AI BriefingDaily AI briefing
Editor's picks come first, followed by stories repeatedly covered across days and sources.
Top stories
05- 01GPT-6 Astra Improves Qwen3.5-0.8B Math Performance 6x via Weight Space Programming
GPT-6 Astra improved the MATH-500 score of Qwen3.5-0.8B to 48.1% by modifying its weights without training.
PyTorchKR
- 02Vercel Launches Security Review in Public Beta for GitHub Repositories
Available in public beta for Pro and Enterprise teams, the feature uses Vercel Agent to detect logic flaws and offer fixes via dashboard or CLI.
Vercel Blog
- 03Revolut shifts 2,000 engineers to self-hosted GLM, cutting AI costs to a tenth of previous spend
The bank deployed 56 Nvidia B200s to run GLM internally, reducing token costs to 30 cents per million and improving engineering lead time by 70%.
Hacker News
- 04LangSmith BYOC expands to Azure and Google Cloud, joining AWS GA
LangChain manages the data plane infrastructure while keeping traces, evals, and agent deployments in the customer's own cloud account.
LangChain Blog
- 05Evernote reports 10-20% higher trial-to-paid conversion from ChatGPT Ads
Campaigns focused on a single feature drove 25% of desktop visits to new accounts, with trials starting August 28-September 12, 2026.
OpenAI Blog
Rising repositories and models
02- floci-io/floci
LocalStack alternative with 24ms startup and 13MiB memory
- Usagi-org/ai-goofish-monitor
AI-powered second-hand marketplace monitoring set up with a single line of natural language