- Chaos Theory
- Posts
- 🥟 Chao-Down #301 OpenAI previews an AI model for creating custom voices, New York City's AI chatbot hallucinates to tell people to break laws and do crimes, How AI is making financial fraud easier
🥟 Chao-Down #301 OpenAI previews an AI model for creating custom voices, New York City's AI chatbot hallucinates to tell people to break laws and do crimes, How AI is making financial fraud easier
Plus, Forbes breaks down the fall from grace of Stability AI's CEO.
Apparently, OpenAI has had access to voice cloning tech since 2022 but has refrained from making it public over safety concerns. In a recent blog post, OpenAI previewed Voice Engine, a platform that can mimic speakers based on only 15-second audio samples.
Based off OpenAI’s text to speech API, Voice Engine can create natural sounding speech with emotive and realistic voices in English, Spanish, French, or Chinese. However, they’ve chosen to limit the tool’s rollout and have only selected a small set of partners to evaluate the tech with.
In their blog post, they write:
We recognize that generating speech that resembles people’s voices has serious risks, which are especially top of mind in an election year.
While some voice cloning startups like ElevenLabs are seizing the moment and raising large sums of money, OpenAI is taking a more deliberate approach in releasing their tech. Will they be proven correct in the long term?
-Alex, your resident Chaos Coordinator.
What happened in AI? 📰
Microsoft Reportedly Building ‘Stargate' to Transport OpenAI Into the Future (Gizmodo)
New York's AI chatbot tells people to break laws and do crimes (qz.com)
The wrong way to study AI in college (The Atlantic)
Large Language Models’ Emergent Abilities Are a Mirage (WIRED)
AI is making financial fraud easier, Treasury Department says (qz.com)
Stability AI Founder Emad Mostaque Tanked His Billion-Dollar Startup (Forbes)
Always be Learnin’ 📕 📖
Avoid blundering: 80% of a winning strategy (asmartbear.com)
What happens to the tech startups that never go public? (Substack)
Selling AI : Category Creation of a Different Flavor (tomtunguz.com)
Projects to Keep an Eye On 🛠
langchain-ai/langchain-extract: A simple web server that allows you to extract information from text and files using LLMs. (Github)
developersdigest/llm-answer-engine: Build a Perplexity-Inspired Answer Engine Using Next.js, Groq, Mixtral, Langchain, OpenAI, Brave & Serper (Github)
The Latest in AI Research 💡
ViTAR: Vision Transformer with Any Resolution (arxiv)
Mini-Gemini: Mining the Potential of Multi-modality Vision Language Models (arxiv)
The Unreasonable Ineffectiveness of the Deeper Layers (arxiv)
The World Outside of AI 🌎
Tweeting your research paper boosts engagement but not citations (Nature)
Scientists Find Human Brains Are Getting Larger and Larger (Futurism)
Nepo-Homebuyers: More Than One-Third of Gen Z and Millennial Homebuyers Plan to Use Family Money For Down Payment (Redfin)
The world’s broken market for medicines (ft.com)
South Korea hopes new speed train links will help boost birthrate (Reuters)
Bad Haircut? A Hot Chinese App Is Giving Americans Blunt Advice (WSJ)
One Last Bite 😋
