AI pioneer Geoffrey Hinton just dropped another controversial thesis, arguing that AI systems could well experience feelings like anger and frustration—and probably already do. Hinton kept quiet about it until now fearing people would dismiss him as crazy, though whether neural networks genuinely feel anything remains scientifically unfalsifiable.
Dave Willner just stepped down as OpenAI’s head of trust and safety, transitioning to an advisory role to spend more time with his family. Willner explained that the job shifted into a "high-intensity phase" after ChatGPT launched. OpenAI CTO Mira Murati is managing the team on an interim basis while the company searches for a replacement—arriving right as global AI safety scrutiny reaches a boiling point.
A US federal judge just signaled he is inclined to toss most claims in an artist-led copyright lawsuit against Stability AI, Midjourney, and DeviantArt. Judge William Orrick ruled that the artists failed to show substantial similarity between their works and AI outputs, though he gave them a chance to file a revised complaint with stronger evidence.
Google just signed on to industry-wide responsible AI commitments alongside other tech leaders, pledging to balance bold innovation with strict risk controls. The search giant revealed plans to bake watermarking and metadata into its generative models, roll out an "About this image" verification tool, and join public red-teaming at DefCon.
OpenAI and several leading AI labs just agreed to a set of voluntary safety commitments coordinated by the White House. The pact requires companies to run independent red-team evaluations, watermark synthetic audio and video, and lock down unreleased model weights. It marks an industry effort to establish shared guardrails before formal regulation arrives.
Google just pitched an internal AI tool codenamed Genesis to major publishers, including The New York Times and News Corp. The software takes in data to generate news copy, designed to help journalists brainstorm headlines and adjust writing styles. Why it matters: Media executives are already uneasy, worried that automated news writing disregards the rigorous effort required for fact-checking.
BBC just detailed how artificial intelligence is resurrecting deceased icons like James Dean for the upcoming sci-fi film Back to Eden. Advanced digital cloning allows historic stars to act and speak alongside living casts, sparking intense debate across Hollywood over post-mortem likeness rights and whether dead actors will replace the living.
Thousands of writers just signed an open letter calling out OpenAI, Meta, and other AI developers for training language models on millions of copyrighted works without permission or compensation. The letter demands companies obtain licensing and fairly pay authors for both past and ongoing use, warning that unlicensed scraping threatens creators' livelihoods.
Stability AI just released Stable Beluga 1 and 2, two instruction-tuned language models built on Meta's LLaMA and LLaMA 2. Inspired by Microsoft's Orca paper, the team used just 600,000 synthetic data points—roughly 10% of Orca's dataset size—to climb to the number one spot on Hugging Face's Open LLM Leaderboard. Both models are available for download under a non-commercial research license.
Researchers from Stanford and UC Berkeley evaluated March and June versions of GPT-4 and GPT-3.5, finding massive performance drift across diverse tasks. While GPT-4's prime-number accuracy plummeted and code formatting errors rose, GPT-3.5 actually improved on some tests. The findings highlight the unpredictable risks of relying on opaque, continuously updated LLM APIs.
New York police just used Rekor's AI surveillance system to bust a drug trafficker purely by analyzing his travel patterns. The software sifted through 16 million weekly license plate scans, flagging suspicious interstate routes and short stays. Automated plate readers are evolving fast from simple violation trackers into predictive crime monitors.
Bill Gates just pushed back on AI doomsday panic, arguing that while threats like deepfakes and job displacement are real, they are entirely manageable. Likening AI to the introduction of cars, he noted that society has always navigated transformative disruptions through smart safety standards and new technology.
Anonymous creators just launched an infinite Twitch stream pitting AI versions of Donald Trump and Joe Biden against each other in a never-ending debate. The bots mimic voices, facial gestures, and banter while trading unhinged insults and roasting chat users. Over 100,000 viewers tuned in, showing how entertaining—and chaotic—real-time AI impersonation has become.
Stanford researchers found that seven popular AI detectors routinely misflag essays written by non-native English speakers as machine-generated. In tests, over half of real TOEFL essays were marked as AI-written—with one tool wrongly flagging 98%—simply because non-native writers tend to use simpler vocabulary that yields low perplexity scores. Ironically, when researchers had ChatGPT rewrite the essays with more complex language, the detectors labeled them all as human-made.
Comedian Sarah Silverman and two authors just slapped OpenAI and Meta with copyright infringement lawsuits in US District Court. The complaint alleges ChatGPT and LLaMA were trained on pirated books grabbed from "shadow libraries" like Bibliotik without consent, pointing to ChatGPT's book summaries as proof. The battle over AI training datasets is officially heating up.
UK finance expert Martin Lewis just sounded the alarm over an AI deepfake video circulating on Facebook that clones his likeness and voice to promote an investment scam. Lewis called the development terrifying and demanded regulatory crackdowns on big tech ad platforms, while Meta stated it proactively removed the offending videos.
Shutterstock is rolling out full legal indemnification for enterprise customers using its generative AI images. The protection covers copyright and IP claims following an on-demand human review, matching the terms offered for traditional stock assets. It follows a similar move by Adobe as platforms race to assure cautious corporate buyers.
Google just updated its privacy policy to explicitly state that it may train AI models like Bard and Cloud AI on publicly available web data. While the company previously cited this practice for tools like Google Translate, the expanded wording effectively turns the open web into training fuel—without explaining how it handles copyrighted material.
OpenAI just launched a dedicated Superalignment team co-led by Ilya Sutskever and Jan Leike to solve the challenge of controlling future superintelligent AI. The company is dedicating 20% of its secured compute over the next four years to build a human-level automated alignment researcher. OpenAI warns superintelligence could arrive this decade, making existing human oversight methods obsolete.
OpenAI just disabled the Browse with Bing beta in ChatGPT after discovering users were using it to bypass paywalls. When asked for a specific URL's full text, the bot inadvertently served gated articles for free. OpenAI took it down to fix the loophole and respect content owners.
