Welcome to another exciting week in the world of AI. From massive launches by Meta and Apple entering the AI arena to AI's continual integration into the entertainment sector and OpenAI's challenging week, there's plenty to delve into. Let's dive right in!
AI in Entertainment: A Writers' Dilemma
The ongoing writers' strike underscores the rising concern about AI-generated content. It's inevitable—AI is soon to play a significant role in content creation. Twitter user Nathan Lands showcased some incredible examples, and this week showed the possibilities of AI video.
While the fear writers harbor is understandable, storytelling—with its emotion, suspense, and empathy—is a domain AI is yet to conquer.
The Southpark AI Recreation
An astonishing application of AI in entertainment came from Fable Studio this week. They published a white paper detailing their AI-powered recreation of Southpark using stable diffusion, ChatGPT, and Runway Gen2. Check out the paper for some exciting examples.
Apple's New AI Competitor
Moving away from Siri, Apple is developing its own AI, codenamed Ajax, to compete with the likes of OpenAI, Microsoft, and Google. Though Ajax's go-to-market strategy is still in development, its integration with the Apple ecosystem could ensure adoption by billions.
The Quality-Safety Tradeoff in ChatGPT
ChatGPT has indeed been on the decline, as you may have noticed. A fundamental tradeoff exists with large language models: the more safety measures incorporated, the lower the quality. As ChatGPT has amplified safety over recent months, Stanford and UC Berkeley researchers have found an apparent decrease in quality. Learn more in the published paper.
OpenAI's Dilemma
OpenAI's week was further complicated by The NY Times report about its delayed release of an advanced ChatGPT version capable of analyzing images. The delay was due to the AI's ability to match faces with names and personal information—potentially infringing on privacy laws. This issue was particularly concerning for the partnership between ChatGPT and Be My Eyes, an app helping visually impaired people identify objects.
The Rise of HuggingFace
AI startup HuggingFace, dubbed the GitHub for AI and ML models, is set to raise a massive $200m round at a $4b valuation, led by Ashton Kutcher’s venture capital firm, Sound Ventures. Check out the Forbes article here.
Stripe's Internal LLM
Stripe, the prominent payments API company, has built an internal LLM tool, confirming the massive potential for consultancy and services in helping companies worldwide implement secure, private, and open-source LLMs for internal use.
Petals: Democratizing LLMs
This week also saw the launch of Petals, a project promising to further democratize open source LLMs. Petals enable running large models on any computer using torrent technology. Each device stores a small piece of the model, and collectively, they could form the most powerful supercomputer in the world. Petals already support Llama 2, so don't miss the tutorial video!
Llama 2: Meta's New Release
Speaking of Llama 2, our main story is Meta's launch of this new version. Llama 2, which was used with 40% more data and brought significant quality improvements, is available in two flavors (base and chat) and three sizes (7b, 13b, and 70b). It's commercially viable as long as you have less than 700 million users. You can start experimenting with Llama2 here.
