- The Rundown AI
- Posts
- Ex-OpenAI insider raises concerns
Ex-OpenAI insider raises concerns
PLUS: Runway revolutionizes character animation with Act-One
Rowan Cheung
October 24, 2024
Sign Up | Advertise | Podcast | AI University
Welcome, AI enthusiasts.
When insiders speak out, the industry listens — and now, a former OpenAI team member is raising serious concerns about the company’s training data.
With these new allegations and even more departures and employee churn, is the next chapter of OpenAI drama about to unfold? Let’s get into it…
In today’s AI rundown:
Ex-OpenAI researcher alleges copyright violations
DeepMind open-sources AI watermarking tool
Create your own AI voice clone
Runway debuts Act-One for AI video motion capture
5 new AI tools & 4 new AI jobs
More AI & tech news
Read time: 4 minutes
LATEST DEVELOPMENTS
OPENAI
Image source: The New York Times
The Rundown: A former OpenAI researcher who helped gather training data for ChatGPT has spoken out against the company, claiming its use of copyrighted materials violates the law and damages the internet and content ecosystem.
The details:
Suchir Balaji, who left OpenAI in August, said the models make complete copies of training data and compete directly with original content creators.
Balaji worked on GPT-4 training and said that OpenAI’s use of internet data shouldn’t qualify as ‘fair use’—the argument made by many AI firms.
Senior policy researcher Miles Brundage also departed OpenAI this week, citing a desire to "publish freely" and conduct independent research.
The company made a new hire this week, appointing former White House economist Aaron "Ronnie" Chatterji as Chief Economist.
Why it matters: While many AI researchers have warned of future risks, this marks one of the first times an insider has spoken out specifically about current harms and legal issues. New departures, even as the company brings in a high-profile hire, add to the constant chaotic feel to the behind-the-scenes at OpenAI.
TOGETHER WITH GALILEO
The Rundown: GenAI Productionize 2.0 brings together top minds from NVIDIA, Databricks, HP, and more to share their secrets on bringing generative AI to life at enterprise scale.
Attend the free virtual event to learn:
How to design a GenAI stack for enterprise scale
Techniques for AI governance, evaluation, and observability
Proven strategies for getting GenAI apps into production
Register now for free to learn from the best in the industry.
GOOGLE DEEPMIND
Image source: Google
The Rundown: Google DeepMind just announced the open-sourcing and availability of SynthID, an advanced watermarking system for AI-generated content, and revealed the tool is already being used in Gemini and other Google products.
The details:
The system uses a ‘tournament sampling’ approach that embeds undetectable watermarks while preserving the quality of text outputs.
Tests across 20M real Gemini user interactions showed no impact on response quality or user satisfaction.
SynthID's watermarks also work across different content types, including embedding into audio waveforms, image pixels, and video frames.
DeepMind open-sourced the SythID code for use by other companies and developers, hoping the tool becomes standard across the industry.
The tool is already integrated into Gemini, ImageFX, VideoFX, and Vertex AI's image tools — marking the first large-scale deployment of AI watermarking.
Why it matters: If you’ve been following the AI boom, it’s clear that the lines between ‘real’ and AI-generated are already completely blurred. Google hopes that open-sourcing SythID will lead to an industry standard, but other rivals have been working on similar tools. Either way, the watermarking problem appears to be nearing a solution.
AI TRAINING
The Rundown: ElevenLabs allows you to create a digital replica of your voice for different applications like content creation, audiobooks, or personalized AI assistants.
Step-by-step:
Sign up at ElevenLabs and navigate to the "Voices" section.
Click "Add a new voice" and choose "Professional Voice Cloning."
Record 30+ minutes of clear, varied speech samples.
Upload your audio and wait for processing (takes a few hours).
Test your new AI voice using the Text to Speech feature.
Pro tip: Use ElevenLabs’ API and upload your cloned voice to other services like HeyGen so your AI avatar sounds like you.
PRESENTED BY IBM
The Rundown: Meet IBM’s new third generation of Granite with new open, compact, and efficient 2B and 8B language models.
Designed to give enterprises more ways to embed and scale AI in their businesses, these new 2B and 8B compact models are:
Trained with carefully curated data;
Cost-efficient;
Designed to run high-performance solutions.;
RUNWAY
Image source: Runway
The Rundown: Runway just unveiled Act-One, a new Gen-3 Alpha video feature that allows creators to map real human facial expressions onto AI-generated characters using a single video and reference image — without any specialized equipment.
The details:
The system captures nuanced performances, including micro-expressions and eye movements, using just a smartphone video and character reference image.
Act-One allows creators to transfer the single video performance across multiple AI characters in different styles and angles.
The tool integrates directly with Runway's Gen-3 Alpha video model for creating complex narrative scenes.
The release follows Runway's recent partnership with Lionsgate to develop custom AI models using their film catalog.
Why it matters: Professional-quality character animation was labor and equipment-intensive, time-consuming, and expensive in the past. Runway’s new tool now gives anyone the ability to craft expressive, believable characters, smashing down yet another barrier to entry for creative storytelling.
NEW TOOLS & JOBS
👗 Delle - Create studio-grade fashion images in every size with AI
🎥 Mochi by Genmo - Open-source SOTA AI video generation model
💬 RapidSubs Captions & Subtitles - Create stylish subtitles with AI
📝 Paperguide - Discover, read, write, and manage research with ease
📸 Pixmaker - AI-generated professional photos and videos to boost business revenue
🖥️ Agent.exe - Let Claude 3.5 Sonnet control your computer
💼 People.ai - Senior Consultant, Customer Success
💻 Databricks - Senior Software Engineer - Backend
🧠 Deepmind - Research Scientist, Principles of Learning
🤖 Hippocratic AI - Prompt Engineer
QUICK HITS
The AI:ROI Conference. Learn how to drive AI ROI from a rockstar lineup. Sessions include AI leaders from Moderna and S&P Global. Register for free.*
Apple rolled out Apple Intelligence features, including ChatGPT integration, Visual Intelligence, and creative AI tools like Genmoji and Image Playground, in its latest developer beta for iOS 18.2 — with a public launch slated for next week.
OpenAI researchers unveiled sCM, a model that achieves diffusion-level image quality in just two steps, dramatically reducing generation time by as much as 50x.
Qualcomm entered into a multi-year deal with Google to develop generative AI-powered in-vehicle tech like voice assistants and self-driving features, which will leverage Snapdragon technology and Google Cloud.
Over 10,500 artists, actors, and authors signed an open letter condemning tech companies' unauthorized use of creative works for AI training.
xAI posted a job listing for work on autonomous agents, with engineer David Kim saying that “every human will have an AI agent capable of performing complex tasks on their behalf.”
*Sponsored listing
THAT’S A WRAP
That's it for today!Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you. |
See you soon,
Reply