Daily Digest: YouTube heist, Apple stole your videos.

Want to get in front of 100k AI enthusiasts? Work with us here

Hello folks, here’s what we have today;

PICKS

1. New tutorials just published:

   1. Create a custom image generation model - Learn how to train a personal AI image model on your artwork with Exactly AI.

   2. Create custom graphics for websites and presentations - Learn how to use Recraft to turn your brand colors into custom graphics for your next project.

2. Big tech firms secretly used YouTube videos to train AI. An investigation from Proof News found that major AI companies like Apple, Nvidia, Salesforce and Anthropic used subtitles from over 170,000 YouTube videos to train their AI models—without creators' knowledge or consent.🍿Our Summary (also below)

3. Mistral has released two new baby monsters:Codestral Mamba and Mathstral. Both have 7B parameters and open weights. Codestral Mamba experiments with the Mamba architecture where Mathstral is specifically fine-tuned for mathematical reasoning.

4. Andrej Karpathy is starting an AI education company - Eureka Labs. Karpathy says that Eureka Labs will be an AI native school and his work at Tesla and OpenAI were side-quests for his full-time job/passion—teaching. The first product, LLM101n, promises to be an undergraduate-level course, teaching students to build their own AI assistant.

5. Zapier believes computers should do more work for humans. So they distilled their AI products into a new digital energy drink: ZAP, AI in a can. Just crack one open to bestow the power of AI automation to your computer, so you can be unstoppable. Claim your free sample today! (sponsor)

TOP TOOLS

• Liminal* - Workflow productivity tools provide secure access to gen AI anywhere work gets done. Enterprise-grade data security & governance.

• Anthropic’s Claude is now available as an Android app.

• Claude Engineer 2.0 - Use Claude in your CLI with a code editor, execution agents, and dynamic editing.

• Morphic Studio - An AI-powered platform to create controlled videos.

• Mem0 - The memory layer for LLMs and AI agents.

• Buildbox - Just type to create, design and build games with AI.

• Ssemble - Reassemble a long video into engaging shorts.

• Sketch2scheme - Turn your diagram sketches into digital schemes.

View more →
*sponsored

NEWS

• Applications are now open forbatch 4 of AI Grant (due on August 9th).

• Gemini, please analyze my poem - One powerful way I, as a poet, have embraced AI.

• Google announces Project Oscar for managing open-source projects at I/O Connect Bangalore.

• Using Claude 3.5 Sonnet and Supabase to build an e-commerce app with AI-powered related-items search.

• AI is not the sole reason fordata centre emissions - Jeff Dean.

• FTC is asking for details on the Amazon-Adept deal, but informally.

• Reflection AI’s Misha Laskin on the AlphaGo Moment for LLMs.

• Exa raises $22M to build the search engine for AI.

• SF Compute has raised $12M to build a trading platform for large-scale GPU clusters.

• Brev.dev has been acquired by Nvidia.

• Vectara raises $25M Series A for enterprise RAG.

View more →

QUICK BITES

Turns out, AI giants have been sneakily snagging YouTube vids to teach their bots. Creators are not happy, and the legal waters are murky.

What's going on here?

An investigation found that major AI companies like Apple, Nvidia, and Anthropic used subtitles from over 170,000 YouTube videos to train their AI models - without creators' knowledge or consent.

What does this mean?

Big names in tech (we're talking Apple, Nvidia, Anthropic) used a dataset called "YouTube Subtitles" to train AI models. This dataset includes content from 48,000+ channels, including major YouTubers, educational channels, and even some conspiracy theory videos.

Many creators had no idea their content was being used this way. Some are calling it theft. The dataset was part of a larger compilation created by Eleuther AI called "The Pile," which also includes stuff like Wikipedia articles and Enron emails.

YouTube's terms of service prohibit this kind of automated data scraping, but companies argue it's different when using a pre-compiled dataset. You can see if your videos were in the dataset using Proof News tool.

Why should I care?

This is a big deal for content creators and the future of AI. It raises questions about consent, compensation, and the potential for AI to replicate (or replace) human-created content. Plus, it shows how even deleted videos can live on in AI training data. As AI gets smarter, the debate over using creators' work without permission is only going to heat up.

Share this story

Ben’s Bites Insights

We have 2 databases that are updated daily which you can access by sharing Ben’s Bites using the link below;

• All 10k+ links we’ve covered, easily filterable (1 referral)

• 6k+ AI company funding rounds from Jan 2022, including investors, amounts, stage etc (3 referrals)