17:[[47548448,{"by":"will_pharaoh","descendants":1,"id":47548448,"kids":[47549166],"score":5,"time":1774646621,"title":"Why AI may increase competition but not success rates","type":"story","url":"https://arxiv.org/abs/2603.23685"}],[47494696,{"by":"noemit","descendants":22,"id":47494696,"kids":[47498409,47496895,47569769,47540979,47518087,47561130,47540200,47499371,47499342,47499000,47548754,47508348,47532964,47526878,47508063,47498203,47500805,47522625,47497365,47498274],"score":41,"text":"I ran an experiment to see if CLI actually was the most intuitive format for tool calling. (As claimed by a ex-Manus AI Backend Engineer) I gave my model random scenarios and a single tool "run" - i told it that it worked like a CLI. I told it to guess commands.

it guessed great commands, but it formatted it always with a colon up front, like \n:help\n:browser\n:search\n:curl

It was trained on how terminals look, not what you actually type (you don't type the ":")

I have since updated my code in my agent tool to stop fighting against this intuition.

LLMs they learn what commands look like in documentation/artifacts, not what the human actually typed on the keyboard.

Seems so obvious. This is why you have to test your LLM and see how it naturally works, so you don't have to fight it with your system prompt.

This is Kimi K2.5 Btw.","time":1774297908,"title":"LLMs learn what programmers create, not how programmers work","type":"story"}],[47536365,{"by":"stackdestroyer","descendants":20,"id":47536365,"kids":[47542795,47540909,47541028,47536376,47558964,47548724,47538378,47548346,47536481,47537910,47539172,47543001,47538722,47540686,47538685,47540934,47571331],"score":10,"text":"I've been interviewing for technical candidates in my current role, and I'm seeing what I believe is a relatively high rate of AI assistants/overlays during interviews. The role I'm hiring for is a Sr. Release Engineer, and even (what appear to be) qualified candidates are showing up obviously reading from prompts. I say obviously for two reasons:

1. I can SEE them reading. Their eyes go back and forth, and it's clear they are reading.

2. They NEVER stop talking. Normally getting answers out of engineers requires a legitimate question with a sound purpose. I'm getting _immediate_ answers with well thought out, multi-point rationale.

So...I'm rejecting all of those candidates the instant I detect it. How are everyone else doing this?","time":1774562388,"title":"Ask HN: How do you deal with obvious AI assistant usage in interviews","type":"story"}],[47530639,{"by":"huize","descendants":1,"id":47530639,"kids":[47530725,47530699,47530640],"score":2,"time":1774533970,"title":"PDF Prompt Injection Toolkit – Hidden LLM Payloads in PDFs","type":"story","url":"https://github.com/zhihuiyuze/PDF-Prompt-Injection-Toolkit"}],[47529107,{"by":"kathir05","descendants":15,"id":47529107,"kids":[47529319,47530518,47532950,47529269,47533193,47530891,47551645,47536707,47563760,47531267,47531097,47545590,47530234,47531358,47529855,47529521],"score":8,"text":"We all start with an idea, 0 customers, 0 followers, no clarity on long term on our startups.

AT some point, you get conviction, this will work and worth pursuing full-time even though 0 revenue to begin with.

For HuntYourTribe, after 2 months, we got 10 portfolios built by strangers and some even kept their link in Github profile. This is it!

Even with 0 paying customers, we decided to purse this startup maze full-time, now 10+ paying customers.

What's your inflection point?","time":1774523912,"title":"Ask HN: At what point you decided to pursue your startup fulltime?","type":"story"}],[47530222,{"by":"roschdal","descendants":2,"id":47530222,"kids":[47531231,47530283,47530285],"score":2,"time":1774531708,"title":"Ask HN: What are the single point of failures of current AI infrastructure?","type":"story"}],[47530156,{"by":"KalskiTheDan","descendants":1,"id":47530156,"kids":[47530186],"score":4,"time":1774531397,"title":"Why Linguistic Context Outperforms Raw Data for LLM Decision-Making","type":"story","url":"https://www.prereason.com/evidence/research"}]]

user:	KalskiTheDan
created:	March 26, 2026
karma:	5
about:	Solo dev building prereason.com. Financial context API for AI agents. Pre-analyzed market briefings instead of raw data feeds.

1.	1 points by KalskiTheDan 51 days ago \| discuss
2.	Why Linguistic Context Outperforms Raw Data for LLM Decision-Making(prereason.com) 4 points by KalskiTheDan 53 days ago \| 1 comment