AI News: ChatGPT “Her”, Llama 3.2, Meta's New AR Glasses, and more OpenAI Departures



Here’s all the AI news you missed over the last few days!

Join My Newsletter for Regular AI Updates 👇🏼
https://www.matthewberman.com

My Links 🔗
👉🏻 Main Channel: https://www.youtube.com/@matthew_berman
👉🏻 Clips Channel: https://www.youtube.com/@MatthewBermanClips
👉🏻 Twitter: https://twitter.com/matthewberman
👉🏻 Discord: https://discord.gg/xxysSXBxFW
👉🏻 Patreon: https://patreon.com/MatthewBerman
👉🏻 Instagram: https://www.instagram.com/matthewberman_ai
👉🏻 Threads: https://www.threads.net/@matthewberman_ai
👉🏻 LinkedIn: https://www.linkedin.com/company/forward-future-ai

Media/Sponsorship Inquiries ✅
https://bit.ly/44TC45V

source

33 thoughts on “AI News: ChatGPT “Her”, Llama 3.2, Meta's New AR Glasses, and more OpenAI Departures”

  1. I know Moshi Voice-to-Voice Chat is a simple and small model with lots of problems, but I find it unfortunate that when it was demoed, everyone dumped on it harshly, especially for a fledgling, truly open source voice model.

    Everyone was even more hyperfixated on the "Open" AI demo at that time that wasn't even released. I'm not nearly as impressed with the reality of it. It's not nearly the image they built up. Buggy, laggy, jittery, no video or stills, interruptions don't hold over long enough, lots of refusal for things like singing, voices feel less natural, etc.

    I personally think the gap in the voice-to-voice experience isn't that big. The model supporting Moshi is really dumb, sure, but it's also actually open source and they were aiming on trying to get it up and running locally on consumer grade hardware in months from scratch. There has to be more potential in this project.

    To me, it's a marval that a tiny not-for-profit research lab's team pulled it together so fast. A lot of people are talking about how we need to support open source, so here we have a chance to keep Moshi or the possibility of other open source models in our awareness.

    I'm following this project, and interested to learn of any open source projects that are working on voice-to-voice.

    Bearing in mind that Meta's voice mode source code can likely be reviewed for research purposes, but may have strict restrictions on redistribution.

    The Lab for Moshi:
    https://kyutai.org/

    [Source Code (GitHub)](https://github.com/kyutai-labs/moshi)

    This guy followed up with a good overview when the source code got released: [YouTube](https://youtu.be/JKA_v5Bb_tI?si=MgfAYMA8QFLOSekC)

    Reply
  2. Frontier model company valuations will look silly in a few short years when they're commoditized by open source.

    The real value is, as Satya said, in how you leverage LLM's as tools and not in building LLM's.

    Reply
  3. Anyone who works with these models every day, will be very reluctant to claim that openai is so "ahead" of Anthropic. I still prefer claude for most of my work. Some times I even get better results than o1

    Reply

Leave a Comment