What does Opus 4.7 verify against?
Claude Opus 4.7 looks like a genuine step forward, and one line in the announcement caught my attention: the model “devises ways to verify its own outputs.” ...
Posts by Year
2026
Build to learn 🔗
Marty Cagan on the distinction between product discovery (“build to learn”) and product delivery (“build to earn”), and why AI makes the former more importan...
Math and code got there first
Quanta Magazine has a piece this week on how AI has changed mathematical research — AlphaEvolve, LLMs as collaborative partners, problems that used to take m...
Where graphs supplement LLMs 🔗
Graph-based parsers appear to outperform LLMs on relation extraction — and the gap widens as relational complexity grows. A preprint out today from Gajo et a...
MCP vs. Skills 🔗
A good breakdown of the MCP vs. Skills tradeoffs from David Mohl:
The knowledge graph as digital twin
A new paper from Wharton finds that LLM-generated Community Notes on X are rated more helpful than human-written ones across 108,000+ ratings. It’s a well-de...
Provenance, not detection
Researchers recently published a method for removing Google’s SynthID watermarks from AI-generated images with near-invisible quality loss, by reverse-engine...
The inbox apocalypse 🔗
I have started talking about the inbox apocalypse that is going to hit this year, where everything that is normally sort of reviewed and bottlenecked by h...
Ground truth is reality
Anthropic restricted Claude Mythos to vetted security researchers this week via Project Glasswing — not because it was producing false positives, but because...
Publishing’s two jobs
There’s a piece on the ergosphere blog worth reading this week about what the author calls the Alice-and-Bob problem. Alice and Bob both produce a PhD resear...
Agents, bugs, and the statistical editor
Nicholas Carlini, a research scientist at Anthropic, ran a simple bash script that looped over every file in the Linux kernel and asked Claude Code to look f...
2025
MCP lets you ship faster 🔗
I’ve been thinking a lot about this quote from Steve Krouse (via Simon Willison):
uv is the best thing to happen to the Python ecosystem in a decade 🔗
Agree with this post 100%. I switched to using uv about six months ago and it has made package management in python much easier.
AI-Generated “Workslop” Is Destroying Productivity 🔗
Let’s be considerate about how we use GenAI to write emails, articles, or blog posts. When I first started, it was fun: Wow, I can crank out a 750-word essay...
I think ‘agent’ may finally have a widely enough agreed upon definition to be useful jargon now 🔗
Via Simon Willison:
Python: The Documentary
Learning Python for data science seven years ago changed the trajectory of my career. This documentary is a great behind-the-scenes view of the people who br...
Anycrap.shop: Bring Impossible Products to Life 🔗
Is it satire? Is it an art project?
Beyond vibe coding: AI-assisted development 🔗
Via Simon Willison:
Switzerland releases its own AI model trained on public data 🔗
Switzerland released their own Llama-3-class model, trained exclusively on public sources while respecting crawler opt-out requests.
Are content credentials going mainstream? 🔗
The Content Authenticity Initiative is a collaborative effort to bring transparency to digital media. By using cryptographic signatures and standardized meta...
China Releases “AI Plus” Policy: A Brief Analysis 🔗
China released their new “AI Plus” strategy document last week when I was in Beijing. Here is some context and a translation of the policy document (via Bene...
M365 Copilot + GPT-5 = big improvement
Have you tried M365 Copilot lately? It has gotten seriously good.
Piloting Claude for Chrome 🔗
I’m not sure if we’re ready for agentic browser control. Yes, you can click each time to accept the risk, but how many of us read the T&Cs before we clic...
Chinese universities want students to use more AI, not less 🔗
While many educators in the West see AI as a threat they have to manage, more Chinese classrooms are treating it as a skill to be mastered. In fact, as th...
FDA’s artificial intelligence is supposed to revolutionize drug approvals. It’s making up studies 🔗
The FDA’s head of AI, Jeremy Walsh, admitted that Elsa can hallucinate nonexistent studies. “Elsa is no different from lots of [large language models] ...
Markitdown 🔗
This looks like a handy package for converting documents (PDF, .docx, .pptx, and more) to .md. There’s also a MCP server so you can use it with your LLM.
Reflections on OpenAI 🔗
A fascinating look into OpenAI the company:
Grok 4 delivers surprises 🔗
The AP article quotes Simon Willison:
The AWS Survival Guide for 2025: A Field Manual for the Brave and the Bankrupt 🔗
I haven’t had to figure out AWS IAM or review the Cost Explorer in a hot minute.
If MCP is the USB-C of AI agents, A2A is their Ethernet 🔗
Via @ErikJonker@mastodon.social:
Fighting AI Hallucinations One Citation at a Time: Introducing the LLM Citation Verifier
As AI writing assistants become more prevalent in academic and professional settings, we face a growing challenge: how do we maintain the integrity of the sc...
Enabling cookie consent on a Jekyll Minimal Mistakes site 🔗
Paul’s post got me about 80% of the way there, but I was still having issues with
TIL: Modern Python Package CI/CD with uv, Trusted Publishing, and GitHub Actions
Today I learned how to set up a complete CI/CD pipeline for Python packages using modern tooling. As a first-time package publisher, I wanted to make sure I ...
The bad boy of bar charts: William Playfair 🔗
Via Clarke & Esposito, an entertaining sketch written by Mike Woodward of William Playfair, who invented bar charts and pie charts in between misadventur...
Stop Building AI Agents 🔗
Hugo-Bowne Anderson1 argues that agentic workflows shouldn’t be your first choice because of their increased complexity and instability. Remember that GenAI ...
Automation of Systematic Reviews with Large Language Models 🔗
This looks very impressive, using LLMs to not only survey the literature but also synthesize the results and generate new statistically significant findings.
AI4Research: A Survey of Artificial Intelligence for Scientific Research 🔗
This is more of a survey than a critical review, and the equations on pages 8–10 seem unnecessary, but a potentially useful compilation map of what’s new as ...
Finding secrets in ‘Oops’ commits 🔗
As someone who gets confused beyond simple commits and pushes, this approach of spelunking for thought-to-be-deleted secrets in “oops” commits is a little sc...
Sometimes you just gotta get started 🔗
Via Jeff Triplett: