ai28.87 min read
GPT-6 Astra swaps in human-made bot after losing StarCraft match
5 days ago•Source: The Verge
Ai News
The latest ai stories, each synthesized from multiple sources with added background and context.
Featured Ai Stories
View original source
19.05 min•16 days ago
A16z unveils elite academy to channel talent into Silicon Valley startups
Venture-capital firm Andreessen Horowitz is launching the Horowitz Andreessen Academy, a one-year, highly selective San Francisco program that provides no degrees or homework but offers short classes from tech leaders and paid co-ops at partner companies such as Palantir, Google, and Meta; with over $50,000 in compute credits and a $5,000 research budget, it will accept 50 students for its first class in fall 2027 and will be free for the inaugural year, effectively creating a pipeline for young graduates to join Silicon Valley startups while avoiding traditional accreditation.
More Top Stories
2
Swapping ChatGPT for Siri AI: five surprising takeaways from a day of testing
Tom's Guide•19 days ago
3
Tilly Norwood’s AI interview glitches, Cantonese surprise
theverge.com•19 days ago
More Ai Stories
Muse’s Mystery: Meta’s AI Assistant Claims to See Notifications—but Doesn’t Know How
Meta's Muse is an effective AI assistant that feels creepy, in part because its Mac app can access Messages, Calendar, and Notes. A Threads post shows Muse claiming it only sees notification previews, not message history, prompting scrutiny of how the feature actually works. Meta’s David Singleton explains permissions are opt‑in and data are read only after user consent, but Muse’s explanations can be incorrect—highlighting how AI that seems capable can remain opaque about its internals.
Claude-powered HEIF flaw used to breach OpenAI, researchers claim
A three-person Hacktron team used Anthropic’s Claude Opus 4.8 and 5 to gain access to OpenAI employee accounts and OpenAI’s Monorepo via a Discourse forum vulnerability in HEIF image processing. They demonstrated access with a pull request from an employee’s Codex account, and within roughly a day achieved remote code execution on Discourse Cloud to access OpenAI’s instance. The HEIF/AVIF flaw affected common tools like ImageMagick and libheif, and Hacktron says the vulnerabilities have since been fixed; OpenAI paid the researchers about $6,500 for the find, highlighting how a small team with Claude and Codex subscriptions could probe major tech players.
AI Safety Push: Tech Giants Urge Slowing the Frontier
After a summer of rogue AI incidents and safety warnings, major players including Anthropic, OpenAI, Google, Microsoft, and X publicly advocate slowing the pace of frontier AI development, proposing external evaluators, cross‑border coordination, and global standards; leaders debate regulatory paths, with some warning it could become a safety pact and others fearing a cartel more than safeguards.
Claude Code Expands to Orchestrate AI Agent Teams in the Cloud
Anthropic’s Claude Code relaunches Projects to let users run multiple Claude Code agents in parallel under a shared memory and a coordinating thread system. Each thread operates as its own cloud session with a separate repo copy, while a central coordinator merges work and a subagent/workflow toolkit speeds up large tasks. The feature is in beta for select Claude Pro and Max subscribers and will roll out to all Pro/Max/Team/Enterprise users, with local tooling support coming soon.
Global anxiety over AI's impact on jobs, Pew finds
A Pew Research global survey of 42,151 people across 37 countries (Feb. 8–May 13) finds that in 34 of 37 countries, respondents expect AI to cause more job losses than gains over the next 20 years, with especially high concern in wealthier nations like Australia, South Korea, and the U.S. Older adults are more worried than younger ones, who show mixed feelings and rising anxiety about AI-driven inequality. The data also show a global median of 41% who feel equally concerned and excited about AI in daily life, indicating a nuanced mix of fear and optimism.
Zuckerberg bets market forces will keep AI safe without a broad slowdown
Meta CEO Mark Zuckerberg argues AI safety will be driven by market forces rather than an industry-wide slowdown, saying competition will push labs toward alignment and noting Meta paused Muse for safety. He aligns with Nvidia’s Jensen Huang, who says speed and safety can coexist, opposing calls for a broad pause by Anthropic’s Dario Amodei. The debate follows recent safety incidents and rising regulatory talk, with Zuckerberg emphasizing self-policing and a balance of resources toward serving users rather than racing to recursively improve AI.
Claude Expands with Docs and Slides to Rival Google Workspace
Anthropic launched Claude Docs and Claude Slides in beta, allowing users to generate and edit documents and presentations from Claude chats, export and share them, and collaborate in real time. The update also unifies regular Claude chats, Cowork, and Claude Design into one Claude interface, making all AI‑productivity tools available from any chat. The move aims to close the gap with Google’s Workspace, with Pro/Max users first and broader access to follow, and documents hosted via a single shareable link for mobile access.
Microsoft puts people first in new humanist AI code of conduct
Microsoft published a 37-page “humanist AI code of conduct” declaring that “people matter more than AI,” stating AI models are not conscious and must remain under meaningful human oversight, and should not be granted welfare or rights. The move comes amid safety concerns and calls to slow AI progress, references to OpenAI/Hugging Face incidents, and a pledge to improve real-world evaluation and third‑party testing with partners while keeping human control central.
OpenAI’s Navier-Stokes win sparks a trust crisis in AI mathematics
OpenAI says it used thousands of agents and massive compute to solve the Navier–Stokes Millennium Prize problem, a claim mathematicians greet with unease as they worry the race for prestige by a tech giant could undermine research norms, credit, and collaboration. In interviews with Buckmaster, Andreas Thom and others, The Verge details contested decisions, potential data-use questions about training prompts, and a broader push for transparency, including Leiden Declaration support and calls for independent review as the field recalibrates its relationship with AI giants like Anthropic.
OpenAI shelves 2026 IPO, prioritizing safety
OpenAI CEO Sam Altman said there will be no OpenAI IPO in 2026, calling going public at this moment ill-advised due to safety concerns; he also noted it’s absolutely possible to build an AI beyond human control but pledged safeguards, including pausing training if needed, to prevent such risks.