AI Friday
The AI Friday Brief

Good Sunday, NOLA. July 19th brings a genuine shocker: GPT-5.6 just closed a 30-year gap in convex optimization using a prompt-engineered approach. Meanwhile, Fable 5 and GPT-5.6 Sol go head-to-head on an NP-hard problem, someone built a guide for letting Claude Code control your spare Mac, and Simon Willison shipped an AI-powered SQLite query explainer. It's a weekend—but the AI world didn't get the memo.

Research Breakthrough

GPT-5.6 Closes a 30-Year Gap in Convex Optimization

After OpenAI's CDC proof announcement, GPT-5.6 used a carefully crafted prompt to solve a 30-year-old open problem in convex optimization. This isn't just a parlor trick—it's a demonstration of how frontier models can contribute to real mathematical research when guided properly. The HN discussion is full of mathematicians weighing in on what this means for the field.
Hacker News

Tools & Hands-On Guides

SQLite Query Explainer: AI-Powered Query Plan Breakdowns

Simon Willison built an AI tool that explains SQLite query plans in plain English. Inspired by Julia Evans saying "Maybe one day I'll learn to read a query plan," this tool takes the mystery out of EXPLAIN QUERY PLAN output. If you've ever stared at a query plan and thought "what does this actually mean," this is for you.
Simon Willison

Setting Up Your Spare Mac for Claude Code: A Step-by-Step Guide

If you've got an old Mac lying around, here's a detailed walkthrough for setting it up as a dedicated machine Claude Code can control. This lets you experiment with agentic code execution without worrying about it touching your main workstation. Smart approach for anyone who wants to try Claude Code but values their sanity.
Hacker News

Model Showdowns & Benchmarks

Fable 5 vs. GPT-5.6 Sol on an NP-Hard Problem: Does /goal Help?

A hands-on test comparing Fable 5 and GPT-5.6 Sol on a complex NP-hard problem, with a twist: does the /goal directive actually improve reasoning? The results are nuanced—both models struggled in different ways, but the takeaway is that explicit goal-setting seems to help structure the problem-solving process. Worth reading if you're thinking hard about prompt engineering for reasoning tasks.
Hacker News

Industry & Business

What AI Did to Stack Overflow, in a Graph

A data visualization showing Stack Overflow's traffic decline in the AI era. The drop is stark and unmistakable—ChatGPT and other AI tools have fundamentally changed where developers go for help. The HN thread is full of developers reflecting on what this means for knowledge-sharing communities.
Hacker News

Dave Eggers Told OpenAI Staff ChatGPT Was 'Silencing an Entire Generation'

Last year, Sam Altman invited author Dave Eggers to speak to OpenAI staff. Eggers didn't pull punches—he told them ChatGPT risks "silencing an entire generation" by making it too easy to outsource thinking and writing. Whether you agree or not, it's a provocative counterpoint from someone who's spent a career crafting words. The Verge has the full story.
The Verge AI

Also Worth Reading

Will AI Fix Prior Authorization—or Make It Worse?

The government is piloting a program that uses AI for insurance-coverage decisions. Ars Technica digs into whether this will streamline a broken system or just automate the frustration. If you've ever dealt with prior auth delays, this is a story worth following—the stakes are real for millions of people.
Ars Technica AI

LLM-Integrated Multivariable Calculus Course

An experimental calculus course built with LLMs integrated throughout. Students can ask questions, get explanations, and work through problems with AI assistance. It's an interesting glimpse at what AI-native education might look like—interactive, personalized, and available 24/7.
Hacker News

Also

Pass it on

Share this brief

Share this brief →