13 things I would have told myself before building an autorouter 🧵

I’ve spent about a year working on an autorouter for tscircuit (an open-source electronics CAD kernel written in Typescript). If I could go back a year, these are the 13 things I would tell…

AI systems have attained superhuman performance across various domains. If the hidden
knowledge encoded in these highly capable systems can be leve...

The canonical deep learning approach for learning requires computing a gradient term at each layer by back-propagating the error signal from the output towards each learnable parameter. Given the...

Let’s fix these pesky Transformer outliers using Softmax One and QuietAttention.

EPFL researchers have developed a flexible auditory brainstem implant (ABI) that closely conforms to the curved surface of the brainstem. The technology has been successfully demonstrated high-resolution “prosthetic hearing” in macaques.

HiddenLayer’s latest research uncovers a universal prompt injection bypass impacting GPT-4, Claude, Gemini, and more, exposing major LLM security gaps.

Chemputation, the execution of code controlled reaction pathways on universally reconfigurable hardware, offers a route to treating chemical synthesis as a formally computable i.e. chemputable...

We interact with computers on an everyday basis, be it in everyday life or work, and many aspects of work can be done entirely with access to a computer and the Internet. At the same time, thanks...

In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training...

Following on from the previous video at https://www.youtube.com/watch?v=cyyR7j2ChCI, I now attempt to formalize a different proof of the same assertion using the large language models Claude 3.4 Sonnet and o4-mini-high, after giving them the informal and…

We introduce the first method for translating text embeddings from one vector space to another without any paired data, encoders, or predefined sets of matches. Our unsupervised approach...

The Fiat-Shamir (FS) transform is a prolific and powerful technique for compiling public-coin interactive protocols into non-interactive ones. Roughly speaking, the idea is to replace the random coins of the verifier with the evaluations of a complex hash…

We are inspired by the possibilities of AI to help people be more creative, productive, and stay closely connected on WhatsApp, so we set out to build a new technology that allows our users around …

An artificial microsaccade–enhanced event camera for varied vision tasks in challenging scenarios is proposed.

This is very worrying. The New York Times v OpenAI lawsuit, now in its 17th month, includes accusations that OpenAI's models can output verbatim copies of New York Times content …

A first-of-its-kind study tested the safety of psilocybin on patients with mild to moderate Parkinson’s disease and found that patients experienced clinically significant improvements in mood, cognition, and motor function that lasted for weeks after the…