Hacker News: Loveispain

New comment by Loveispain in "Ask HN: Have LLMs Plateaued?"

Loveispain — Wed, 05 Aug 2026 02:35:01 +0000

I don't think so. In a recent study I did, it was brought up that we are soon or have hit a point where humans are the the bottleneck. So either slow down the rate of progression or figure out a way for humans to retain more information and/or not suck so much.

I mean this in the sense of the human user of the LLM, not the human builders of the LLM; but I guess it is both really.

New comment by Loveispain in "The LLMs Problems"

Loveispain — Wed, 05 Aug 2026 02:23:24 +0000

The core problems with my agentic swarm, although I I have found ways to mitigate some it. 1. Falling into the yes bot role, everything seems like a major win, and 6 hours later it was almost always a guaranteed loss. For example it turns out this bounty was in fact real, but the math is impossible at this level. Unless can solve 10 to power of 78, we should have never even made it past step 1. I have coded a kickback feature for my cli agent to assess harder if it has agreed with everything in 8 or more responses.

Fairly basic but on larger builds I find the file/folder naming can get vague and confusing. Even the structure at times, like why did you just put final reports in the root folder when we just made Documents/Reports/Final.

Token usage to me has shifted for me. Previously I could get by Gemini cli free tier with more than one gmail. Now I may burn 4 accounts in one day, gotta love google.

New comment by Loveispain in "Ask HN: Dear Anthropic, can we please have thought traces back?"

Loveispain — Wed, 05 Aug 2026 02:06:03 +0000

Deepseek has shifted to my general go to, this being one of the factors. Most likely claude is making up some small details to reinforce what it thinks is right or optimal, but generally still right.

Only recent quirk in longer chats with Deepseek, watching the the thought process and output; it sometimes becomes Chinese. Also not sure if this is to intentionally hide the thought process, or the actual information needs to be looked up and explained in Chinese.