How do you plan on making money with Open AI API?

A plugin for Microsoft Teams called "Waffle" that watches the transcript of the meeting. If you suddenly realise you have been asked a question, or are supposed to say something, but you havent been paying attention, you press the "Waffle" button and ChatGPT feeds you some anodyne lines to say, based on the last few minutes of the meeting transcript.

Well I think you've all raised some really interesting points, maybe we should circle back and think about the new issues listed in the risk register

Perhaps a much better question is ‘how do you plan on making money from people who plan on making money from ChatGPT?’

IMO one of the killer use cases of GPT is reformatting information from any format X to any other format Y, and we're using this superpower in the relatively "boring" space of data extraction:

https://kadoa.com allows you to extract any data from any website in the format you want.

Focus on other things while everyone else is jumping on the AI train.

Here's my guess: ad money. Generating loads of mediocre, simple articles for SEO purposes for ad money. Then you can hire other kinds of bots to pump up your page views. If someone is smart about it they will target articles in the topic of "ways to make money online/passive income" (which is mostly click bait but also pays the most) and then produce videos on YT to push people to their content. ChatGPT can write your video script, probably farm it out to voice actors, add some free b-roll video and post it for you.

One I've been seeing recently (regarding AI generated images) is people making up "fake movies" they blatantly say it's not a real movie... but you produce these videos with text to speech audio... ehh it sucks content farms/waste of bandwidth but whatever. Just like the people restreaming a live nasa page or restreaming spacex/etc... oh well.

It's like those movie summary channels idk why those get under my skin... just a waste.

Oh here's an example Chrome Lords 1988 on YT saw this one recently

Maybe I'm annoyed because it works. You take a bunch of public things like public images of airplanes and make several channels/videos on each plane and facts... you could automate this and eventually something will work. The TTS voice from medium in particular (guy voice).

My idea is to leverage ChatGPT for "thought aggregation". For example, say there are 10K comments on a particular post or video (or any topic). We usually don't value thoughts from "the masses" very much but there could be important ideas emerging / buried in these comments. ChatGPT LLMs could likely be leveraged to aggregate these comments into themes and summarize them in a meaningful way.

One way to think about this idea is multiple choice surveys are great because they are easy to summarize. Their weakness is they overly constrain responses and the aggregated results may not accurately represent the opinions of the people being polled. The idea is to reverse engineer something like survey from free form comments.

I quit my day job and created an email auto reply extensions back in September using GPT-3 and we had a couple hundred people use it. My running total for OpenAI didn’t exceed $20 (their free tier) until a couple weeks ago. The cost of intelligence is going to 0.

The problems most companies in the space are solving are “nice to haves”. The real money is building products that aren’t for a tech adjacent audience. I’ve seen so many interesting use cases and I just had another person reach out who’s debating hiring staff (previously tried) and wants to see what AI can do.

With some custom training we can completely automate aspects of his business. We’re talking about a potential 10k+ customer. I’m convinced there’s millions of these.

1. Buy Microsoft stock.

2. (not me personally) Tuning and personalisation of LLMs. - As a company, I want private data to be combined with internet-scale data so that I can get actionable data - As a company, I want internal identity management to secure who is told what by the LLM that is trained on private data.

3. (not me personally) Quick buck. Skin + API call + marketing = short term cash until people clock on that it is cheaper elsewhere.

My likely unpopular opinion is that it will be hard to make money with Open AI API. The main moat, i.e. competitive advantage in such applications is the model, and infra serving it. Application won't be that hard to reproduce.

If you find a good application with more or less wide market, it will be quickly copied by larger players. Also, the majority of profits will likely go to OpenAI unless there will be other competing providers.

Curie can roughly match davinci quality when fine tuned. Fine tuning's cost is mostly on initial training data.

After that, it's half the price as davinci on paper, but you need only a fraction of the tokens. They've said they plan on making it cheaper in the future too.

I don't think there's a money making play with Chatgpt for the small guy. The move is to improve your output by having it help you with your tasks and adding it's functions to existing apps.

It reminds me of the browser. With browsers there is no money making move but you can use it to improve your work. That and you can't trust the output. Someone needs to babysit it.

I'm building a product using their APIs. Aiming to launch this month: https://inventai.xyz.

The UI will set it apart, and more features are going to be built after launch to further differentiate InventAI from other products.

ChatGPT is just the beginning and a glimpse of what is to come. In its current form, it is rather generic and more of a tech demo. In the future, we will see much more specialized AI applications based on ChatGPT or similar technology, that are geared towards attractive use cases: - software developer support (StackOverflow on steroids) - News aggregation - the automatic newspaper editor or blog curator - Legal support - automatic law clerk

All jobs that use natural language as a tool or resource are fair game.

Another issue that no one is talking about is the quota limit. Say you build a succesful product with thousands of people using at the same time. How to get around the usage limit?

I built a travel-planning site over Christmas: https://www.daystodo.com/

I'm planning on posting a blog about my experience here at some point, but it was fun. You can find a draft of it on the site if you're interested in my thoughts on using it as datasource for an app.

Basically, I agree - you need to use DaVinci to get good results and it's expensive. That means restricting the amount of queries a user can make (I do it through restricting the inputs a user has) and saving the results, so other users get the saved results instead of hitting OpenAI.

For more free-form inputs, I think the only option is to make users pay a small fee (which is tough, because even DaVinci struggles sometimes and I don't want users paying for errors). I'm also experimenting with AdSense but I doubt it'll cover the costs.

I wouldn't say I plan to 'make money' but with luck it might be my first side-project that will break even

How does OpenAI protect sensitive information that is sent to them? This seems like a scandal waiting to happen.

Just yesterday I released a new feature for https://www.clozetesting.com - one can now create gap-filling exercises using AI. E.g. create a story based on keywords, or generate a dialogue or example sentences regarding a specific grammar theme (e.g. superlatives). I hope teachers are going to love this. Check out the demo video https://youtu.be/ur3fP2xfDEY

I found out that it is really good at recommending Movies or Series, based on things that I liked. Maybe that could be turned into something.

I'm doing design for the team at Kadoa [0] which uses AI API to extract unsctuctured data from whatever sources you throw at it, and prepare it for further automated processing. Seems like a valid use case.

https://www.kadoa.com/

I wonder if OpenAI should offer some OAuth flow to allow users to easily grant a certain allowance of credits for an app to use (and later revoke on per-app basis).

With that it would be less effort to create fun and useful little apps without needing to charge the user directly.

All the ideas I can come up with somehow involve…spam.

Recursive prompting works pretty well for niche websites. There is a millionaire computer scientist who sells AI written articles for 50 cents a peace. Can generate those with a few API calls.

Smaller models like Curie can work quite well too. But they are less "instruct-like" models so you will need to properly use few-shot learning (aka "prompt engineering") in order to get good results: https://nlpcloud.com/effectively-using-gpt-j-gpt-neo-gpt-3-a...

I takes a bit more work though, and it makes your requests bigger so more expensive.

Word wiggler. Pipe chatGPT from one computer to another via keyboard emulation and fool the corporate spy software with long email drafts that take you hours.

Using similar tech (Named Entity Recognition, Keywords, Summary ,etc) for finclout.io and are currently evaluation GPT3 for same. At this time, our tech is still more fine-tuned to the task.

But we are evaluating GPT3 for earnings call summaries and comparisons.

We've been working to solve the resume and have seen 10X more job seekers use it in the past few months - https://www.rezi.ai/

Im using GPT-3 to enhance specific features. Users only have adhoc need for those features, so the cost is negligible, but the value is great.

i.e. the product isnt based around GPT, but uses it in specific low-volume places.

Make any FAQs and Tech Documentation humanish interactive. What say?

I'm just speculating but i'm certain there will be quite a few deep-pocketed tech and finance types working out deals for lower rates. Think AWS for LLMs.

How well would it do as a “expert system” trained only in literature for a particular field. Say medicine?

“Doctor GPT3 I feel faint whenever I eat a muffin…”

I’m making a note taking app… sustainable income would be nice, but in the meantime it’s a fun project grugnotes.com

By engaging in SEO against it?

(I say this because that's the way to make money from Google search).

Not sure what's your use case but have you tried fine tuning the cheaper models?

Not plans yet. Still figuring out how to make money on Stable Diffusion.

make it about beauty and fashion to attract females... (skincare routines, beauty advice... ) integrate it into an ecom or recom platform

AI aided shopping "experience"

Levered long MSFT

At this pricing, it might only make sense to embed it in a larger application (e.g. Duolingo) or use it for internal IT.

AI talent obviously

I built my own OpenAI compatible alternative, https://text-generator.io to help with that pricey ness. Also self hostable which helps people cut costs.

It also analyses linked web pages and image content automatically so helps people build these web integrations or auto img alt tag describers or reiept analysers etc without doing as much of the prompt engineering/crawling themselves.