An open-source project, tamserve, lets many cameras share one SAM 2-style tracking model with flat memory, so it runs indefinitely. Its author reports up to 3.8x faster than Meta's code at the same accuracy, cutting cost per camera per day on one RTX 4090 from $2.40 to $1.07 at 30 fps.
Chinese AI chipmaker SigmaStar Technology plans a Hong Kong listing as soon as this year to raise up to $300 million, people familiar said, as Chinese tech firms seek a fundraising window. Timing and size may still change.
Anthropic released Claude Fable 5.2, which scored 65 on the Artificial Analysis Intelligence Index, surpassing Google's Gemini 4 Carbon at 60. The model generates output 30% faster and is priced at $8/$40 per million input/output tokens, with Jump Trading, Balyasny and Box as users.
Several studies assess AI tools across multiple cancer diagnostic applications, highlighting medical imaging and diagnostic support as key areas for clinical AI adoption.
Harris County, Texas is considering a broader AI policy to speed government services, but experts flag concerns over taxpayer spending, data privacy and accountability. The item was pulled from the Nov. 12 agenda; the current policy was approved in November 2024.
The Hacker News community discussed a piece arguing that an AI crash should be expected, centering on concerns about overheated valuations and investment in the sector.
Edge0 open-sourced Edge0-145B and Edge0-120B models with a unified inference framework. Edge0-145B is the world's first open-source 145-billion-parameter on-device model, delivering 80+ tokens/s and 28GB peak memory on a laptop, with capabilities rivaling frontier cloud models.
A developer released RealtimeVAD, which uses a 10ms energy detector instead of silero voice activity detection to signal end-of-speech sooner. Benchmarked on about 3,200 turns from the author's assistant logs, it was initially about 110ms faster and still about 100ms faster after a fix for quiet speech, at the cost of more false positives.
An OpenAI board member warns that an AI takeover could be only months away. The article offers no verifiable details or specific basis for the timeline.
A Reddit post says Baidu's new Ernie 5.5 Flash Preview, with 280B total and 14B active parameters, performs strongly on an aggregate 'Agentic Coding' index spanning 12 coding benchmarks such as Terminal-Bench, DeepSWE and SWE-bench Pro. The poster contrasts it with the AA index, which includes only two coding benchmarks.
SpaceX plans to invest $74 billion in data center campuses in Texas ($24B) and Arizona ($50B). Its AI business now exceeds half of revenue, with Anthropic committed to pay up to $120.5B through 2029 and Google paying $920M monthly for compute.
The post only mentions Nvidia CEO Jensen Huang attending a board meeting today, without details on discussions or decisions.
That is everything