- The Renaissance Times
- Posts
- YouTube vs. OpenAI
YouTube vs. OpenAI
The Renaissance Times
April 05, 2024
Dear Artisan,
OpenAI certainly lives by the motto “all press is good press.”
A few weeks ago their CTO Mira Murati gave an interview and was pressed on the training data for OpenAI’s upcoming video model, Sora. She was coy to say the least, and left viewers with an uncomfortable taste in their mouth as to how ethical or legal the company was acting.
Today, YouTube confirmed that if indeed OpenAI is using YouTube videos to train its model, it will be violating YouTube’s rights and could be pursued in court.
Let’s dive in.
Hot Off The Press
YouTube Says OpenAI Training Sora With Its Videos Would Break Rules
NYC’s chatbot is advising businesses to break the law
Israel used AI to identify bombing targets in Gaza
Grok is detailing users how to make bombs, concoct drugs
AI-generated Asians were briefly unavailable on Instagram
The One Big Thing
One factor that gets overlooked in much of the AI discussion is the importance of data. We talk a lot about amazing and fancy new models, about the latest and greatest in hardware, but very little about data.
In the end, the greatest models and chips mean nothing if they do not have appropriate data to be trained on and people to oversee the training and provide feedback.
For most foundation models, we have no idea what data they have been trained on. This makes them a black box.
Many of the things we may take to be a model “reasoning” about could be pieces of data it has seen in the past. This includes ChatGPT and most every other model we use today. And it will apply to their next model, Sora.
Sora is a text to video model, and the results are gorgeous. We have written about them many times previously. Yet to train a model to create beautiful video ouputs you need a plethora of video. And the place with the most video is YouTube.
Now, as you can imagine, YouTube / Alphabet will not be very happy to find out a competitor is about to use all of their data (despite being publicly available) to train their models for no cost. And they have stated as such, claiming it would be a violation of their rights.
How exactly they might prove this to be the case remains to be seen, but this is not the end we have heard of this battle.
The Gallery
Let's talk AI Music.
People are creating some amazing AI music and they are insanely good.
10 examples: twitter.com/i/web/status/1…
— Min Choi (@minchoi)
3:23 PM • Apr 4, 2024
Tools
Must have tools for every artisan to add to their toolkit:
Audialab Emergent Drums: A tool to create royalty-free drum samples
Legal Graph: A tool to visualize connections between legal concepts, cases, statutes, and regulations
BannerGate: A tool to create animated HTML5 display banners for advertising.
WhatColors: An app for color season analysis, nail color recommendations based on your skin tones
SWE-agent: A tool to autonomously resolve bugs in GitHub repositories
Deep Tech
The newest and coolest in the research world that you need to know about:
Google announces Training LLMs over Neurally Compressed Text
MiniGPT4-Video: Advancing Multimodal LLMs for Video Understanding with Interleaved Visual-Textual Tokens
ReFT: Representation Finetuning for Language Models: 10x-50x more parameter-efficient than prior state-of-the-art parameter-efficient fine-tuning methods
AutoWebGLM: Bootstrap And Reinforce A Large Language Model-based Web Navigating Agent
PointInfinity: Resolution-Invariant Point Diffusion Models
Closing Thought
IP rights call for a new incentive structure…
Work With Us!
The AI Renaissance is coming and we are building the best community of the people making it happen.
Contact us to sponsor your product or brand and reach the exact audience for your needs across our newsletter and podcast network.