🥟 Chao-Down #355 Meta releases Llama 3.1 completely in open-source, OpenAI offers free fine-tuning for GPT4o-mini, YouTube looks to legally license music from content labels for AI song generator
Plus, AI-based lie detectors are better than humans at spotting lies.
The long-anticipated day of Meta’s latest open-source release is here.
Developers have been waiting for the 405 billion parameter model from Meta to drop ever since it was announced back in April. Promising a model comparable to GPT4 that people could run on their own and have full control over, Meta overdelivered by updating its entire class of small, medium, and large offerings.
What’s unique about this release, in addition to the model beating benchmarks of all types, is that Meta is allowing for people to use the output of Llama 3.1 as synthetic training data for downstream tasks, removing some of the stringent restrictions of building with these large language models. Now people can create smaller versions of their own models using data that Meta’s Llama 3.1 generates.
Seeing the threat of developers moving away from GPT4, OpenAI is responding to Meta’s announcement by offering their customers free fine-tuning until September 23 of their latest GPT4o-mini model.
So when are we going to see GPT5?
-Alex, your resident Chaos Coordinator.
What happened in AI? 📰
Introducing Llama 3.1: Our most capable models to date (Meta)
AI arms race escalates: OpenAI offers free GPT-4o Mini fine-tuning to counter Meta's Llama 3.1 release (VentureBeat)
YouTube tries convincing record labels to license music for AI song generator (Ars Technica)
AI lie detectors are better than humans at spotting lies (MIT Technology Review)
The developers suing over GitHub Copilot got dealt a major blow in court (The Verge)
OpenAI’s China Block to Reshape AI Scene as Big Players Like Alibaba Pounce (Bloomberg)
Always be Learnin’ 📕 📖
Inside the maths that drives AI (Nature)
Beyond LLMs: Building magic (Foundation Capital)
Truth Machines, from Llull to Boole to ChatGPT (nautil.us)
Projects to Keep an Eye On 🛠
GitHub Copilot Extensions are all you need (visualstudio.com)
tegonhq/tegon: Tegon is an open-source, AI-first alternative to Jira, Linear (Github)
mufeedvh/code2prompt: A CLI tool to convert your codebase into a single LLM prompt with source tree, prompt templating, and token counting. (Github)
The Latest in AI Research 💡
The Llama 3 Herd of Models | Research (AI at Meta)
Understanding Alignment in Multimodal LLMs: A Comprehensive Study (arxiv)
Revealing Fine-Grained Values and Opinions in Large Language Models (arxiv)
The World Outside of AI 🌎
Gen Z sours on tech jobs (qz.com)
How to Avoid Getting Sick This Summer (WIRED)
When Does Heat Make Summer Sports Intolerable? *The Atlantic)
People Are Feeling Stuck in Their Jobs. Bosses Are Starting to Worry. (WSJ)
Apple Tries to Rein In Hollywood Spending After Years of Losses (Bloomberg)
‘Recession pop’ explained: How music collides with economic trends (CNBC)
One Last Bite 😋
