Hugging Face has a tiny two-person research unit called H4 building open-source ChatGPT alternatives like Zephyr-7B-α. Armed with a 1,000-A100 GPU cluster, the duo focuses on model alignment recipes and shares their training code directly with the open-source community.
GitHub just rolled out three AI-powered features for GitHub Advanced Security to help developers resolve vulnerabilities faster. Code scanning now uses LLMs to generate one-click fixes for JavaScript and TypeScript alerts right in pull requests, while secret scanning uses AI to spot unstructured leaked passwords that bypass standard rules.
The FTC just submitted a rare comment to the U.S. Copyright Office, warning that training AI on copyrighted material and deceptive synthetic mimicry could violate antitrust and consumer protection laws, even if defended as fair use. Critics argue the agency is overstepping its mandate.
SAG-AFTRA and Hollywood studios just hit another wall in strike negotiations, and AI likeness rights are the breaking point. Studios reportedly proposed owning digital scans of higher-earning performers in perpetuity—even letting them resurrect deceased actors without estate consent. SAG-AFTRA walked away, demanding clear consent and pay every single time an actor's face is reused.
Scarlett Johansson is taking legal action against developer Convert Software after its app, Lisa AI, used an unauthorized clone of her voice in an ad. The promo spliced real footage with an AI voice pitching features behind a flimsy disclaimer. Stars are finally drawing a line against deepfake ads.
Instagram is developing a customizable "AI friend" chatbot, according to newly leaked app screenshots. Users can tweak traits like age, gender, personality style, and specific hobbies before picking an avatar and jumping into open-ended chats. While Meta just launched celebrity AI personas last month, this shifts the focus to personalized everyday companionship.
World leaders just backed the Bletchley Declaration at the AI Safety Summit in Bletchley Park, agreeing to collaborate on risks posed by frontier foundation models. US Commerce Secretary Gina Raimondo also announced a new AI safety institute under NIST to work with international partners on aligning safety policies.
Collins Dictionary just named "AI" its word of the year for 2023, reporting that usage of the term quadrupled over the past 12 months. The publisher noted that artificial intelligence has evolved from a futuristic concept into an everyday presence alongside email and streaming, dominating global conversations on innovation and safety.
A federal judge just narrowed the landmark copyright lawsuit brought by visual artists against Stability AI, Midjourney, and DeviantArt. While dismissing claims that all AI outputs automatically constitute infringing derivative works, the court allowed direct infringement claims over model training against Stability AI to proceed. The artists will have a chance to amend their complaint.
Meta AI research head Joelle Pineau argues that traditional open-source licenses aren't built for models trained on vast amounts of web data and copyright risks. While critics slammed Meta's Llama 2 as not being truly open source due to usage caps, the Open Source Initiative agreed that legacy licenses fall short and confirmed it is drafting an AI-specific definition.
President Biden just signed a landmark executive order imposing strict oversight on frontier AI systems. Developers training high-risk foundation models must now alert the federal government and share their red-team safety results under the Defense Production Act, while officials develop mandatory watermarking standards to catch AI-generated fraud.
Yan LeCun just called out corporate leaders for lobbying to regulate open-source AI R&D under the guise of doomsday fears. He argued that AI platforms will hold human knowledge and must stay open like Wikipedia, warning that killing open source will leave global information in the hands of a few tech monopolies.
Stability AI just announced that CEO Emad Mostaque will attend the UK Government's AI Safety Summit at Bletchley Park. As a prominent open-source player, the company plans to push policymakers to support safety across the entire ecosystem—advocating that open models keep AI benefits from being monopolized by Big Tech.
Microsoft just shared an update on its AI safety practices ahead of the UK AI Safety Summit, highlighting how it is operationalizing its commitments. The company expanded its independent AI Red Team, added automatic provenance disclosure for AI-generated images in Bing Image Creator, and funded 125 research projects across 13 countries. Why it matters: As global regulators push for enforceable safety standards, frontier labs are working to show that voluntary guardrails are already delivering results.
Google just expanded its Vulnerability Rewards Program (VRP) to cover generative AI attack scenarios, releasing new guidelines on what qualifies as an in-scope bug. The company is also rolling out prototypes for model signing via Sigstore and SLSA to secure AI supply chains against tampering. Why it matters: Google is putting cash behind crowdsourced safety to catch jailbreaks and hallucinations before attackers do.
British Prime Minister Rishi Sunak just announced plans to establish the world's first AI Safety Institute, aimed at evaluating frontier models for bias, misinformation, and extreme risks before they are released. The government is also proposing a global expert panel modeled on the Intergovernmental Panel on Climate Change to track AI science, with Sunak defending the inclusion of China to ensure any serious safety strategy covers all major AI powers.
OpenAI just formed a dedicated Preparedness team led by Aleksander Madry to tackle catastrophic risks from frontier AI models. The group focuses on severe threats like cyberattacks, chemical and biological hazards, and autonomous replication, while rolling out a strict Risk-Informed Development Policy.
Google just made its "About this image" feature available to all English-language users worldwide. The tool lets people check when an image first appeared in Search, how other sites describe it, and metadata showing if it was AI-generated. Approved fact-checkers and journalists are also getting access via dedicated APIs.
Grammarly just announced a new feature called "Personalized voice detection and application" rolling out to business subscribers by year's end. The AI analyzes your day-to-day writing to build a custom voice profile, letting you rewrite any draft in your own distinct tone. Why it matters: it makes AI output feel authentic, but opens up tricky questions about identity and style cloning at work.
Anthropic, Google, Microsoft, and OpenAI just appointed Chris Meserole as the first Executive Director of the Frontier Model Forum and launched a new AI Safety Fund with over $10 million in initial backing. The initiative will grant funds to independent researchers developing red-teaming techniques and evaluation tools for frontier models. Why it matters: Big tech wants to establish collective safety and governance standards before regulation forces their hand.
