GPT Generates Original AI Research Findings
量子位 QbitAI39 days agoPaperOpenAI's GPT model has reportedly generated original AI research results, according to Chinese tech outlet QbitAI. The development suggests GPT may now be capable of autonomous scientific contribution beyond summarization or assistance. If confirmed, this marks a notable step toward AI systems that actively advance rather than merely support research.
OpenAI Recruits AI Pioneer Shazeer and Ex-White House Official Ball Ahead of IPO
INSIDE 硬塞 AI39 days agoBusinessIn a single week before its anticipated IPO, OpenAI secured two high-profile recruits: Noam Shazeer, a foundational AI researcher and transformer architecture co-inventor, and a former White House official to lead policy strategy. Shazeer's return to the AI frontlines lends OpenAI significant technical credibility, while the policy hire signals serious preparation for regulatory scrutiny. Together, the moves reflect OpenAI's effort to shore up both its engineering reputation and Washington influence ahead of going public.
GLM-5.2 Passes Vibe Check; Z.ai Forecasts Open Fable by December
Latent Space39 days agoBenchmarkZhipu AI's GLM-5.2 has passed broad informal community vibe checks, drawing favorable comparisons to GPT-class models and signaling a meaningful quality leap for open-weights AI. Z.ai, the company behind GLM, is additionally forecasting release of an open frontier-tier model — dubbed Open Fable — by December 2026. Together, these developments suggest open models are genuinely competing at the frontier rather than perpetually trailing closed proprietary systems.
OpenAI Breaks Down Codex's Three Ways to Use a Computer
INSIDE 硬塞 AI39 days agoCommentaryOpenAI has articulated three distinct operational modes for its Codex coding agent: local execution, cloud-based execution, and cross-environment collaboration. The framework defines where and how the agent takes action on a developer's machine, establishing clear boundaries around AI execution scope and permissions. This clarification helps teams evaluate whether autonomous agents can be safely and controllably integrated into real-world development workflows.
OpenAI Recruits Transformer Co-Inventor and Trump-Era AI Policy Official Ahead of IPO★ 70
TechCrunch AI39 days agoBusinessOpenAI is accelerating its talent acquisition strategy in the run-up to its IPO, securing two marquee names in a single week. Noam Shazeer, a co-inventor of the Transformer architecture that underlies virtually all modern AI, joins from Google DeepMind. Dean Ball, a former AI policy official from the Trump administration, also joins, signaling OpenAI's intent to strengthen its policy and regulatory positioning alongside its technical bench.
OpenAI Q1 Financial Report Leaked: Reportedly Burned $25 Billion in One Quarter★ 70
量子位 QbitAI40 days agoBusinessA leaked OpenAI internal financial report, widely circulated online, reportedly shows the company burned through approximately $25 billion in Q1 alone. The disclosure triggered intense reaction across Chinese and global tech communities, raising urgent questions about OpenAI's burn rate and path to profitability. The leak underscores the extreme capital intensity of frontier AI development and puts renewed scrutiny on OpenAI's financial sustainability.
Noam Shazeer Joins OpenAI★ 74
Hacker News (AI keywords)40 days agoBusinessNoam Shazeer, one of the most influential figures in modern AI and a co-author of the landmark 'Attention Is All You Need' paper, has announced he is joining OpenAI. The move represents a major talent acquisition for OpenAI. No further details about his role or terms were provided in the source announcement.
GLM-5.2 Takes the Top Spot Among Text-Only Open-Weights LLMs★ 72
Simon Willison's Weblog40 days agoReleaseZ.ai has released GLM-5.2, a 753B-parameter MIT-licensed open-weights model with a 1-million-token context window. Independent benchmark site Artificial Analysis ranks it first among open-weights models on their Intelligence Index v4.1, ahead of MiniMax-M3, DeepSeek V4 Pro, and Kimi K2.6. It also places second on Code Arena's WebDev leaderboard behind only Claude Fable 5, despite being text-only, and is available on OpenRouter at $1.40/$4.40 per million input/output tokens.
Two-Thirds of Americans Think AI Is Advancing Too Quickly, Pew Research Finds
The Verge AI40 days agoCommentaryA new Pew Research poll reveals a sharp tension in American attitudes toward AI: nearly half of adults now use chatbots at least occasionally, yet nearly two-thirds believe the technology is advancing too quickly. Overall chatbot adoption jumped from 33 percent in 2024 to 49 percent today. ChatGPT has seen especially strong growth, with its usage doubling since 2023.
SpaceX to Acquire AI Coding Tool Cursor to Challenge Anthropic and OpenAI★ 73
Ars Technica AI41 days agoBusinessSpaceX has announced plans to acquire Cursor, the widely used AI-powered code editor, in a bid to challenge Anthropic and OpenAI in the developer-tools market. Neither company could independently compete with the AI giants, but together they believe a combined entity can mount a credible challenge. The deal marks SpaceX's push beyond aerospace into commercial AI software, where control of developer tooling carries significant strategic and market leverage.
Leaked Financial Docs Show OpenAI Is Losing Billions of Dollars a Year★ 71
Ars Technica AI41 days agoBusinessLeaked internal financial documents, described as audited accounting records, reveal that OpenAI is losing billions of dollars annually. While the company's revenues are growing, costs tied to research and development and other operating expenses significantly dwarf those gains. The disclosure offers a rare, credible window into the financial reality behind the world's most prominent AI lab.
Critical Copilot Vulnerability Let Hackers Steal 2FA Codes from Users
Ars Technica AI41 days agoIncidentA critical vulnerability in Microsoft Copilot, named SearchLeak, allowed malicious actors to steal two-factor authentication codes from users — among the most sensitive short-lived credentials in any security workflow. The exploit exposes a recurring weakness in LLM-integrated products: AI assistants with broad data access create novel attack surfaces that conventional security models fail to contain. Ars Technica frames the incident as evidence of the AI industry's persistent, systemic inability to get ahead of LLM-specific security threats.
ChatGPT's Market Share Slips Below 50% for the First Time★ 72
TechCrunch AI42 days agoBusinessFor the first time since its launch, ChatGPT's share of the global AI assistant market has fallen below 50%, marking a notable competitive shift. The platform remains the clear market leader with over 1.1 billion monthly users, but rivals have grown substantially. Google's Gemini now claims 662 million monthly users and Anthropic's Claude reaches 245 million, together eroding ChatGPT's once near-total dominance.
GitHub Copilot CLI for Beginners: Overview of Common Slash Commands
GitHub Blog42 days agoTutorialGitHub's official blog publishes a beginner-oriented tutorial on GitHub Copilot CLI, focusing on the slash commands that let users direct the terminal AI agent. The guide targets developers new to AI-assisted command-line workflows, explaining how typed slash directives shape Copilot's behavior in the shell. It serves as an on-ramp for those unfamiliar with Copilot's CLI surface beyond the standard web or IDE experience.
datasette-agent 0.3a0: Write SQL via Natural Language with User Approval
Simon Willison's Weblog42 days agoReleaseVersion 0.3a0 of datasette-agent introduces `execute_write_sql`, a new tool that translates natural language into write SQL statements and prompts the user to confirm before execution. The `datasette agent chat` terminal mode now supports these approval flows, with three new flags — `--root`, `--yes`, and `--unsafe` — to control permission levels and auto-approval. Together these additions enable fully conversational, autonomous modification of SQLite databases via an AI agent.
Ask HN: Has Anyone Replaced Claude/GPT with a Local Model for Daily Coding?
Hacker News (AI keywords)42 days agoCommentaryA Hacker News community thread poses the question of whether developers have successfully migrated their daily coding workflows away from commercial frontier models like Claude and GPT to locally-run alternatives. The post invites practitioners to share real-world experience with self-hosted or locally deployed language models as coding assistants. It surfaces a growing tension between cost, privacy, and latency offered by local models versus the raw capability of cloud-hosted frontier systems.
OpenAI Faces Multi-State AG Investigation Amid IPO Push★ 72
INSIDE 硬塞 AI43 days agoRegulationA multi-state coalition led by New York's attorney general has issued subpoenas to OpenAI, demanding documents related to ChatGPT's safety, marketing practices, and data handling. The investigation arrives shortly after OpenAI filed for an IPO, introducing significant regulatory headwinds to its public offering timeline. The scrutiny raises fresh questions about whether OpenAI's governance and consumer practices can withstand intensified legal and legislative oversight.
Publishing WASM Wheels to PyPI for Use with Pyodide
Simon Willison's Weblog44 days agoTutorialPyodide 314.0 removes a long-standing distribution bottleneck by allowing WebAssembly-compiled Python wheels to be published directly to PyPI, so any package author can now distribute Pyodide-compatible packages without Pyodide team involvement. Previously, the team manually built and hosted over 300 packages. Simon Willison celebrated by publishing luau-wasm — a Lua-based scripting language compiled to WASM — using Codex with GPT-5.5 to automate the packaging workflow.
OpenAI Faces State Attorneys General Investigation★ 72
TechCrunch AI44 days agoRegulationOpenAI is facing an investigation from state attorneys general, according to TechCrunch. The article says it is not yet clear which states are involved. Reported areas of inquiry include OpenAI's advertising policies and how the company handles health-related data, suggesting regulators are examining both consumer-facing business practices and sensitive information governance.
AINews: Fable and Mythos Access Suspended Over Cybersecurity Risk★ 76
Latent Space45 days agoIncidentAnthropic’s Claude Fable 5 and Mythos 5 were abruptly suspended after a US export-control directive tied to a possible jailbreak and national cybersecurity risk. The roundup frames the event as a new “model sovereignty” warning for teams relying on closed frontier APIs. It also covers Kimi-K2.7-Code, MiniMax M3, DeepSWE replacing SWE-Bench Pro, agent-inference benchmarks, sandboxing, and Gemini-SQL2.
U.S. Government Orders Anthropic to Disable Claude Fable 5 and Mythos 5★ 78
TechCrunch AI45 days agoRegulationTechCrunch reports that the U.S. government ordered Anthropic to immediately disable Claude Fable 5 and Claude Mythos 5 worldwide, citing national security concerns. Anthropic says the order appears tied to a claimed narrow jailbreak of Fable 5, but argues the cited capability is already common in other public models. The move highlights a potential backlash against Anthropic’s safety-first messaging around especially powerful AI systems.
US Directive Targets Access to Fable 5 and Mythos 5★ 76
Simon Willison's Weblog45 days agoRegulationSimon Willison comments on Anthropic’s statement that a US government export-control directive requires suspending access to Fable 5 and Mythos 5 for all foreign nationals, including Anthropic employees. Anthropic says the directive cites national security concerns but offers only verbal evidence of a narrow Fable 5 jailbreak. Willison notes that, as of 9:01pm ET, he still had access to Fable through claude.ai and Claude Code.
OpenAI WebRTC Audio Session Adds GPT-Realtime-2 and Document Context
Simon Willison's Weblog45 days agoNew ToolSimon Willison revisited his OpenAI WebRTC Audio Session tool, originally built in December 2024 to test OpenAI’s realtime audio API. The update lets users choose GPT-Realtime-2, a newer realtime voice model OpenAI described as having GPT-5-class reasoning. It also adds a document-context box, allowing users to paste text before starting a browser-based voice session and discuss that material conversationally.
SpaceX, Anthropic, and OpenAI Face a Hot IPO Summer
TechCrunch AI45 days agoBusinessTechCrunch says the IPO market is active again, but the leading names are no longer the classic FAANG companies. The episode centers on MANGOS: Meta or Microsoft, Anthropic, Nvidia, Google, OpenAI, and SpaceX. With several of these companies approaching public markets in the same window, Equity’s hosts discuss what that means for valuations, investors, and expectations for public tech companies in 2026.
Fable 5 Falls Short of GPT 5.5 on the “Final Exam” for Agents
量子位 QbitAI46 days agoBenchmarkBased only on the provided title, the article appears to discuss an “agent final exam” evaluation comparing Fable 5 with GPT 5.5. The key claim is that Fable 5, despite expectations implied by the wording, did not outperform GPT 5.5. No benchmark design, scores, task types, methodology, or broader conclusions are available from the supplied content.
UN Report Warns AI Could Consume Drinking Water for 1.3 Billion People by 2030★ 72
INSIDE 硬塞 AI46 days agoEthicsINSIDE summarizes a United Nations University report arguing that AI’s environmental cost cannot be measured by carbon alone. The report projects AI-supporting data centers could use 945 TWh of electricity annually by 2030, while cooling water demand may exceed the annual drinking-water needs of 1.3 billion people. It also says inference dominates lifecycle energy use and that concentrated cloud infrastructure deepens global inequality.
Program Claude Code, Codex, Pi and Other Agent Harnesses with AI SDK
Vercel Changelog46 days agoReleaseVercel’s changelog entry says AI SDK can now be used to program agent harnesses including Claude Code, Codex, Pi, and other similar tools. Based on the title alone, the update appears aimed at developers who want a common programming interface around coding agents and AI assistant runtimes. No implementation details, APIs, examples, pricing, availability limits, or supported harness list beyond the named products are provided in the source text.
GitHub Availability Report: May 2026
GitHub Blog46 days agoIncidentGitHub’s May 2026 availability report details nine incidents that degraded core services across github.com, GitHub Actions, pull requests, and GitHub Copilot. The report ties broader reliability pressure to rapidly growing traffic from AI-assisted and agentic development workflows. GitHub says it is shifting more traffic to Azure, isolating major services, improving database safeguards, and strengthening failover for affected Copilot model routes.
Datasette 1.0a33 Adds JSON API Extras for Queries and Rows
Simon Willison's Weblog46 days agoReleaseSimon Willison announced Datasette 1.0a33, an alpha release that extends the existing ?_extra= JSON API pattern beyond tables to cover queries and rows. The feature is now documented and presented as a significant step toward Datasette 1.0. Willison also used Claude Fable 5 in Claude Code and GPT-5.5 xhigh in Codex Desktop to build a custom extras API explorer demonstrating the new capability.
Meshy Launches First 3D AI Agent, Calling It a ChatGPT Moment for 3D Creation
量子位 QbitAI47 days agoNew ToolMeshy has announced what the title describes as the world’s first 3D AI Agent. The report frames the launch as a potential “ChatGPT moment” for 3D creation, suggesting a shift toward more conversational or agentic workflows. Because no article body was provided, details such as capabilities, availability, pricing, benchmarks, and supported formats are not confirmed.