ArXiv's Next Chapter

Nano Banana 2 Lite

Google DeepMind

Matrix Orthogonalization Improves Memory in Recurrent Models

Personal website of Ayush Tambde

Forestiere Underground Gardens

Hatari – Online Atari ST/STE/TT/Falcon Emulator

Ante: A new way to blend borrow checking and reference counting

Claude Code is steganographically marking requests

I inspected Claude Code for privacy reasons and found hidden system prompt markers based on API base URL and timezone.

I ported Kubernetes to the browser

Almost 100,000 lines of LLM-generated code in 2 months, and none of it is slop.

From brain waves to words: a new path to communication without surgery

Tokyo has only two barley tea makers, we visited one to see how mugicha is made

With barley tea going into glasses as the weather heats up, we find out what goes into making barley tea.

CERN bids farewell to the LHC and enters Long Shutdown 3

Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5

We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We'll begin restoring access tomorrow, and will share an update soon. We’re grateful to our users for their patience, and to everyone who worked with us on

The first early human eggs from stem cells

A look inside the biology, engineering, and validation behind a major scientific milestone.

Scaling Laws, Carefully

Scaling laws are one of the most critical empirical findings in deep learning. The observation is simple in form: the training loss $L$ decreases predictably as we scale up model size $N$, dataset size $D$, and compute $C$, following a power-law curve, which appears as a straight line on a log-log plot. We can view scaling laws as a framework for describing the relationship between compute, loss, model size and data; at its core, it is about how to allocate precious compute optimally between $N$ and $D$.

Pystd, similar-ish functionality with a fraction of the compile time

Google copybara: moving code between repositories

Copybara: A tool for transforming and moving code between repositories. - google/copybara

Leanstral 1.5

An updated Lean 4 formal proof engineering model optimised for automated theorem proving and autoformalization. 119B total parameters, 6.5B active.

Claude Science

Claude Science is your AI workbench for scientific research. Works through your research like a skilled scientist, running the analysis and tracing every step. Spend less time stitching pipelines together, and more time on the science.

How does a pull-back car work? Illustrated teardown

Illustrating the Engineering Around Us

Claude Sonnet 5

Our most agentic Sonnet yet, with top-tier intelligence for coding and everyday professional work.