Tesla’s Remarkably Bad Quarter Is Even Worse Than It Looks
www.theatlantic.com
Reality is catching up with Elon Musk.

Yesterday evening, Tesla reported first-quarter earnings for 2025, and they were abysmal: Profits dropped 71% from the same time last year.

Musk sounded bitter on the call with investors that followed, blaming the company’s misfortune on protesters who have raged at Tesla dealerships around the world over his role running DOGE and his ardent support of far-right politicians.

“The protests that you’ll see out there, they’re very organized. They’re paid for,” he said, without evidence.

Non-paywall link

  • gnutrino@programming.devEnglish
    0·
    2 months ago

    However, EMMA also exhibits certain limitations: it can process only a small amount of image frames, does not incorporate accurate 3D sensing modalities like LiDAR or radar and is computationally expensive.

    And found it wanting compared to systems that incorporate LiDAR

    • NotMyOldRedditName@lemmy.world
      0·
      2 months ago

      That’s not really saying its wanting, it’s just saying it’s limitations.

      The model they’ve created can’t process those extra things. It still performed very well in their early testing.

      Edit: on mobile now (was on desktop) but the article linking to the source said it was out performing other systems, but they didn’t specifically test it against the waymo production system. It also said it would be difficult to feed other sources into it as it’s more complicated than images.

    • NotMyOldRedditName@lemmy.world
      0·
      2 months ago

      I was just looking through the linked PDF and I’ll actually retract what I said earlier about not finding it wanting, and them just listing limitations.

      In the research paper linked it is more explicit

      “Despite these promising results, EMMA is not without its limitations. In particular, it faces challenges for real-world deployment due to: (1) limitations in 3D spatial reasoning due to its inability to fuse camera inputs with LiDAR or radar, (2) the need for realistic and computationally expensive sensor simulation to power its closed-loop evaluation, and (3) the increased computational requirements relative to conventional models. We plan to better understand and address such challenges in future work.”

      This is more clear that it’s not just listing limitations, but it also finds it not doing as well because of it.

      That I’d call wanting.