Thread

1. barney54 · 1790384411 · focus · HN ↗

   Does anyone have any good experiences with copilot? I have never heard anyone say anything good about it.

   1. epihelix · 1790406387 · focus · HN ↗

      Yes, with Enterprise M365 Copilot.

      Select "GPT 5.6 Think deeper" (ugh, so unrevealing, MS) and you've got free, unlimited Sol or Terra (likely a mix depending on load) on tap.

      It writes very nice, clean code with almost no comments.

      I use it a lot.

      1. zerreh50 · 1790434478 · focus · HN ↗

         For me it regularly gives garbled code with random *s and ;s everywhere. I don't get how supposedly frontier models can be made so useless

         1. TurboSkyline · 1790455931 · focus · HN ↗

            This has happened to me, but only a couple of times. I think it’s an issue with the UI, not with the actual LLM output, and a reload of the page fixes it.

            I think it’s related to how they display the LLM output, through some obfuscation layer in the HTML presented to the browser. I tried to save a thread as pdf once and that got me nowhere because of this.

         2. epihelix · 1790484895 · focus · HN ↗

            Yes, it's an absolutely terrible UI, probably the worst ever coupled to an LLM. I have no idea why those *'s pepper the code blocks (but it's not coming from the model: reload, and you'll generally see the codeblock render correctly). The interface is bloated, slow, fugly and massively over-engineered. That's the price of entry.

            But you don't need to get code from the chat window -- that's so 2024. There is an entire agentic coding remote sandbox built into the Enterprise M365 version. It's not talked about, or advertised, but it's there. Just ask for your code changes back in a zip.

            It turns out almost anything is possible with Copilot, if you spend a bit of time thinking it through and exploring its capabilities. It is a surprisingly powerful platform.

Unofficial Hacker News client; not affiliated with Y Combinator.