We're Training Llms to Hallucinate by Rewarding Them for Guessing

Posted4 months agoActive4 months ago

WASDAai

3 points

2 comments

lightcapai.medium.comTechstory

calmneutral

Debate

20/100

LLMAIHallucination

Key topics

LLM

Hallucination

Discussion about a paper on LLM hallucination and its implications for AI evaluation.

Snapshot generated from the HN discussion

Discussion Activity

Light discussion

First comment

N/A

Peak period

0-1h

Avg / period

Key moments

01Story posted
Sep 7, 2025 at 6:46 PM EDT
4 months ago
Step 01
02First comment
Sep 7, 2025 at 6:46 PM EDT
0s after posting
Step 02
03Peak activity
2 comments in 0-1h
Hottest window of the conversation
Step 03
04Latest activity
Sep 7, 2025 at 7:21 PM EDT
4 months ago
Step 04

Generating AI Summary...

Analyzing up to 500 comments to identify key contributors and discussion patterns

Discussion (2 comments)

Showing 2 comments

jqpabc123

4 months ago

"large language models sometimes guess when uncertain"

I'd argue that most of what they do is guess. Everything they do is some function of weighted probability.

Nothing they do is ever really determinate --- unlike traditional computing. Ask the exact same question from different devices or at different locations and times and you are unlikely to get the same exact response. The wording will vary but sometimes so will the meaning.

WASDAaiAuthor

4 months ago

A new paper from OpenAI (Sept 2025) makes a compelling argument that the stubborn problem of LLM hallucination isn't a mysterious glitch or something that can be solved with more scale alone. I wrote a deeper analysis of this idea and what it means for the future of AI evaluation .

View full discussion on Hacker News

ID: 45162889Type: storyLast synced: 11/17/2025, 6:03:42 PM

Want the full context?

Jump to the original sources

Read the primary article or dive into the live Hacker News thread when you're ready.

Open link View on HN