AI Roundup 070: AI whistleblowers
AI Roundup 070: AI whistleblowers
June 7, 2024.
AI whistleblowers
Current and former OpenAI and Google DeepMind employees signed an open letter to foundation model companies, warning of a culture of recklessness and secrecy.
Why it matters:
The letter is the latest symptom of the turnover at OpenAI - it comes in the wake of major departures from the company's safety teams.
At the same time, OpenAI and Anthropic continue to publish safety research on the interpretability of their models.
But it's also a reminder that many leading AI researchers still believe we’re on track to create AGI (and likely destroy humanity in the process) within the next few years.
Elsewhere in AI anxiety:
Cara, an artist-run, anti-AI social platform, picked up over 600K users as artists grow frustrated with Meta's AI policies.
Israel appears to have used AI content and fake social accounts to influence US lawmakers.
A new report found Amazon’s use of AI in its warehouses isolates workers and negatively impacts union organizing drives.
And in case you missed it, here’s the story on how I fell for a deepfake Elon Musk’s crypto scam.
Computex
Nvidia's Computex event was this week. I'm no hardware expert, but a few announcements stood out.
What to watch:
Nvidia's CEO Jensen Huang detailed plans to upgrade its AI accelerators every year, starting with a Blackwell Ultra chip in 2025, and a "Rubin" chip in 2026.
Huang also announced that major computer manufacturers are rolling out new AI infrastructure powered by Nvidia Blackwell GPUs.
Nvidia G-Assist, an AI chatbot that guides you through video games.
Nvidia and AMD are teaming up to bring Copilot Plus features to gaming laptops.
AMD announced two new Ryzen AI chips, aimed at mobile devices and lightweight laptops.
The MI325X GPU, AMD's answer to the Nvidia H200.
And Intel showcased Lunar Lake, an overhauled AI laptop chip.
Elsewhere in AI infrastructure:
Crypto miners are getting swept up in the AI boom, by renting out data centers and selling computing power to AI clients.
Microsoft plans to invest $3.2B in Sweden's cloud and AI infrastructure over the next two years.
And xAI is expected to develop a new supercomputer based in Memphis (which could explain why Elon diverted some of Tesla's GPUs).
Here comes the fuzz
Federal regulators at the Justice Department and the Federal Trade Commission agreed to divide responsibility for investigating Nvidia, Microsoft, and OpenAI.
Between the lines:
In this agreement, the DOJ will investigate Nvidia for antitrust violations, while the FTC will investigate the relationship between Microsoft and OpenAI.
The DOJ’s top antitrust enforcer has said he's looking at "monopoly choke points" in the landscape, including AI, cloud, data, and GPUs.
And to top it all off, the FTC is also opening an inquiry into the Microsoft/Inflection AI deal from last quarter.
Elsewhere in the FAANG free-for-all:
Google rolled out its note-taking assistant NotebookLM to over 200 new countries.
WhatsApp announced new AI features, including tools to create ads in English-speaking markets.
Amazon’s Project PI looks for product defects before they ship.
And sources detailed how Apple fell behind in the AI arms race, as the company gears up to reveal a new OpenAI-powered "Apple Intelligence."
Things happen
Mistral adds model customization options and SDK for fine-tuning. A look at the uptick of AI-generated TikToksfeaturing UK politicians. YC is lobbying Washington to support "Little Tech." The right to not be subjected to AI profiling. ByteDance appears to be renting GPUs from Oracle to bypass US sanctions. Scientists should use AI as a tool, not an oracle. Nvidia's market cap breaks $3T. A look at Sam Altman's sprawling $2.8B+ investment portfolio. How billionaire VCs are fighting over the future of AI. Analysis finds 90K+ images of anime characters on gen AI sites. The AI revolution is already losing steam. White Stork, Eric Schmidt's company building AI-enabled military drones. Q&A with Lisa Su on working with TSMC and competing with Nvidia. Humane has been in talks to sell itself to HP for $1B. AI-generated thumbnails are present on Amazon Prime Video. Asana unveils "AI teammates" for beta testing. Shutterstock made $104M from content licensing last year. The world's first beauty pageant for AI creators. Stability releases Stable Audio Open, a new model trained on fully licensed content. FTX sells its remaining Anthropic shares. Showrunner, a streaming services for AI-generated series. Linux may be the best way to avoid the AI nightmare. ElevenLabs rolls out its Sound Effects tool. Tribeca and OpenAI plan to feature short films made with Sora. Man scammed after Facebook AI confirmed nonexistent support number. How online privacy is like fishing. How it feels to get an AI email from a friend. AI is a False God.
