HN Explorer

Ask HN: Have you seen any useful GPTs yet?

11 points by hubraumhugo ↗ HN
By now I've tried quite a few custom GPTs that were posted by "AI influencers", but haven't found them particularly valuable.

Aren't these essentially pre-configured prompts? I see custom GPTs as helpful, shareable prompt templates. Maybe also a good strategy to convert people to the ChatGPT Pro plan.

I also tried to create a Hacker News GPT that integrates with the HN Algolia API, but had a hard time making it call the correct API endpoints and handle edge cases: https://chat.openai.com/g/g-BIfVX3cVX-hackernews-gpt

12 comments

[ 4.1 ms ] story [ 40.7 ms ] thread
xkcd1963
You mean as alternative to openAI?

I'm really not on board with the current hype. It is not useful for my work (webcoder), but if I've to write in business-attire I find it useful.

atleastoptimal
The only really good ones I would imagine are any one not disadvantaged by having any unique UI elements and can derive their value 100% via a simple linear arrangement of prompt responses and API calls. I feel though that even in the short time it's been out almost all the obvious good ideas have been tried.
loveparade
Like you said, it's mostly influencers trying to cash in on the hype and I don't expect that to change.

> Aren't these essentially pre-configured prompts?

They also include custom knowledge and actions, and I don't think these are simply added to the prompt. But I don't think OpenAI has published how this data is fed into the model internally?

In general, I've had mostly bad experiences with models that include retrieval or external APIs. It's just too brittle because the model needs a very precise understanding to retrieve what you want, not fuzzy language. The by far best experiences is when you can put everything into the prompt.

LucieXLiu
Recently saw a Scale x OpenAI webinar that described some strategies for improving RAG systems. Looks like there are a good number of strategies to try for optimizing the retriever, like optimizing chunk sizes and classifying/ranking chunk relevance, but building a robust RAG system require rigorous experimentation against the specific dataset that feeds the retriever. Not sure how many of these GPTs bothered with this experimentation, and I think you are right that without knowing how that retrieved data is fed into the model they have very few parameters to experiment and optimize with, so they probably just can’t.
Mrirazak1
I think right now their is too many mundane things— just ppl playing around— whilst you have a few big gpts like canva etc.

It’s still nothing major.

Botir
I'm working on gptscatalog.com, a comprehensive catalog of custom GPT models.

It's all about exploring the potential of GPTs!

philomath_mn
There is a lot of potential, but the OpenAI-provided ones are pretty rough. I asked the math tutor some questions about the relationship between sine and cosine. The answers were decent (no different than base GPT4) but when I asked to see a visualization it tried using DALLE instead of a more thoughtfully constructed matplotlib plot.

I think they have a way to go.