Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
utkarshphirke
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
7 ms
·
1.
▲
by
utkarshphirke
2y ago
Absolutely right - we tried estimating LLM confidence and the results are not great. Any process that requires reliability will struggle with LLM OCR. https://news.ycombinator.com/item?id=43350816
2.
▲
by
utkarshphirke
2y ago
LLMs are quite poor at rating their own confidence. Your best bet is to train a task specific LLM and ensure it is not overfit We benchmarked it here - https://news.ycombinator.com/item?id=43350816