Companies train models with YouTube videos, new mini-model from OpenAI — Weekly AI Roundup

AI Video & Visuals


Proof News reports that several major AI companies, including Nvidia, Anthropic, and Apple, are reportedly training their models using subtitles from over 173,000 YouTube videos across 48,000 channels. The practice, which violates YouTube's terms of service, has sparked outrage from affected content creators, including major news outlets and popular YouTube channels.

Predictably, channel owners are angry, and rightly so.

Many creators invest significant time and effort into producing content only to find their work being reused without permission for AI development. The dataset in question was compiled by EleutherAI, a non-profit AI lab focused on model interpretability and tuning.

For more information, read Proof News' full report, available for free here.

In an encouraging development, Representative Jennifer Wexton from Virginia's 10th Congressional District has regained her voice through AI technology.

Wexton, who was diagnosed with progressive supranuclear palsy and has severely reduced speech and motor skills, now uses an AI-generated voice model based on his past speech.

This advancement represents a major step forward in helping individuals facing similar challenges, going beyond traditional text-to-speech solutions.

OpenAI has announced GPT-4o mini, a more compact and cost-effective version of the GPT-4o model. Designed to improve accessibility for both developers and enterprises, the mini model is more affordable and efficient. Starting July 18, it will replace GPT-3.5 for all ChatGPT Free, Plus, and Team users, with enterprise access expected to launch next week.

GPT-4o mini is priced at 15 cents per million input tokens and 60 cents per million output tokens, a 60% cost reduction compared to its predecessor, GPT-3.5 Turbo. With a context window of 128K tokens and supporting up to 16K output tokens per request, this model achieves an impressive score of 82% on the MMLU benchmark, outperforming competitors in the areas of text intelligence and inference.

AI continues to evolve, bringing new possibilities and advancements, so stay tuned for more updates and breakthroughs in this field.

“Beyond Tomorrow” is a weekly newsletter published every Saturday that rounds up everything AI-related from the past week.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *