The daily briefing
Latest publication
How stories are selectedIn focus
Property Finder discovered 96.8% of its evaluation labels reflected a single model's predictions rather than human verification, systematically favoring the incumbent. Rebuilding ground truth and optimizing prompts enabled a more cost-effective model choice.
Source: aws.amazon.com
Meta's Muse AI agent is now available in Singapore with connections to local services including Shopee, Grab, Singapore Airlines and Lazada for tasks like shopping, dining and travel planning.
about.fb.com
Opera's new WebAI standard lets users run open-weight AI models directly in Opera One, processing data on-device rather than in the cloud. The preview supports models including Google's Gemma and Apple's foundation model on Mac.
techcrunch.com
Anthropic's IPO filing acknowledges lack of laboratory experience and drug development challenges, tempering CEO predictions that AI would cure most diseases and deliver a "deluge" of drugs within five to 10 years.
wired.com
Retell AI's contact center platform grew from 30 million to over 100 million monthly calls by September 2026, with OpenAI models generating 80% of responses and customers reporting resolution rates above 80%.
openai.com
The longer view
Historical coverage, revisited. A rotating selection refreshed with each day’s site build.
Historical coverage ·
How OpenAI's September 2024 release of o1 reasoning models marked a fundamental shift in AI development strategy and performance benchmarks.
Revisit the storyHistorical coverage ·
How Anthropic's massive November 2025 infrastructure commitment signaled a new phase in enterprise AI competition and domestic manufacturing.
Revisit the storyHistorical coverage ·
On November 30, 2022, OpenAI launched ChatGPT as a free research preview, swiftly reaching 1 million users and igniting global discourse on AI's potential impact.
Revisit the story