- Chaos Theory
- Posts
- 🥟 Chao-Down #235 Bytedance found to be violating OpenAI's terms of service in using GPT4 to train own AI models, Meta updates its Ray-Ban glasses to now have AI see, Microsoft targets nuclear power
🥟 Chao-Down #235 Bytedance found to be violating OpenAI's terms of service in using GPT4 to train own AI models, Meta updates its Ray-Ban glasses to now have AI see, Microsoft targets nuclear power
Plus, AI comes for the field of wine-tasting.
The dirty little secret to text-based generative AI models is that their output can be easily used to train competing language models. And apparently, Bytedance is capitalizing on it.
The Verge reports that Bytedance, the parent company of TikTok, has been using OpenAI’s large language models to develop its own competing foundation models by directly using the generated text from GPT4. The effort is codenamed Project Seed and internal documents confirm that GPT4 has been used in nearly every phase of development, including for training and evaluating Bytedance’s own model.
The practice is against both OpenAI and Microsoft’s terms of service, but the stakes are high enough that Bytedance is willing to embrace the legal ramifications.
As of today, OpenAI has suspended Bytedance’s account. But is it too late?
-Alex, your resident Chaos Coordinator.
What happened in AI? 📰
ByteDance is secretly using OpenAI’s tech to build a competitor (The Verge)
Meta's Ray-Ban Glasses Added AI That Can See What You're Seeing (CNET)
Microsoft’s AI Chatbot Replies to Election Questions With Conspiracies, Fake Scandals, and Lies (WIRED)
Microsoft Targets Nuclear to Power AI Operations (WSJ)
AI Could Be Helping Deloitte Avoid Mass Layoffs in The Future (Bloomberg)
AI can tell which chateau Bordeaux wines come from with 100% accuracy (New Scientist)
Always be Learnin’ 📕 📖
A Composer’s Helper: Using AI to Create New Harmonies (stanford.edu)
Guide to Prompt engineering (OpenAI API)
Why the U.S. Needs a Moonshot Mentality for Artificial Intelligence (WSJ)
Projects to Keep an Eye On 🛠
Announcing Purple Llama: Towards open trust and safety in the new world of generative AI (meta.com)
havenhq/mamba-chat: Mamba-Chat: A chat LLM based on the state-space model architecture 🐍 (Github)
The Latest in AI Research 💡
Amphion: An Open-Source Audio, Music and Speech Generation Toolkit (arxiv)
ReST meets ReAct: Self-Improvement for Multi-Step Reasoning LLM Agent (arxiv)
DreamTalk: When Expressive Talking Head Generation Meets Diffusion Probabilistic Models (arxiv)
The World Outside of AI 🌎
America Wants Hybrids. Car Companies Don't Want to Make Them. (The Atlantic)
IVF Cost: High Bills Help Black Market Fertility Drugs Thrive Online (Bloomberg)
The Coming Boom in Rare Earths (IEEE Spectrum)
A Good Night's Sleep May Help Control Blood Sugar (Scientific American)
Are your organs ageing well? The blood holds clues (Nature)
Air pollution: map shows 'concrete jungles' more polluted (bbc.com)