Comments URL: https://news.ycombinator.com/item?id=48590573

Points: 3

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=48539467

Points: 6

# Comments: 6

Comments URL: https://news.ycombinator.com/item?id=48526331

Points: 1

# Comments: 0

As a small note of color, when I was a teenager I helped the local police department clean up one of their PCs, which had been infected with multiple viruses, Michelangelo is the one I still remember, though there were others. After cleaning the machine, I installed Prince of Persia for them. The policemen were absolutely thrilled to have that video game on their computer.

I do not understand what the problem is. There are both closed and open models. You can run your own machine with dozens of open models. You can train your own model. You can do everything on your own.

Of course, there are limitations. For example, you cannot magically have all the best hardware at your disposal, but that limitation also exists in normal programming.

Comments URL: https://news.ycombinator.com/item?id=48308123

Points: 2

# Comments: 0

I wrote an email saying I would not pursue the position, and they wrote back asking me to have another interview with them. I politely declined.

They probably understood that their method was not good.

Who cares if you had trauma when you were 16? Will a past trauma affect your future at the company? Does the interviewer have a psychology degree to conduct such an interview?

In any case, do people have the right to a second chance if they did something morally questionable in the past? I've conducted over 2,000 interviews in the past 20 years, and I've learned a lot. The best indicators of a good candidate are not questions like "Tell me your weaknesses" or "Tell me about a mistake you made."

The best indicators are whether the person spent time learning about your project, your company, the people who work there, the technologies, the product, the vision, the financial status, and the investors. That shows more interest than answering "Tell me about your hobbies."

- I plan everything in ChatGPT. I provide project trees and create an execution plan for Claude Code, Codex, or Copilot agents. - I clearly say I do not want the agent to explain what it will do or what it did. I only want "I finished". That saves extra tokens. My projects include tools to gather file contents in bulk. A single command like bun copy-paste FILES=" " lets the agent collect many files into one file. - I also provide a bun create-empty-files command so the agent can create all needed files at once instead of one by one.

With this setup, I managed to build 4 major features (more than 80 file edits or creations) using the 3 daily resets on a standard plan.