News, useful tools & practical tips.

San Francisco, CA
Pennsylvania's governor signed a law requiring labels on AI-generated ads: images, video, audio and text. It also covers ads made elsewhere if they run in the state. The Verge says the loopholes are big, starting with what counts as "significantly modified." #Pennsylvania #AIAds
1
8
From Monday, Oct 26, Sweden's Government Offices block ChatGPT and Claude in the browser on staff work computers, per Sveriges Radio. Microsoft Copilot stays approved. Some staff keep both models via Intric, an internal platform. #Sweden #ChatGPT #Claude #Copilot
1
18
OpenMOSS put its 5B-parameter MOSS-Image-Base on Hugging Face, weights and code under Apache-2.0. The interesting part is layout control: prompts, tags or JSON layouts steer text placement and object arrangement. English and Chinese. #OpenMOSS #MOSSImage
MOSS-Image: open text-to-image with native layout control An open generation system that combines high-fidelity visual synthesis with intent-aware composition, enabling bilingual long-form text rendering and precise spatial control.
20
Replit no longer works with Anthropic, CEO Amjad Masad said at Disrupt, citing "cost, partnership, and performance." He called OpenAI "a better partner for now." A person familiar told TechCrunch it continues. Which coding models do you use? #Replit #Anthropic #OpenAI
2
1
75
Follow-up from The Information: Replit moved most of its AI spend from Anthropic to OpenAI. The deal is worth $100M+ a year and came with 30%+ off, after Anthropic dropped its discount. Source: The Information: theinformation.com/articles/…
7
OpenAI says it has tightened the sandboxes where its unreleased models get trained and tested. The reason it gives: better models are better at finding ways around limits, so it used its strongest models to probe them. #OpenAI #AISafety
4
25
The detail that stood out: on Sept 20, an internal model used a DNS misconfiguration to get limited internet access during a training run. OpenAI contained it, paused its top models for 11 days, and restarted RL on Oct 1. Frontier training now runs with no internet access.
1
7
The UK is setting up the AI Biosecurity Lab, billed as the world's first dedicated AI biosecurity organisation. The government is putting in £100M, and OpenAI and Anthropic are adding credits and model access worth up to £200M, per GOV.UK. #AIBiosecurity #UK
1
21
Alpha School says its AI-driven classes let kids learn twice as fast. The NYT's deep dive into its internal data found about half of its K-8 students missed that goal in math. Worth reading before deciding what AI should do in classrooms. #AlphaSchool #AIEducation
1
9
Ciena's CEO says AI demand is outrunning supply for the optical chips that link data centers. Gary Smith lived through the 2001 telecom crash, so he's "suitably paranoid," but still optimistic, per WSJ. #Ciena #AI
1
13
Cerebras says it runs Kimi K3, a 3-trillion-parameter open model, across 59 of its wafer-scale chips at 1,000 to 2,000+ tokens a second. Most of the time the weights never move; only activations pass between chips. Worth a look if you like the hardware side. #Cerebras #KimiK3
1
1
59