I'm excited to be speaking at the next PyAI Meetup in San Francisco on Nov 11 about using coding assistants to create high-quality, secure code :) Hope to see you there! luma.com/gjtfecc7

Py AI Meetup · Luma

Join us for an evening of cutting-edge AI discussions featuring lightning talks from leading practitioners: Samuel Colvin (CEO of Pydantic) Reducto + Modal…

luma.com

Llion Jones, co-author of the 2017 paper that launched the transformer era, says he’s ‘absolutely sick’ of transformers — and believes the field’s over-fixation on one architecture may be stifling the next breakthrough. venturebeat.com/ai/sakana-ai...

At the Man vs. Machine hackathon, co-hosted by AI nonprofit METR to test if AI helps people code faster and better, the top prize went to an "AI-supported" team (Kylie Robison/Wired) Main Link | Techmeme Permalink

Exciting day for @zed.dev - Now with an awesome debugger. zed.dev/blog/debugger

The Debugger is Here - Zed Blog

From the Zed Blog: Over 2,000 developers asked, and we delivered. Debugging in Zed is now a reality—and it's a big leap toward Zed 1.0.

zed.dev

One of the absolute greats.

Six Colors@sixcolors.com · last yr.

(4) Bill Atkinson dies at 74 (Link)

The playhead on the #SNL 50 broadcast on YouTube TV is a cowbell.

Congrats to the Astral team! uv is awesome.

Charlie Marsh@crmarsh.com · last yr.

uv is one year old today. Hard to believe its only been a year. The growth, adoption, and impact surpassed my wildest expectations. Happy birthday, uv! 🥳

Great overview of different methods to build reasoning a #LLM.

Understanding Reasoning LLMs

Methods and Strategies for Building and Refining Reasoning Models

magazine.sebastianraschka.com

A great overview of all things LLM from Andrej Karpathy, targeted at a (relatively) general audience.

Deep Dive into LLMs like ChatGPT

YouTube video by Andrej Karpathy

youtube.com

#Unsloth fixes bugs in Phi-4, dramatically improving the performance of the small #LLM. Unsloth continues to impress.

Finetune Phi-4 with Unsloth

Fine-tune Microsoft's new Phi-4 model with Unsloth! We've also found & fixed 4 bugs in the model.

unsloth.ai

Regex-based #chess? Sure, why not. Nicholas Carlini stitches 84,688 regex snippets into a 2-ply minimax engine. Pure #hacker whimsy at its finest.

"I gained remote code execution via MIDI messages to trick my synth into playing Bad Apple on its LCD." I love this kind of #ReverseEngineering project :)

Fascinating to see a systematic approach using system safety and control theory by Google #SRE. An approach like this is needed with sufficiently complex systems to help prevent and mitigate complex failures.

The Evolution of SRE at Google

usenix.org

Turns out, at least in this experiment, telling an #LLM to write better code actually helps. (Also, perhaps a rare time Betteridge's Law is disobeyed.)

Can LLMs write better code if you keep asking them to “write better code”?

Most coders want AI to write code faster: I want AI to write FASTER CODE.

minimaxir.com

How small can a #AI language model be?

TinyStories: How Small Can Language Models Be and Still Speak Coherent English?

Language models (LMs) are powerful tools for natural language processing, but they often struggle to produce coherent and fluent text when they are small. Models with around 125M parameters such as GP...

arxiv.org

Now your proof of life includes saving pixelated worlds from the demons of Doom. Once again, it is proven that Doom can run on anthing - even a CAPTCHA prompt.

DOOM® CAPTCHA

Prove you're human by playing DOOM

doom-captcha.vercel.app

Indexing code well is a tough problem, especially at scale. Interesting namespace conflict between products that ... um ... index things. #programming

Indexing code at scale with Glean

We’re sharing details about Glean, Meta’s open source system for collecting, deriving, and working with facts about source code. In this blog post we’ll talk about why a system like Glean is import…

engineering.fb.com

@kagi.com is pretty awesome. Thanks to @gruber.foo for highlighting. Worth a visit and try for any user, and a worthwhile shot at making a straightforward business model work for #search. The `?` operator to trigger an AI summary is just a truly inspired design.

@irisvanrooij.bsky.social argues that #AI should refocus on its original goal: serving as a theoretical tool to advance cognitive science—not just building smarter machines.

Reclaiming AI as a Theoretical Tool for Cognitive Science - Computational Brain & Behavior

The idea that human cognition is, or can be understood as, a form of computation is a useful conceptual tool for cognitive science. It was a foundational assumption during the birth of cognitive scien...

link.springer.com

SemiKong, built on #Llama 3.1, is an open-source #AI tailored for the #semiconductor industry. It streamlines chip design, boosts manufacturing precision, and accelerates workforce training—promising faster innovation and smarter production pipelines. A hopeful leap for smarter semiconductors.

Scaling semiconductor expertise with Llama-powered Domain-Expert Agents

Aitomatic enables semiconductor companies to build Domain-Expert Agents to capture and scale their deep domain expertise using Llama-based foundation models.

ai.meta.com

Not familiar with Betteridge’s Law? Google it (or ask your favorite AI). Today’s #AI is messy, but sound engineering will catch up. Half-baked products skew perceptions, but quality systems are possible. (See s1.ai/purple 😉) The Wright Flyer was pretty wobbly too on its first few flights.

Does current AI represent a dead end?

Eerke Boiten, Professor of Cyber Security at De Montfort University Leicester, explains his belief that current AI should not be used for serious applications.

bcs.org

Sending good energy to the #OpenAI SRE staff who are surely working late tonight. status.openai.com/incidents/6b...

High error rates for ChatGPT, APIs, and Sora

OpenAI's Status Page - High error rates for ChatGPT, APIs, and Sora.

status.openai.com

I should note that the site hosting this post I just referenced has an emulator that runs The Print Shop, and *GENERATES PDFS* of the output. ...awesome. theprintshop.club

Who else thinks the world today is missing the charm of a dot-matrix-printed banner from The Print Shop? #nostalgia

This channel is so great. Watch for a history of #ethernet #networking and why most of us are using twisted-pair cables with RJ-45 connectors.

TWISTED: The dramatic history of twisted-pair Ethernet

YouTube video by The Serial Port

youtu.be

#Minecraft server in #COBOL. This is fun.

GitHub - meyfa/CobolCraft: A Minecraft server written in COBOL

A Minecraft server written in COBOL. Contribute to meyfa/CobolCraft development by creating an account on GitHub.

github.com

Artem Zakirullin argues that reducing cognitive load in code—especially extraneous complexity—is key. Simplify conditionals, use clear names, and avoid deep nesting. Clean code is easy on the brain of a #developer. #programming

It's likely my followers have already read this about #o3, but if not you should take a moment to do so. Yesterday was a *big* day for #AI, #OpenAI, and the path toward #AGI.

OpenAI o3 Breakthrough High Score on ARC-AGI-Pub

OpenAI o3 scores 75.7% on ARC-AGI public leaderboard.

arcprize.org

This is a down-and-dirty look at building your own high-performance #AI #LLM inference engine, from raw #CUDA kernels on up. The result? Beating top-shelf libraries at their own game. Still probably best to use a supported library in production, though.

If you write Python scripts, make yourself a favor and use the Rich library to beautify their output 🐍 🧑‍💻

GitHub - Textualize/rich: Rich is a Python library for rich text and beautiful formatting in the terminal.

Rich is a Python library for rich text and beautiful formatting in the terminal. - Textualize/rich

github.com