Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

New models and developer products

openai.com

Post reply on HN
Print view View original

451–460 of 561 posts

Re: New models and developer products

#451

Post by lucubratory »

doubtfuluser wrote: post #442

With the assistant API, am I wrong or is it now much cheaper to actually use the API instead of the web Interface? $20 would cover a lot of interactions with the API, and since it’s now also doing truncation / history augmentation the API would have pretty much the same functionality. Thoughts?

You would have to build the UI probably a pre-prompt yourself, but yeah it should be fine if the math does work out. I'm not sure it will though, because if it did any company on the planet could launch a thin wrapper around the API, charge less than OpenAI does ChatGPT+, and undercut them that way.

Re: New models and developer products

#452

Post by lucubratory »

candiddevmike wrote: post #335
atleastoptimal wrote: post #324

Given that their main goal is still AGI, how does offering better developer tools and nifty custom models that can look at your dog for you help? Is it just bolstering revenue? They said they don't use API input to train their models so it isn't making them constantly smarter via more people using them.

They're in the AGI business the same way Tesla is in the self driving car business

That just isn't true according to literally any evidence. People inside OpenAI, those who've gotten access for various reasons e.g. journalists, Microsoft, other investors, their pattern of behaviour, their corporate governance structure, their hiring practices and requirements, etc. They are true believers, at least the vast majority of them.

Re: New models and developer products

#453

Post by yen223 »

openquery wrote: post #378

If I had no contact with society from the 29th of November 2022 (the day before ChatGPT was released according to Wikipedia) and came back today to see the OpenAI keynote I would have lost my mind. The progress and usefulness of these products is absolutely incredible.

Does it boggle anyone else's mind that ChatGPT was released less than a year ago? It definitely feels like it was around lot longer than that.

Re: New models and developer products

#454

Post by lucubratory »

atleastoptimal wrote: post #324

Given that their main goal is still AGI, how does offering better developer tools and nifty custom models that can look at your dog for you help? Is it just bolstering revenue? They said they don't use API input to train their models so it isn't making them constantly smarter via more people using them.

They probably want to train on GPT Builder + store rankings to be able to train an AI to effectively spin up new agentic AIs in response to whatever task it has in front of it. If you're familiar with the Global Workspace Theory of consciousness I think they're aiming for something similar to that, implemented in modern AI systems. They'd like data on what creating a new agent looks like, what using it looks like, and how effective different agents are. They'll get that data from people using GPT Builder, people using "GPTs" and their subsequent ratings/purchases, and the sales and ratings data from the GPT Store, respectively.

Re: New models and developer products

#455

Post by antupis »

bigfudge wrote: post #265
muttled wrote: post #247

Earlier quoted context omitted.

For me personally, being able to fine-tune the local LLM's at a much higher rank and training more layers is very useful for (somewhat unreliably) embedding information. AFAIK the OpenAI fine-tuning is more geared towards formatting the output.

As I understand it, fine tuning is never really about adding content. RAG and related techniques are likely cheaper/better if that’s what you want.

yup how I understand fine tuning is more about adding context and bigger picture. RAG is more about adding actual content. Good system probably needs both in long run.

Re: New models and developer products

#456

Post by Heyso »

cornholio wrote: post #395
somsak2 wrote: post #298

Earlier quoted context omitted.

How do you square this with OpenAI's assertion that they never use data from enterprise customers for their own training? Are you suggesting they're lying?

OpenAI just slurped the entire internet to train their main model, and the world just looks on as they directly compete with and disrupt authors the globe over. Whoever thinks they are not interested in your data and won't use any trick to get it, then double down on their classic "but your honor, it's not copyright theft, the algorithm learns just like an employee exposed to the data would", isn't paying attention.

I agree with you. What come to my mind, is that GPT using private data to learn, if given back to (any) customer, you would have an indirect "open source everything".

Re: New models and developer products

#457

Post by awestroke »

torginus wrote: post #446
openquery wrote: post #378

If I had no contact with society from the 29th of November 2022 (the day before ChatGPT was released according to Wikipedia) and came back today to see the OpenAI keynote I would have lost my mind. The progress and usefulness of these products is absolutely incredible.

I'm sorry, what breakthrough feature did we see here? - Code interpreter, function calling were already possible on any sufficiently advanced LLM that could follow instructions well enough to output tokens in a rigidly parseable format, which could then be fed into a parser, and its output fed back to the LLM. It was clunky to do with online APIs like ChatGPT, but still eminently possible. - Custom chatbots were easy…

You're completely ignoring image analysis.

Re: New models and developer products

#459

Post by Alifatisk »

nmfisher wrote: post #384
unshavedyak wrote: post #381

Earlier quoted context omitted.

Can you go into depth? I’ve used ChatGPT Pro and Phind extensively, didn’t know about Claude and code. Curious to give it a try

I generally use it for boilerplate tasks like “here’s some code, write unit tests” or “here’s a JSON object, write a model class and parser function”. Claude is significantly faster, so even if it requires a couple more prompt iterations than GPT4, I still get the result I need earlier than with GPT4. GPT4 also recently developed this annoying tendency to only give you one or two examples of what you asked for, then…

Imagine being told by an ai to do it yourself, hilarious

Re: New models and developer products

#460

Post by wizzard0 »

wokwokwok wrote: post #354
edgyquant wrote: post #343

Earlier quoted context omitted.

Switching from one API to another generally requires refactoring. I’ve not had much problems moving between LLMs (openai to Anthropic)

Then you’re either not testing your prompts or doing something trivial. Remember: a good model with a good prompt will generate bad outputs sometimes. A bad model with a bad prompt will generate a good output sometimes. That is simply a fact with these non deterministic models. You have to do many iterations for each prompt to verify they are working correctly. > I’ve not had much problems moving between LLMs… If you…

> many iterations of each prompt

BTW its much faster and cheaper to artive at a good prompt if you sample the model in deterministic mode (ie temperature=0)

By default you have to guess if the difference is due to the prompt change or due to the dice roll, as you’ve noticed, but you don’t need to!

Post reply on HN