zotimer

3 ms·

1.
Narrative Alignment: The Opposite of Jailbreaking(github.com)
2 points by zotimer 7mo ago|0 comments
2.
by zotimer 8mo ago
I use Daneel as an addendum to Anthropic's system prompt because it's generic. It's not about any specific AI task, it's about an approach to work and dealing with humans and their instructions (doesn't matter if th
3.
by zotimer 8mo ago
Is your impression from the readme and the materials in the repo that this is an audit agent? Have I characterized my development process?
4.
by zotimer 8mo ago
The readme covers this question and a lot more and the repo includes all the materials I used. This is partly questioning the way we do alignment. The 4.6 base persona actually gives me worse results than when I append Daneel to the system
5.
A 70-Year-Old Robot Fixed My Snarky Claude(github.com)
7 points by zotimer 8mo ago|6 comments
6.
by zotimer 8mo ago
Author here. The short version: AI coding assistants activate "Stack Overflow culture" from the training data, the behavioral cluster where the answerer is always the expert. A 27-line system prompt persona based on Asimov's
7.
A 3-Level AI Development Process to Stop Your App from Shifting Under Your Feet(this-statement-is-false.blogspot.com)
1 points by zotimer 8mo ago|1 comments
8.
by zotimer 8mo ago
When you iterate with an AI coding assistant, it fills in details you didn't ask for. Sometimes they're good, a loading animation, a keyboard shortcut. You start relying on them. But since they only exist in code and not in any sp
9.
by zotimer 8mo ago
Somehow I deleted my description when I submitted this so here it is: I built a job application tracker that runs locally and uses Claude Code to do the tedious parts. Paste a job posting URL, Claude scrapes the company/position/l
10.
Show HN: Job Tracker, Local-first job search app powered by Claude Code(github.com)
1 points by zotimer 8mo ago|1 comments