[models] · · By ByteBulletin Editor
The 280B-parameter sparse MoE model offers a 1M-token context and novel attention mechanisms designed for efficient fine-tuning.
Read the story →editor’s picks
[research] ·
Hacker News users report that paid Gemini subscribers' chat logs are used for model training and human review unless specific activity settings are disabled, challenging the assumption that a subscription guarantees data privacy.
[funding] ·
The acquisitions integrate AI agent observability and security testing into the Falcon platform, addressing the gap in enterprise controls for autonomous AI workflows.
[research] ·
Three new papers from arXiv evaluate how AI agents handle operational state, long-term memory, and real-world collaborative workflows, revealing significant gaps in current model performance.
By cutting token costs to 30 cents per million, Revolut removed usage rationing, allowing developers to apply AI to all tasks and improving lead time by 70%.
The company's SEC filing states that internal estimates suggest it may soon meet its own definition of dramatically accelerated AI progress, prompting new safety commitments.
The Sequoia-backed startup claims its continual learning platform allows non-technical experts to improve open-source models without hiring specialized research teams.
Publishers allege Brave sold access to its scraped index to major AI labs and enabled paywall bypassing without consent.
OpenAI has reduced its fundraising goal from $30 billion to $20 billion as institutional investors show hesitation toward heavy AI infrastructure commitments.
The week was defined by the intersection of corporate transparency and technical security, as major labs disclosed financial and safety risks while developers grappled with supply-chain vulnerabilities and new agentic interfaces.
Google's prototype space data center is online, executing inference tasks with a 20-minute power constraint that highlights the engineering hurdles of orbital AI.
A new benchmark using Claude Opus 5.5 reveals that framework choice significantly impacts the token cost and time required for AI agents to build web applications.
A supply-chain compromise exposed up to 5,000 developers to a persistent SSH backdoor, with Anthropic yet to issue a public warning.
The AI lab's IPO prospectus blends a 'radical honesty' strategy with 94 pages of risk factors, including explicit mentions of potential human extinction.
Google is launching a public preview of its Connected Apps Platform, allowing developers to build and submit integrations that function directly within Search and Gemini using the Model Context Protocol.
The startup pivots from prompt-to-app builder to a personal agent that generates custom mini-apps for daily tasks.
One short email when it matters. No recaps of recaps.
