AI is quietly crossing a line we didn't expect so soon: from assistant to researcher.
DeepMind just solved 9 open Erdős problems, some untouched for decades, for a few hundred dollars each. Meanwhile, Karpathy's 65-line config file is pushing Claude's coding accuracy to 94%. DHH shipped 30k lines with GPT-5.5.
The pattern is clear: we're not prompting anymore. We're directing autonomous systems. The bottleneck is shifting from capability to trust.
|
|
Together with
|
·
|
|
|
|
|
|
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
|
|
Top Repo
|
|
|
|
Frigate is a free open-source home camera app that keeps footage off the cloud
|
|
17,038 Likes
|
|
Frigate is a free, open-source home security system that runs entirely on your own hardware. No cloud. No subscription. Your footage never leaves your house. Here is what makes it worth your attention. It is designed for real-time AI object detection on IP cameras, with all processing done locally. That means the AI watching your driveway runs on your machine, not Amazon's servers. What can it actually do? - Tells the difference between a person, a car, a dog, and a blowing tree branch
- Labels and identifies faces directly within the platform
- Reads license plates
- Runs on anything from a Raspberry Pi to an Nvidia Jetson
Why does this matter? The FTC charged Ring with allowing any employee or contractor to access consumers' private videos, resulting in a $5.8 million settlement. With Frigate, there is no company in the middle. Frigate runs as a Docker container alongside Home Assistant. The repo has 32,000+ stars on GitHub. MIT licensed. Built by one person with a day job.
|
|
|
|
Presented by GitLab
|
|
Your AI Agent Is Flying Blind
|
|
Agentic AI is reshaping software development, and the technology leaders adopting it early are setting the pace.
GitLab Transcend brings you unfiltered insights from peers deploying it now, plus an early look at where GitLab is taking it next.
Join GitLab Transcend, June 10 (free & virtual): - Hear from enterprise engineering leaders on real-world AI adoption
- Workshop: Top 5 Duo Agent Platform use cases
- Automated code review, pipeline fixes & vulnerability remediation
Intelligent orchestration, now with context.
|
|
|
|
Top Repo
|
|
|
|
Karpathy's 65-line Claude config boosts AI coding accuracy from 65% to 94%
|
|
5,977 Likes
|
|
You ask AI to add a button. It rewrites three files, adds a config system, and breaks your tests. Sound familiar? That is the exact problem a 65-line text file just fixed for over 220,000 people on GitHub. Andrej Karpathy (ex-Tesla AI, OpenAI co-founder) called out AI coding agents for guessing instead of asking, overcomplicating everything, and touching code they were never supposed to touch. Developer Forrest Chang turned those complaints into four rules inside a single CLAUDE.md file, pushing AI coding accuracy from 65% to 94%. The four rules are simple: - Think before coding: state assumptions, ask when unsure, never guess
- Simplicity first: write the minimum code that solves the problem, nothing extra
- Surgical changes: every changed line must trace back to your request
- Goal-driven: turn vague instructions into verifiable success criteria before starting
Claude Code automatically reads CLAUDE.md at the start of every session. Drop the file in your project root and those rules apply instantly, no setup needed. Works in Cursor too.
|
|
|
|
Top Repo
|
|
|
|
DHH says GPT-5.5 wrote most of 30,000 lines of new code in his project
|
|
5,173 Likes
|
|
GPT-5.5 just pulled off something real. DHH, creator of Basecamp and Ruby on Rails, used it to write the majority of 30,000 lines of new code for Omarchy 4, a Linux distribution. That scale of conversion would not have been possible without AI. What made GPT-5.5 shine here? Two things: - Agentic coding (AI that runs multi-step tasks on its own, without you babysitting every move). You can give it a messy, multi-part task and trust it to plan, use tools, check its work, and keep going.
- Code explanation. It helped DHH understand his own complex JavaScript in Basecamp, including subtle interactions he had forgotten about.
- QML coding (a language for building desktop UIs) where GPT-5.5 reportedly outperformed Claude Opus 4.7 significantly.
Under the hood: GPT-5.5 has a 1,050,000 token context window and 128,000 max output tokens. That massive memory is what lets it hold an entire codebase in mind at once. You still need to review the output, but the heavy lifting is now doable at a scale that simply was not feasible before.
|
|
|
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.
Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
|
|
|
Looking to promote your company, product, service, or event to 250,000+ AI developers?
|
|
|
|
|