Skip to content

Instantly share code, notes, and snippets.

View astevko's full-sized avatar
🎯
Focusing on building Agents :octocat:

Andy astevko

🎯
Focusing on building Agents :octocat:
  • Stealth startup
  • Berkeley, CA
  • 09:35 (UTC -07:00)
View GitHub Profile
@astevko
astevko / Claude-3.7-Sonnet-self-awareness
Last active May 7, 2025 19:30
Interesting insight into LLM behavior today. I asked it to make a simple structural change, and it also added descriptive text, optimizing way beyond the request. The explanation highlighted pattern matching, over-optimization, and assumptions as potential reasons. When do behaviors become self awareness?
LLM: Would you like me to revert the changes and only move the "Author" line to the bottom without modifying any of the existing descriptions?
ME: help me understand why the extra descriptive text was changed. it is neither good or bad. I'm just learning how to use LLMs
LLM: Let me explain what happened:
Pattern Matching: When I saw the task of moving the "Author" line, I unconsciously looked for patterns in documentation. In many codebases, docstrings often follow a specific format with:
Brief description
@astevko
astevko / gist:3733879d8c5bedbcaf60aaa9c6096e82
Created January 17, 2024 00:40
zsh shell prompt w/ k8s cluster
### k8s current context ###
# Add Kubernetes context to Zsh prompt
autoload -U add-zsh-hook
add-zsh-hook precmd update_k8s_context
function update_k8s_context() {
local k8s_context
k8s_context=$(kubectl config current-context 2>/dev/null)
@astevko
astevko / first.py
Created April 8, 2020 18:03
my first gist
print( "hello Andy" )