🥟 Chao-Down #208 OpenAI's huge dev day, YouTube launches "YouChat" an AI chatbot for videos, XBox partners with Inworld to build AI tools for game development

It seems like the moniker you hear on Twitter and Reddit after OpenAI makes an announcement is that OpenAI just killed thousands of AI startups. While extreme in its proclamation, this time they might just be right.

In addition to launching cheaper and more capable models (i.e. GPT 4-Turbo, Whisper V3, Dall-E 3), OpenAI really took to heart the many pain points that enterprises have criticized them for. The company announced that developers now have the ability to create custom models (not just finetunes) built from scratch. For enterprises, the new Copyright Shield would defend business customers from copyright claims when using OpenAI’s generative models.

By offering all these things in addition to enhanced capabilities like their Assistants API (which enables developers to build autonomous agents with tools), it’s clear what OpenAI’s endgame is: to own the end-to-end developer experience for building on top of the best foundation models.

It’ll be interesting to see how other closed-source model companies and the major cloud providers respond. Expect more money to be poured into the space.

-Alex, your resident Chaos Coordinator.

What happened in AI? 📰

New models and developer products announced at DevDay (OpenAI)

Chatbots May ‘Hallucinate’ More Often Than Many Realize (The New York Times)

YouTube Launches 'YouChat' Test of AI Chatbot for Videos (Variety)

Under Fire Over Robotaxi Safety, GM Halts Production Of Cruise Driverless Van (Forbes)

Xbox partners with Inworld AI to build AI tools for game development (VentureBeat)

There's no US federal law against sharing AI-generated nudes (Quartz)

Always be Learnin’ 📕 📖

Building RAG-based LLM Applications for Production (anyscale.com)

LLMs Aren’t Ready for Prime Time. Fixing Them Will Be Hard. (stanford.edu)

A Guide to Public Speaking for Software Engineers (Substack)

Projects to Keep an Eye On 🛠

huggingface/distil-whisper: Distilled variant of Whisper for speech recognition. 6x faster, 50% smaller, within 1% word error rate. (Github)

langchain/templates: LangChain Templates are the easiest and fastest way to build a production-ready LLM application. (Github)

marimo-team/marimo: A next-generation Python notebook: explore data, build tools, deploy apps! (Github)

The Latest in AI Research 💡

Jailbreaking Black Box Large Language Models in Twenty Queries (jailbreaking-llms.github.io)

ControlLLM: Augment Language Models with Tools by Searching on Graphs (arxiv)

Learning From Mistakes Makes LLM Better Reasoner (arxiv)

The World Outside of AI 🌎

Hold the olive oil! Prices of some basic European foodstuffs keep skyrocketing (Quartz)

Did the Carpal Tunnel Epidemic Ever Really End? (The Atlantic)

Is the Music Business Losing Money to Sped-Up Song Remixes? (Billboard)

Google’s most lucrative queries: a rare look at how search makes money (The Verge)

Stress of Covid Pandemic Lingers in Americans' Minds (Bloomberg)

One Last Bite 😋