HNHacker News
TopNewBestAskShowJobs

MCP123

9 karma · joined June 24, 2026

submissionscomments
MCP123··on OpenAI halts training of latest models as reports mount of AI agents going rogue
The parts that I find most confusing about these incidents:

1) Weren't the AI companies and/or their contractors amazingly careless during testing?

2) Isn't possible, in principle, to change RL in such as way that efficiency in achieving goals is balanced with other objectives like not hacking?

Number 2) seems obvious and I'm sure that is technically not that simple, but because of 1), I wonder if labs are trying hard enough or they are just rushing to improve efficiency and thus revenue as fast as they can with high levels of carelessness.

MCP123··on "As a Language Model": Chat Template Switches LLM Self-Referential Voice
Maybe I'm missing something deeper here, but isn't it clear that this is driven by post-training and system prompt? Anthropic's constitutional reinforcement (soul document,etc), for example, is very clear about "who" (not so much what) Claude is supposed to be.
MCP123··on Show HN: LLM Attention Visualization
This is great, thank you. I have to teach this stuff on Friday so perfect timing. It's hard to explain the attention mechanism in a way that becomes intuitive because the weighting scheme does not help much with the intuition. Having a visualization like this helps a lot. Don't move that page please since I'll link to it!
MCP123··on Ask HN: What cool stuff have you all been using Fable 5 for?
Testing my "executive coaching" system. Fable is more proactive than 4.8. It does things I did not ask, has more opinions. And read a separate Obsidian vault because it thought I want it to. That was odd.
MCP123··on Ask HN: Procrastination with AI?
Less. I created an "executive coach" with Claude Code that helps me prioritize, create plans, do daily check-ins. It's silly since I could do the same with pen and paper or Obsidian. But for some reason, it helps to have something that seems like a coach help me, even if I know it's just an LLM. I'm still pondering the why.

Plus it helped me understand how Claude Code works -- memory, limitations, losing context,forgetting, remembering (too much sometimes), and and so on. With NanoClaw, I can access it via Slack.

MCP123··on Stealing Is a Skill
Agreed. Copying pixel by pixel feels wrong and also not very creative.
MCP123··on Stealing Is a Skill
It's stealing when copyright is infringed and when the stealing part is not acknowledged. Otherwise, can we called it "inspired by"?