‹ BackHN Continuity

Thread

Microsoft abandons personal AI chatbot race with Copilot reboot

157 points · 154 comments · sbulaev

  1. TuringNYC · · focus · HN ↗
    My blood pressure rises everytime i use Enterprise Copilot at work. It seems they truncate message history heavily (to save input token costs?) and Copilot forgets the last thing you told it in the chat. It is like a real-life version of Momento (<a href="https:&#x2F;&#x2F;www.imdb.com&#x2F;title&#x2F;tt0209144&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.imdb.com&#x2F;title&#x2F;tt0209144&#x2F;) mixed with Waiting for Gadot (<a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Waiting_for_Godot" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Waiting_for_Godot), pretending to be in league with Anthropic.
    1. browningstreet · · focus · HN ↗
      Doing research work with it that outputs any kind of artifact is.. really painful. It constantly crashes, has to regenerate.. and then the output is often very difficult to carry to other Microsoft products, like Sharepoint. I end up creating PDFs that I import.

      I&#x27;ll go ahead and burn working hours on it because I&#x27;m getting paid, and people like the things I create.. but it&#x27;s a very crippled and painful version of what you get with the frontier products and most OSS options.

      1. poly2it · · focus · HN ↗
        It boggles my mind that Microsoft can fail to deliver even bare minimum working software so consistently, with all the engineering talent they house. What are the product teams doing?
        1. TuringNYC · · focus · HN ↗
          I feel it is even worse than bare minimum -- it is so bad it infuriates the user. It is hard to get away from that reaction when one uses better alternatives in their personal life.
        2. nsbdhciebsbdhc · · focus · HN ↗
          Vibecoding.
        3. discordance · · focus · HN ↗
          incredible right? - not to mention the complete access they have to OpenAI&#x27;s IP
        4. Droobfest · · focus · HN ↗
          It boggles my mind that it doesn&#x27;t seem to effect their bottom line at all
      2. homeonthemtn · · focus · HN ↗
        Oh it&#x27;s so bad. So so bad.

        If you prompted an AI to return answers &quot;as though you habitually ate soap and crayons while mouth breathing&quot; you&#x27;d still get a better response than copilot

        1. robotnikman · · focus · HN ↗
          Its so bad you would probably get better answers using a 2b parameter local model
        2. epihelix · · focus · HN ↗
          That used to be the case, but the current Enterprise version, with the poorly-named &quot;GPT 5.6 think deeper&quot; option, is very decent indeed. Can&#x27;t tell if it&#x27;s Sol, Terra or a mixture of the two, but it&#x27;s certainly not Luna.

          (Stick with the default &quot;Auto&quot; and you probably get Phi-4 -- it&#x27;s lobotomised AF.)

          I use Copilot a lot now. Unlimited, free, high-quality, enterprise-protected inference is not to be sniffed at. It also writes some of the nicest code I&#x27;ve ever seen from an LLM, that&#x27;s almost entirely free of comments.

          1. rmaxdev · · focus · HN ↗
            My previous company used copilot enterprise

            My experience was great because I had premium license and only used thinking models

            Most colleagues didn’t have premium and used default&#x2F;auto model

            People stick to their very first bad experience unfortunately

          2. Jenk · · focus · HN ↗
            Are you referring to GitHub Copilot? I believe this thread is referring to Windows Copilot, such as the AI feature embedded into every Office progam, which is not the same experience.
            1. epihelix · · focus · HN ↗
              No, good ol&#x27; bog-standard Enterprise M365 Copilot. The one with the worst chat interface in history. That one.

              I don&#x27;t think many people are aware of just how much it can do; it&#x27;s certainly not advertised, and very likely not intended.

              1. sterlind · · focus · HN ↗
                could you give some examples?
                1. epihelix · · focus · HN ↗
                  Try sending this prompt to &quot;GPT 5.6 Think Deeper&quot; in the interface:

                  Provide me with optimised python code to perform a prime number search. Test the code and report on performance. Benchmark different search solutions. Provide the code in a downloadable zip archive.

                  A silly toy example, but it should help give a hint of what&#x27;s possible.

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.