Nvidia Shatters Records with $3 Trillion Valuation Amidst Global AI Chip Fever
Major AI Outages and Legal Battles Erupt: OpenAI Wrestles With ChatGPT Glitch, Meta Faces EU Lawsuit, Justice Department Investigates Tech Giants & more
Today's highlights:
🚀 AI Breakthroughs
Nvidia Surpasses Apple with $3 Trillion Market Value, Leads in AI Chip Demand
• Nvidia Corporation hits a $3 trillion market cap, becoming the first computer-chip company to reach this milestone shares surged 147% this year. ;
• CEO Jensen Huang plans annual upgrades for AI accelerators, emphasizing generative AI's role in the next technological revolution. ;
• Nvidia's growth fueled by AI chip demand, now competing to surpass Microsoft Corp. as the world's most valuable company.
Global ChatGPT Outage Affects Millions, Fix Underway by OpenAI
• ChatGPT outage began early yesterday, with initial reports surging at around 2:30 AM ET, indicating widespread disruption
• OpenAI has identified and is actively working on resolving the issue, with updated status shifts from 'mitigating' to 'continuing to work on a fix'
• The service disruption impacts a significant user base globally, with reports of issues from the US, UK, France, and beyond.
⚖️ AI Ethics
Whistleblowers Claim OpenAI Prioritizes Profit Over Safety in AGI Development
• A group of nine insiders at OpenAI are raising alarms over a purported culture that prioritizes rapid growth and secrecy surrounding its advanced AI developments
• These insiders are concerned that OpenAI's drive to develop artificial general intelligence (AGI) is overshadowing safety measures to prevent potential dangers
• Allegations have surfaced that OpenAI employs coercive tactics, including strict nondisparagement agreements, to silence employees' concerns about their AI technology.
Meta Faces Legal Heat for AI Data Usage Without User Consent, Breaches EU Privacy Laws
• Meta Platforms faces 11 complaints over privacy policy changes enabling AI training with user data without consent, potentially violating EU privacy laws
• Advocacy group NOYB urges immediate intervention by privacy regulators before Meta's policy becomes effective on June 26
• Meta claims a legitimate interest in using private data for AI development, despite previous EU court rulings against such practices.
Justice Department and FTC to Investigate Nvidia, Microsoft Over AI Concerns
• The Justice Department and FTC step up AI scrutiny, targeting Nvidia, OpenAI, and Microsoft in separate investigations
• Increased regulatory action reflects concern over AI's rapid development and its broad implications for jobs and public information
• Despite proactive domestic measures, the U.S. remains behind Europe in implementing comprehensive AI regulations.
DOJ Refuses Biden Interview Audio Release, Citing Deepfake Risks and Executive Privilege
• The DOJ argues against releasing audio of President Biden's interview, citing risk of deepfake creation and misuse
• DOJ acknowledges ample public audio exists for Biden and Hur deepfakes, but fears new release could hinder false version detection
• Transcript of Biden's interview is public, revealing several embarrassing moments despite claims of minor adjustments made by the DOJ.
Security Flaws Emerge in Microsoft's New AI Feature "Windows Recall"
• Windows Recall, part of Microsoft's Copilot Plus, faces heavy criticism for potential security vulnerabilities just two weeks post-unveiling
• Security experts and white-hat hackers express concerns, with tools like TotalRecall emerging to exploit Recall's data collection methods
• Amid controversy, detailed guides on how to disable the Windows Recall feature are circulating online, suggesting significant user apprehension.
🎓AI Academia
Selective Knowledge Unlearning: A Balancing Act for Safer, Efficient Large Language Models
• Large Language Models (LLMs) show promise in diverse fields but struggle with generating harmful content from problematic prompts. ;
• Introducing Selective Knowledge negation Unlearning (SKU), a method aiming to omit harmful knowledge in LLMs while preserving overall performance. ;
• SKU improves the balance in LLMs by effectively identifying and removing harmful content without sacrificing response quality on benign prompts. ;
Generative Large Language Models as Alternatives for Entity Matching in Data Integration Pipelines
• Generative large language models (LLMs) provide a robust alternative to pre-trained language models (PLMs) for entity matching, requiring fewer task-specific data. ;
• LLMs, such evaluating GPT4, show higher robustness in matching, able to function well in zero-shot scenarios or with minimal training. ;
• GPT4 can generate structured explanations for decisions within entity matching, aiding in the identification and correction of errors. ;
LlamaCare: A New Fine-Tuned Medical Language Model and Classification Module
• LlamaCare, a fine-tuned medical language model, equaled ChatGPT’s efficiency using a less power-intensive 24G GPU setup
• The new Extended Classification Integration module by the team tackles redundancy and enhances LLMs' performance in categorical tasks
• A dataset for one-shot and few-shot training, targeting fields like PubMedQA and USMLE Steps 1-3, is now publicly available at https://github.com/Stephen-SMJ/LLamaCare.
Comedians Discuss Impact of AI on Creativity and Bias at Edinburgh Festival
• Twenty comedians were interviewed at the Edinburgh Festival Fringe regarding their use of AI in writing comedy, expressing concerns about AI's creativity support capabilities and tendency towards bland, biased content
• Participants criticized moderation strategies in safety filtering of language models for reinforcing mainstream views and marginalizing minority perspectives, calling it a form of censorship
• The study highlighted the need for community-based value alignment and data ownership to develop AI tools tailored to artists' needs, differentiating between harmful speech and "offensive" language used in satire.
About us: We are dedicated to reducing Generative AI anxiety among tech enthusiasts by providing timely, well-structured, and concise updates on the latest developments in Generative AI through our AI-driven news platform, ABCP - Anybody Can Prompt!
We curate this AI newsletter daily for free. Your support keeps us motivated. If you find it valuable, please do subscribe & share it with your friends using the links below!