Hacker News Reader

Comment by zaptrem | Hacker News Reader

I think a personal subscription model where ChatGPT Plus subscribers are given personal API access for use with third-party open source apps (e.g., Chrome extensions, Siri Shortcuts, VSCode extensions) would be extremely valuable and much more lucrative than selling to specific companies.

OpenAI has the opportunity to become the next fundamental utility provider for society. In the same way everybody has a data plan to make their apps work, in the future everybody could have an OpenAI subscription to enable advanced capabilities across their apps. A system where individual companies pay for access makes many of the most incredible use-cases economically infeasible.

Future children will grow up oblivious to their AI subscription in the same way current children are oblivious to the ramifications of their internet subscriptions.

We are going through the AOL stage again.

View on HN

I don't agree with you about how foundational this is, I think chatGPt (as a concept, not a given instantiaiton) is mostly a toy people will get bored with (and not a stepping stone to something more like AGI)

But whether I'm wrong or not I'm very excited about the idea of a ground-up paid service like this that could potentially have a b2c business model based on people actually paying for it instead of being a product sold to advertisers. I hope we're at the beginning of a shift to widespread paid "information" products and away from the race to the bottom an ad-supported tech ecosystem has become

Edited to be more precise

> I think chatGPt is mostly a toy people will get bored with.

I think that would be correct, if ChatGPT doesn't continue to improve.

Given the rate of progress so far, most of us are expecting that there will be much more progress, and it will continue to add more value. This is where it becomes foundational.

Or Skynet :-)

What are examples of that rate of progress? Are you talking about ChatGPT itself or the field?

Both. Everything around ai is moving incredibly fast.

Could you give some concrete examples in ChatGPT's? This field is very abstract to me.

I think you're wrong, many programmers already prefer ChatGPT over StackOverflow.

Stack Overflow is an amazing tool to find solutions to problems that worked 5 to 10 years ago, while newer versions of those questions get closed for being duplicates.

As they should be. No sense in fragmenting questions just because the best answer may have changed.

I don't.

I asked it a solution to a problem using Flask and it gave me a partially working and an incredibly naive solution. I went to Google and found a few naive solutions each with discussions about why they will not work in certain cases. I spotted a blog by a trusted expert, and found the answer I wanted. GPT was convinced, however, that is naive solution was fine.

Dunno if I agree about the "toy" part.

I was learning a new programming language the other day and ChatGPT was able to provide much more focused/helpful responses than Google. Specifically about details of Rust borrow checker, certain syntax etc

It's true that it can give false info at times, but everything it fed to me was true at the time. Time to meaningful response much faster than Google for certain categories of questions...

It may not be a step towards AGI at all, but it's certainly useful

It's certainly at least a step to far better knowledge aggregation, with SEO rapidly ruining search engines as they currently exist.

I think the filler SEO pages rubbed onto ChatGPT giving it a sort of roundabout filler tone.

Agreed. Same experience. It can often ouput some falsities but when one knows what they are doing... It can accelerate coding quite a lot or be a great help in designing algorithms.

Especially since it can ouput code. (from a statement of requirements in natutal language)

Pretty useful.

Faster than going through links after links on a search engine.

> It may not be a step towards AGI at all

It caught on quite quickly for it not to be something very important.

>not a stepping stone to something more like AGI

Why does AGI even need to be part of any of these discussions? It's a ridiculous pop-sci topic, and anyone who works in ML knows we're nowhere near achieving AGI.

That said, what ChatGPT/GPT-3/LLMs represent is a potential new way for people to interact with a corpus of documents that goes far beyond traditional search and knowledge graphs.

I think it's easy to see it as a toy, that's basically how I played with it to start with, it's fun to get it to write stuff about all kinds of things. But it actually has a fair amount of knowledge that has crossed over a threshold where it is useful to ask it questions on a bunch of topics, including programming.

It bears mentioning that the mission of OpenAI is to work towards AGI. ChatGPT is, in the view of OpenAI, one step among many along the way. How it fits into their vision, no clue. But they are certainly happy to monetize this toy to continue funding their real mission. To this outsider, it looks like a win win.

> OpenAI’s mission is to ensure that artificial general intelligence (AGI)—by which we mean highly autonomous systems that outperform humans at most economically valuable work—benefits all of humanity.

> We will attempt to directly build safe and beneficial AGI, but will also consider our mission fulfilled if our work aids others to achieve this outcome.

https://openai.com/about/

> OpenAI has the opportunity to become the next fundamental utility provider for society. In the same way everybody has a data plan to make their apps work, in the future everybody could have an OpenAI subscription to enable advanced capabilities across their apps.

I don't think so. An open source competitor is OpenAI's worst nightmare. We have seen this with Stable Diffusion against DALL-E 2. So if a free open-source version of GPT-4 comes around and it is better than GPT-4, then OpenAI's moat is gone.

There will always be competition and an open-source model will surely come on top, especially something like Stable Diffusion.

> Future children will grow up oblivious to their AI subscription in the same way current children are oblivious to the ramifications of their internet subscriptions.

I'm very sure 99.9% of people are aware that they need to pay for access the internet in their home.

> We are going through the AOL stage again.

I don't think an OpenAI subscription and a AOL subscription is even the same thing or even comparable.

As far ChatGPT is concerned they certainly have the first movers advantage but I doubt if this wil last long. Now if I get a busy can’t talk from dear old chatGPT I just shuffle over to https://you.com/search?q=who+are+you&tbm=youchat&cfr=chat

I didn’t know about you.com, good recommendation thanks!

Selling SaaS services B2B vs B2C is a large topic, and which one's more lucrative for OpenAI is hard to pin down. How many individuals do you know pay for Google Workspaces or Office365, vs how many companies pay for it? Or how about GitHub CoPilot? Or Trello? Obviously those are different products which makes direct comparison hard, but people don't pay for things, that's why there's stealing, and why so many things are ad-supported or ad-"enhanced".

We are in the midst of tech giants fighting, with the consumer as the loser. (Quick, try and buy a book in the Kindle app on iOS!) I don't see a future where it's possible to run (whatever evolves into being) speech ChatGPT on an Apple device with Apple support.

I don't mind paying for AI services, but I want it to be transparent. I want to see the source code, the training data and so on. What I'm paying for should be the massive compute infrastructure needed to train and maintain the model, but something like this is way too important and powerful to be a 'blackbox' SAAS product.

If we think we have problems with Facebook or Google's hold over us, this has the potential to be much much worse if we get the design of the products and the business models wrong.

StabilityAI's solution to provide base models everyone can fine-tune seems more useful. APIs are too restrictive and expensive.

Exactly. I see this being eclipsed by the Stable Diffusion equivalent for ChatGPT. You could then have cloud providers offering whatever specialised compute you need to easily run one of these as a server. Host your own instance in other words.

Obviously a lot of people will just want a simple SAAS product still, we won't want or need everyone, hosting their own apps (too energy intensive for a start). This can then be a SAAS offering, but backed by the same open source software.

Reply on news.ycombinator.com