AI Reading for Friday January 31
A deep dive into DeepSeek … the public narrative is a bit simplistic - SemiAnalysis
Wouldn't be shocking if it were a top-down initiative with ties to government, CCP, access to industrial espionage. when someone spends $1b and open sources a ton of IP, it's probably for strategic advantage, not a direct moneymaker.
Open source zealots gonna zealot. But yeah, could be the core LLM is hard to monetize and it will go the way of Linux and Chrome with a bunch of open source models and platforms building profitable ecosystems on top of them.
o3-mini maybe today? - Wired
o1 now free in Microsoft Copilot as 'think deeper', for 5 queries a day I think - LaptopMag
Amodei advocates for tight GPU export controls on China - Windows Central
The US seems to question whether all of DeepSeek’s Nvidia chips were legal for export. - Bloomberg
Lot of Nvidia exports through Singapore. Just because the last training run used 2000 H800s doesn't mean they didn't use a lot more chips and export-restricted ones for research.
Google reports on who it sees criming with Gemini - The Register
Google widens Gemini 2.0 beta - TechInAsia
)
Waymo to test its driverless system in ten new cities in 2025, starting with Las Vegas and San Diego - Engadget
Adding AI to Google Sheets - Google Workspace Updates
Distilled DeepSeek models in Amazon Bedrock - Amazon Web Service
Not the real deal R1 which is now in Microsoft AI Foundry, wouldn’t have thought Microsoft would be first out of the blocks.
Follow the latest AI headlines via SkynetAndChill.com on Bluesky