God Emperor of Mastodon (new account)

@michal@sapka.pl

This is my new, future account. While most interaction still happen on the bsd.cafe one, I aim to migrate fully here.

I am posting mostly culture, emacs, bsd and anti-ai based ramblings. I don't curse and I try not bait people into pointless anger rages.

I am a nerd (or dork? I don't know), a software engineer and I started to hate what technology has bacame.

This is a self-hosted, single-user instance running on snac2.
Messages auto deleted every few weeks
Homepage: ✔ https://michal.sapka.pl

341 following, 908 followers

0 ★ 1 ↺
AI boosted

[?]God Emperor of Mastodon (new account) » 🌐
@michal@sapka.pl

It seems Jev AI omits most of the criticisms of LLMs? No theft, no hallucinations, much better energy usage, not Altman?


    ...

    [?]Peter O'Shaughnessy [he/him] » 🌐
    @peter@toot.cafe

    @michal

    Re. No theft: They must still be trained on vast quantities of text, in order to 'understand' the input, even if they output JSON rather than human language. I doubt their training corpus is much more ethical than those used for LLMs?

    Re. No hallucinations: They could still hallucinate things or be influenced in weird ways from things in the input, affecting the decision output. And then you're left with just black box decisions, without any record of why they were reached.

      ...
      1 ★ 0 ↺

      [?]God Emperor of Mastodon (new account) » 🌐
      @michal@sapka.pl

      I'm trying to understand it, and it seems that TypeSafe AI trained the Jev foundation model on synthetically generated data?
      Almeida says Jev is trained exclusively on synthetic data using a technique he calls “reinforcement learning from calibrated decisions.”
      https://techcrunch.com/2026/09/18/a-new-kind-of-ai-model-from-a-chatgpt-inventor-is-thrilling-developers/

      ...

      [?]Peter O'Shaughnessy [he/him] » 🌐
      @peter@toot.cafe

      @michal Oh interesting, I didn't realise that thanks.

      There's also other benefits like no chatbot psychosis (as it doesn't talk to users). And a lot safer due to type safety and lack of actual agency...

      I'm trying to wrestle with it all too.

        ...
        0 ★ 0 ↺

        [?]God Emperor of Mastodon (new account) » 🌐
        @michal@sapka.pl

        Yes. It seems to be the ChatGPT we were supposed to have before it all became a nightmare - it's a decision AI, not a generative one.

          History

          about this site - powered by snac/2.89