Hacker News Reader

Comment by davidweatherall | Hacker News Reader

https://x.com/dWeaths/status/2102415625301717065 - Here's my use case of jev, being able to accurately detect nouns (with adverbs, adjectives etc.) in realtime as the user is typing it, genuinely feels like it's running locally with how fast it comes back. I've sent over 1000 requests to Jev and its cost me $0.01 (probably rounded up!).

View on HN

I'm curious why you didn't use a spaCy for this. It would probably be faster, and you'd avoid a third party API dependency.

For my specific example: "the tall dark handsome man wanders into a dark gloomy bar. he orders the biggest beer in the world and sits down on a bar stool surrounded by irish dancers" spacy splits out "the biggest beer" and "the world", whereas I want "the biggest beer in the world" to be the singular noun.

So, you send Jev a string of text and it sends back what? I thought it only output probabilities of a classifier.

When the user is typing the string, I seperate the last 10 words into reversed joined words, e.g. tall dark handsome man becomes "man", "handsome man", "dark handsome man" and "tall dark handsome man", and I send the full string to Jev and ask it to tell me with probabilities of each option which is the best one to fully capture the descriptive noun

I mean now with modern LLMs a lot of good packages and tools get forgotten about. When you have hammer everything looks like a nail. Even if said "old" tools are actually orders of magnitude faster, and sometimes better too for that specific task. (And I remember spacy being basically SOTA for generalist NLP tasks not that long ago, like 2020/2021).

Reply on news.ycombinator.com