- The Daily Update
- Posts
- Tech Giants accused of scraping YouTube data
Tech Giants accused of scraping YouTube data
Spotlight: Nvidia researchers develop efficient fine-tuning method
Jack Shields
July 17, 2024
Welcome back!
Some major tech companies could be in trouble for scraping data from thousands of YouTube videos. We’ll also check out a huge breakthrough at Nvidia. Let’s dive in.
In today’s Daily Update:
🗞️ Tech Giants allegedly used thousands of YouTube videos to train AI
🤖 Anthropic launches Claude Android app
📸 Nvidia researchers develop more efficient way to fine-tune AI models
🚨 AI Roundup: Four quick hits
Read time: 2 minutes
TOP STORY
🗞️ Tech Giants allegedly used thousands of YouTube videos to train AI
DALL-E 3
Several tech companies including Apple, Anthropic, Nvidia and Salesforce allegedly used over 170,000 YouTube videos to train AI systems.
What you should know:
According to an investigation by Proof, YouTube subtitles from more than 48,000 channels were used to create a massive training dataset.
YouTube CEO Neal Mohan says the use of any content to train AI violates the platform’s terms of service.
The subtitles dataset includes content from popular creators like MrBeast and Marques Brownlee, and major news outlets like ABC News, the BBC and The New York Times.
Alongside its investigation, Proof published the tool it used to search for creators in the YouTube AI training dataset.
The big picture: YouTube hosts over 800 million videos on its platform and is widely considered a gold mine for AI training data. This investigation is interesting because the subtitles dataset is part of a larger collection of training material published by EleutherAI, meaning the companies that used the data technically avoid fault. Regardless, tech companies’ data scraping practices are facing growing criticism and potential legal repercussions.
AI TOOL OF THE DAY
🤖 Anthropic launches Claude Android app
Source: Anthropic
Anthropic just released its Claude Android app in an effort to convince users to abandon ChatGPT by expanding Claude’s availability.
Key points:
The Claude Android app includes free access to Claude 3.5 Sonnet, Anthropic’s most advanced AI model.
Users will be able to sync conversations with Claude across devices, and upload photos or files to the app.
The app also offers real-time language translation, a feature currently missing from ChatGPT’s mobile app.
Power struggle: Anthropic is overachieving with less funding and a smaller team than OpenAI. Claude 3.5 Sonnet actually outperforms OpenAI’s GPT-4o across numerous benchmarks in reasoning, coding and math. Despite this, the company has struggled to attract new users. The Claude Android app could help turn the tide.
RESEARCH SPOTLIGHT
📸 Nvidia researchers develop more efficient way to fine-tune AI models
DALL-E 3
Nvidia researchers have introduced a method for efficiently fine-tuning AI models for specific tasks called “DoRA.”
The details:
Low-Rank Adaptation (LoRA) is a widely popular method for fine-tuning language models that adjusts the magnitude and direction of pre-trained models proportionally.
Weight-Decomposed Low-Rank Adaption (DoRA) expands on this approach by optimizing magnitude and direction individually.
This allows DoRA to make more precise adjustments during the post-training fine-tuning process.
DoRA achieves higher accuracy than LoRA without increasing computational costs.
What this means: Think of the fine-tuning process as adjusting the volume on a stereo system. A LoRA stereo would have a few simple knobs that are quick and easy to use, but can’t make very precise adjustments to individual settings like bass or treble. DoRA is an upgrade to the stereo controls that allows for increased control and more detailed adjustments. This means you can get closer to the exact sound you want. In AI, this means general-purpose models can be more closely fine-tuned to complete specific tasks.
MORE TRENDING NEWS
🚨 AI Roundup: Four quick hits
DALL-E 3
Former OpenAI engineer starts Eureka Labs, an AI-integrated education platform.
Disney hackers say concerns about the company’s approach to AI motivated their attack.
French competition authority confirms an ongoing investigation into possible anti-competitive practices by Nvidia.
Microsoft facesUK antitrust probe after hiring the core team behind Inflection AI.
THAT’S A WRAP