Discover more from Artificial Ignorance
AI Roundup 034: Meta Connect
September 29, 2023
Meta's Connect event was this week - a conference originally about VR - and the company showcased a whole bunch of AI announcements.
All the ones I could find:
AI-generated stickers across Meta's apps, plus AI-editing tools for Instagram.
T-Pain as Mark Zuckerberg's Jarvis AI assistant.
A look at the Llama ecosystem's past, present, and future.
Also, they're only mildly AI-related, but these Smart Glasses are pretty cool.
Why it matters:
AI may end up being Meta's saving grace; generative AI stands to make the Metaverse workable with generated worlds and games.
Beyond that, Meta is chipping away at OpenAI's dominance with its Llama 2 models. It will have help - Mistral AI released a new open-source LLM that supposedly outperforms Llama 2.
But it remains to be seen whether Meta's consumer products will find product-market fit or not. Despite being an early adopter, Snap's AI chatbot was only used by about 20% of its MAUs as of June.
Elsewhere in the FAANG-free-for-all:
Google lets webmasters opt of AI training with a robots.txt flag.
Microsoft is working on smaller, cheaper AI models for features like Bing Chat.
And Amazon plans to invest up to $4 billion in Anthropic at an undetermined valuation.
Despite not organizing a conference, OpenAI had its own slew of updates this week.
Spotify is partnering with OpenAI to translate podcasts in other languages (while keeping the host voices the same).
And Sam Altman may be in talks with Jony Ive and Masayoshi Son to build "the iPhone of AI."
Between the lines:
As cool as GPT-Vision is, the related paper reveals some potential biases and flaws in the model (though OpenAI is still working hard on safety).
Google Gemini has been generating plenty of hype recently, with reports that it will be multi-modal and a true GPT-4 competitor. OpenAI may be trying to take the wind out of Google’s sails.
And the company is reportedly looking at a private share sale at a valuation of over $80 billion.
People are worried about AI content
After months of negotiations, the Writer's Guild of America reached a deal with Hollywood studios, though the actors are still on strike. The new contracts state (among other things) that AI can't write or rewrite literary material.
Why it matters:
The WGA has struck a landmark deal, and it may serve as a template for performers, voice actors, and others.
The deal also allows for potentially using writers' work to train models, with guarantees for credit and compensation.
And other writers are following suit - Medium blocked AI companies from using its content until it can get similar concessions.
Elsewhere in AI anxiety:
Dozens of VC firms are working with the US Department of Commerce on "responsible AI" guidelines.
Google was accidentally leaking its Bard chats into public search results.
And despite a lawsuit against AI-image generators, Getty Images has launched its own AI-image generator, trained only on its legally licensed photos.
Prompt engineering for Claude's huge context window. Palantir wins a $250M contract for military AI services. Cloudflare's new tools to build and deploy AI models. AI-generated subliminal messages are going viral. French billionaire plans to build an Nvidia cloud supercomputer. The AI assistant revolution is over 50 years in the making. Bumble CEO on how AI will "supercharge" love. A look at the "Surveillance AI Pipeline". Microsoft's multi-agent conversation framework. AI to automate scientific literature review. Prisoners are training AI. Workday's new HR AI features. Google is using AI-answers for top search results. LLMs may outperform existing lossless compression formats. An LLM agent paper list.
Artificial Ignorance is reader-supported. If you found this interesting or insightful, consider becoming a free or paid subscriber.