I’m Hunter Perrin. I’m a software engineer.

I wrote an email service: https://port87.com

I write free software: https://github.com/sciactive

  • 7 Posts
  • 336 Comments
Joined 1 year ago
cake
Cake day: June 14th, 2023

help-circle







  • If it’s just hours, that’s fine. I’ve spent months on a system before that ultimately got scrapped. When I was at Google, they accidentally had two teams working on basically the same project. The other team, with about 40 engineers, having worked on it for about a year, had their project scrapped. My team was meant to do the same work, with about 23 engineers. So if you’re ever wondering why Hangouts Chat launched kinda half baked, that’s why.




  • I feel like if ChatGPT were the only LLM on the market, they’d have a real path to profitability, but it’s not even the best LLM on the market. And the open source models are nearly as good, meaning the vast majority of people who need an LLM can run it on their own hardware.

    It’s kind of like trying to make a profitable business out of offering a special sauce that isn’t as good as your competitors sauce, and is barely better than the free sauce from Taco Bell. Oh and it costs you millions of dollars to produce a single bottle.


  • If you’re transferring files over a socket (like through SMB or SFTP), the receiving end usually has a small buffer, like 64KB. It’ll just pause the stream if it’s receiving data faster than it can push it to disk and the buffer gets full. So usually a file transfer won’t use much memory.

    There is some poorly written software that doesn’t do that, though. I ran into a WebDAV server that didn’t do that when I was writing my own server. That’s where you could run into out of memory errors.