimi.it/ramon-alvarado…

Update w/updated itinerary: 5 weeks, 4 countries, 10 cities, 7 talks, 1 panel. Week 2, talk 3 and panel participation at IACAP done.

My last Q&A in Noto, Sicily for the University of Messina. I talked about what a non-anthropocentric epistemology of science that accounted for computational methods would look like. Great group of researchers in the philosophy of computer science led by Nicola Angius.

So nice to see Herbert Simon’s book in there. I’ve been reading and working with its ideas for more than a decade and continue finding uses and insights.