Live data from Hacker News

GPT Store Launch

news.ycombinator.com

11–20 of 69 posts

Re: GPT Store Launch

#12

This looks like an interesting opportunity to learn something new. Anyone knows about any open source projects in this area?

GPTs is just a glue layer. It adds some but small values on top of GPT4, emphasizing on small.

If you don’t have GPT4, no value is added

Re: GPT Store Launch

#13

This looks like an interesting opportunity to learn something new. Anyone knows about any open source projects in this area?

It depends on what you are looking for here. If you want to build with LLMs, there are a number of open source options which can be self hosted (although they may not perform on par with GPT-4).

If you are referring to the assistants API which adds some more complex behavior, there is LangChain as others mentioned, but also some more turn key, self hosted options (which I have not tried) such as

- https://github.com/stellar-amenities/assistants

- https://github.com/transitive-bullshit/OpenOpenAI

If you are referring to the marketplace itself, most developers are currently rolling their own web apps with billing and auth while they wait for OpenAI’s offering.

Finally, as a shameless plug, I’ve been working with some friends on a marketplace which provides auth and billing but decouples you from a specific model provider and the high platform fees they may plan to charge. It isn’t open source but we think it might strike the right balance. https://market.interactwith.ai/

Re: GPT Store Launch

#14

chatgpt is too lame for this to work you can't even ask it to spell words incorrectly if you ask gpt4 to spell words incorrectly, it fusses about it. i'm commenting on the fact that the company is too stuck on appearances

> you can't even ask it to spell words incorrectly

Yes you can.

https://chat.openai.com/share/cd28dc1e-cb50-45b0-986b-275d53...

Re: GPT Store Launch

#15
post #10

chatgpt is too lame for this to work you can't even ask it to spell words incorrectly if you ask gpt4 to spell words incorrectly, it fusses about it. i'm commenting on the fact that the company is too stuck on appearances

Why does a store need apps with incorrect spelling?

Also, you could probably get a ChatGPT instance to create and run a Python script to randomly replace letters in the words that ChatGPT itself creates. So technically, you can get it to misspell words. It seems like the person you replied to is forgetting the point of pre-training transformer models on instructions.

Re: GPT Store Launch

#16
post #7

Earlier quoted context omitted.

GPTs are essentially a proprietary moat-building of several open-source workflows, most especially system prompt engineering, tool-usage, and RAG. Unfortunately there's no magic open-source solution since there's a lot of moving pieces involved that are bespoke to a given use case, and the ones that claim to be magic are libraries like LangChain, which aren't.

This is the most accurate and clear-eyed take I've seen on GPTs so far. They might be useful, but they're not magic and they're intended to enhance OpenAI's moat-building operations to make it harder for people/companies to walk away from the platform amid future competition.

I believe GPTs are an attempt by OpenAI to generate training data. How can you get data at level N+1 when you have a model at level N? You give it more resources - more tokens (CoT), more dialogue rounds, code execution, web search, local KB, human-in-the-loop. A model with feedback from human and tools can do so much more. And by training on this data they can incorporate these skills in the next generation. It's like RLHF in the sense that the training data contains portions generated by the model itself (specifically model errors) and feedback. It's on policy data, generated with the involvement of the model, not something you can scrape from the web.

Let's do an estimation - if they have 100M users and each of them generates 10K tokens in a month, that's 1T tokens per month. In a year they have generated 12T tokens, which is very close to the GPT-4 training set size of 13T. Looks like they can generate serious data with this method. They don't even need to train directly on it, they could rewrite it as high quality training examples, without copyright and PII risks, because LLMs are great at rewriting and rewording and MS has already shown that synthetic data is better.

Google lost the start and they don't have the human-AI chat logs OpenAI sits on. So they are trying to do the same trick but without the human in the loop. Hence the declarations that Gemini will use some techniques from AlphaZero. They are teaching models by feedback too.

Re: GPT Store Launch

#17

chatgpt is too lame for this to work you can't even ask it to spell words incorrectly if you ask gpt4 to spell words incorrectly, it fusses about it. i'm commenting on the fact that the company is too stuck on appearances

> you can't even ask it to spell words incorrectly Yes you can. https://chat.openai.com/share/cd28dc1e-cb50-45b0-986b-275d53...

Not going to lie, those spelling mistakes appeared very natural. Had they been part of an assignment by a 5-year-old, they would have looked quite authentic.

Re: GPT Store Launch

#18
post #8

This close to launch and no information on how they'll audit GPTs. I just want to know their review process and how they'll enforce their usage policies, specifically regarding privacy (and GDPR compliancy). How will they prevent a developer from sending and storing private information from their GPT to their server through APIs? How will they prevent doxing? Such as a GPT posting hyperlinks that direct to a server o…

Also, who is on the hook when your app outputs copyrighted material? https://garymarcus.substack.com/p/things-are-about-to-get-a-...

They are, via their "Copyright Shield" program: https://openai.com/blog/new-models-and-developer-products-an...

Re: GPT Store Launch

#19

chatgpt is too lame for this to work you can't even ask it to spell words incorrectly if you ask gpt4 to spell words incorrectly, it fusses about it. i'm commenting on the fact that the company is too stuck on appearances

There's a lot of theories floating out there that OpenAI will make gpt-4-turbo free (or at least free when used via custom GPTs) so the default won't have to be 3.5-turbo (which is terrible at following system prompts and pales in quality compared to 4/turbo).

Re: GPT Store Launch

#20
I received the same email. My assumption is that all of the GPTs in the store will still require a paid ChatGPT plus plan to access, but I have no way to confirm this. Does anyone have a source verifying whether this will be the case?
Post reply on HN