EP72: Croc Test with Gemini 1.5 Experimental, Flux Destroys Midjourney & GPT4o Model Updates

Поделиться
HTML-код
  • Опубликовано: 4 фев 2025

Комментарии • 23

  • @MarcusAndersonMusic
    @MarcusAndersonMusic 6 месяцев назад +4

    This podcast is among the very best for staying up to date on the latest advancements in generative AI for both developers and the technically minded comsumer. Please keep them coming, the views will surely follow.

    • @sacariah123
      @sacariah123 6 месяцев назад +1

      One might even say it’s average. 😂

    • @chrissharkeyai
      @chrissharkeyai 6 месяцев назад

      Thank you that's nice of you to say!

  • @thetroll999e
    @thetroll999e 6 месяцев назад +4

    This episode, while otherwise average (of course), reminded me of an excellent dinosaur joke. A musem visitor asks a curator how old a particularly impressive dinosaur skeleton is thought to be. "Sixty-five million and twenty-two years", he replies. "Wow, how can you be that precise?". "Ah, well - when I started work here 22 years ago, they told me it was 65 million years old".

  • @glenwilson
    @glenwilson 6 месяцев назад

    Great info guys. Great to see you visiting in my neck of the woods (sunny coast) seeing the big crocs. You got yourself a new sub, and subbed to the podcast.

  • @justtiredthings
    @justtiredthings 6 месяцев назад +1

    AI Explained put forward a very strong hypothesis that 3.5 Sonnet's position on the leaderboard is due to its high refusal rate, not its intelligence

    • @chrissharkeyai
      @chrissharkeyai 6 месяцев назад

      I'm sure they thought it through well but this isn't my experience. I rarely get a refusal and its answers are always top quality. I also judge it, anecdotally of course, by which model I gravitate to when I have something "serious" to work on, and it is always 3.5 Sonnet. I have access to all the top models and I always choose it. Anyway I guess it doesn't really matter we should just feel lucky to have access to so many good ones.

    • @Ed-sf02
      @Ed-sf02 6 месяцев назад

      Yep, the (supposedly) high refusal rate is an hypothesis to explain why it is not higher in the ranking

    • @chrissharkeyai
      @chrissharkeyai 6 месяцев назад

      @@Ed-sf02 Sorry I understand what you're saying now!! That makes total sense

  • @remelin75
    @remelin75 6 месяцев назад

    I can see myself using voice chat if it's always there and I don't have to start it up for every question. It also needs to understand what I'm working on at the moment. So, when it's like another person sitting beside me looking at my screen, that's when I would talk to it all the time. Sam said something about just having it on all the time and that it was awesome. A companion that's always in your head. Like second conscience. I have a feeling the government loves that idéa.

  • @KolTregaskes
    @KolTregaskes 6 месяцев назад

    37:33 Midjourney apparently has no investors. Also, you can website access for all is coming soon. :-) But the threshold is very low, so won't take you long to get onto the site.

  • @tomcraver9659
    @tomcraver9659 6 месяцев назад

    Maybe the reason OpenAI is kind of all over the place, is that Sam Altman runs it internally like Y-Combinator?
    People propose stuff, the most interesting/different get some internal resources, run off and do the project, and it all gets 'thrown out there' to see what the market thinks.

  • @micbab-vg2mu
    @micbab-vg2mu 6 месяцев назад +1

    Gemini 1.5 Experimental is useful - )

  • @michaelhamilton1589
    @michaelhamilton1589 5 месяцев назад

    With image to image reference for flux. Is that use case reference associated with uploading a random image to flux and have it optimised or processed? Is that what you guys mean for image to image?
    Great show by the way!!!😊😊

  • @FriscoFatseas
    @FriscoFatseas 6 месяцев назад

    Quiet month

  • @TimKendall
    @TimKendall 6 месяцев назад

    Croc-theory

  • @baconandhex
    @baconandhex 6 месяцев назад

    You guys haven’t used googles ai studio? Seems odd. They have had a 2m context window on 1.5 for ages. This is just a more recently built version.

  • @scrollop
    @scrollop 6 месяцев назад

    I think that openai had that new voice release to Trump Googles voice release the next day - it was nowhere near release, and they lied and said it was coming out within "weeks."

  • @scrollop
    @scrollop 6 месяцев назад +2

    Lmsys leaderboard is useless now - I wonder whether google and openai somehow got questions/answers from lmsys to train their latest models on (or some other shenaningans).

  • @jdurnwald
    @jdurnwald 6 месяцев назад

    another straight down the middle average (but not mean) podcast.

  • @Jacobo99191
    @Jacobo99191 5 месяцев назад

    Why is Mike always getting sick? Get some sleep man.