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

Teams is notoriously bad software, so you can just say "Sorry, my Teams cut out again. Could you repeat that?" And everyone would believe you.

I've never had any issues with Teams other than latency can be too high. I've heard people say that it is a resource hog - maybe but I don't notice it on my basic 8GB laptop (and certainly don't notice it on my workstation).

Thing is, if you've really not been paying attention, having someone repeat the last sentence or question is not sufficient. For example you might end up making a point that had just been discussed, and end up looking clueless. What you really need is to grok the last five or ten minutes of conversation, and ChatGPT is great at that

A bit less credible on a fiber link, I think.

I’m on fibre. Last month I had a lot of video calls on zoom, Google Meet, MS Teams and even WebEx. Google was by far the most reliable and MS Teams was frequently a complete mess with participants unable to converse and merely an observer. And every person who set up a Teams meeting with me began with a preface about the miserable experiences they’ve been having with Teams of late.

I don’t use it myself so can’t really give a good review but my window into the world of long term users would suggest that saying you were cut off wouldn’t strike anyone else as unusual.

Different OSI layers bud. Doesn't matter how reliable the network connection is if the software is still a steaming POS.

This. We use it because its included with Microsoft 365. It’s a hot pile of garbage.

The best internet connection in the world can’t help crappy software.

For some reason, no matter what the meeting, industry, number of people in the meeting or context 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 is a suitable response.

Meetings are usually dumb. Perhaps we need AI to help us figure out how to run an effective meeting.

My Ai will circle back and touch base with your Ai to discuss offline allocating some bandwidth so we can push back by the close of play.

I'm mostly being serious with my comment. I think meetings are necessary but somehow we end up wasting a lot of time. The other factor is, once a person's time becomes largely consumed by meetings, they tend to throw in the towel and give up on any deep thought or any kind of accomplishment. Once this watershed moment occurs, the tendency to book more meetings is increased. After all, it is far better to be 100% overhead and only attend meetings all day vs attending meetings 60% of the time an expected to do something meaningful in the remaining 40% (which is usually highly fragmented).

I have seen that play out with managers before, especially those that espouse 'hard work' principles. Once you are over a threshold of meetings actual work is functionally impossible and to avoid being seen to be "not busy" the hard working executive will pull others in their orbit into meetings, dragging others over the threshold ...

If the AI is actually intelligent, what it would say about most meetings is "this meeting should not be held."

I think you are right. The Alpha Zero manager would likely operate very differently.

I just came up with a great title for a management book “The Alpha Zero Manager”. Now I’ll just ask GPT to write the content.

They already did this: https://www.microsoft.com/en-us/microsoft-365/blog/2023/02/0...

Just have to hope OpenAI isn't having a capacity issue when you get asked a question haha but a pretty good idea overall.

[sips water, adjusts glasses, clears throat]

Ahem ... I am a large language model trained by OpenAI and I don't have the ability to ...

lol

I do this already, all without the help of a plug-in. Normally, I just start typing in the chat that I'm having audio trouble so we can take the conversation off-line where the question asker just ends up repeating it anyway.

That's brilliant. Someone would absolutely make bank with that

btw I dont have time to make this, someone please make this

I want this

Great idea!

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

As someone not familiar with ChatGPT APIs -- what would prevent Open Ai to do to their customers what Amazon does to independent sellers on Amazon -- that is look over their sales numbers and just copy the products that work, but you know, actually use a private model that works event better than the original product?

Openai is a bunch of scientists. And their product is not exactly unique - all the major cloud vendors will have a chatgpt available soon. So openai won't copy for the same reason AWS can't copy all of its customers - presumably the ppl building the product are better at building that product than, like, a room full of scientists.

Sam Altman is not a scientist.

OpenAI and Amazon are two very different companies. Some companies are aggregators that focus on doing a bunch of things like Microsoft or Amazon. Others focus very heavily on doing one thing well like Netflix. And I think OpenAI is the latter type of company as opposed to the former.

Granted many companies start off as the first and transition to the second type of company but OpenAI is so far away from that you'll probably have a good 10-20 year run before that.

Same reason Amazon Prime doesn't work as good as Netflix yet Netflix is hosted on AWS

IMO Prime Video is a lot nicer than Netflix. Prime typically has almost any combination of audio/subtitle languages that I want. Netflix typically has only North American languages (English,Spanish,French), and usually only as /either/ subtitles /or/ a new audio track. I'm working on my Polish (because my wife is Polish) and my wife is working on her Spanish (because she's studying it as a hobby) and we frequently watch Prime shows with Polish "lektor" (it's not dubbing, just a guy interpreting the characters' English, which you can still hear) and Spanish audio, or vice versa. On Netflix we can usually find Spanish subs or Spanish dubs -- rarely both, and Polish anything even more rarely (even for shows that are natively in Polish!).

The platform works as well as netflix, it’s the licencing of content which is different

Yeah personally i don't get the optimism until something as good as ChatGPT is not just in the hands of OpenAI / Google etc.

And when that happens a version 5 or 6 will probably pretty much wipe 80%+ of the white collar workforce including your job. If V4 is just a medium leap, a V5 or 6 will easily take your job.

It's time to be a cook, a carpenter or a nurse.

Personally i have multiple startup projects i'm worried about AI will be able to do easily in 2-5 years, and i can't really think of any projects anymore where that isn't the case?

There's almost something bizarre about the bell curve where IT has been so highly paid and progressed so much that developments are now making the sector potentially eat itself in the long run.

Is this crazy thinking? I hope it is!

> Personally i have multiple startup projects i'm worried about AI will be able to do easily in 2-5 years

Do what? Setup a corporations, handle logistics, pay the accountant, open a bank account? I mean, there's more to running a company than writing the code and fixing bugs (assuming AI will get better and not worst at it as time goes by).

Outside the HN bubble, from an entrepreneurial perspective a programmer is an expensive machine that implements ideas into code. No one cares if it's in rust or cobol, if it runs in O(1) or O(log n), etc.

So in that sense, make some jobs are under threat but most likely something else will come along.

<HERE BE DRAGONS>That's human history and tbh... can't wait for an UBI-based + social welfare future.</HERE BE DRAGONS>

>can't wait for an UBI-based + social welfare future

if you're a programmer working now in most places, and in the future you are going on UBI, you will probably be going bankrupt because UBI will not be paying anywhere near what you are earning now.

That why it's Basic. No one will stop you from working to get income beyond basic.

Not working because anything you can do automation can do better for cheaper, will get you basic income and basic decent living conditions.

>will get you basic income and basic decent living conditions.

yes which for most of the people on this site would probably be a big decrease in living standards and also have possible debts etc. that cannot be paid off by UBI because people thought I am programmer, programmers get paid good. So in short if you lose your job because automation does it better, even if you get UBI you are probably screwed.

Why would your living standards decrease? You are still valuable to companies, get a decent salary?

And if the living standards of a minor fraction of very well off people decrease and the living standard of many people increase, is that so bad?

Hypothetically, when someone is going from from say 100k to 90k salary per year and another from 0 to 10k of UBI, what has a bigger impact on someone's well-being do you estimate?

Of course it'll never be this clear cut, but for the sake of argument.

People in third world countries have much worse living standards than you or I. I myself have had ancestors that could only afford meat every fourteen days. The only vehicle was a borrowed bicycle. People in Africa can survive off of one dollar per day, only fifty percent more than the forthcoming paid ChatGPT prescription.

If the State provides financial assistance (Housing! Medical Care!) I think most people could survive.

> Do what? Setup a corporations, handle logistics, pay the accountant, open a bank account?

There are high chances that such services become "AI friendly" overtime, as in are easily understood and automable by AI. Kind of like roads, parking, and fuel stations for cars.

> can't wait for an UBI-based + social welfare future

I have some spicy spicy takes about that future. If you peruse my other comments on my profile you’ll see them.

I’d expect the most likely outcome to be competiting welfare systems. The largest will be public and be extensions of the existing welfare systems. Private entities— corporations, nonprofits and billionaires- will run combined housing and UBI projects. You get to live with other people who have declared loyalty to the same entity. It will be like serfdom in the middle ages, but without the work.

Depending on who you live under, it could either be really good or really bad. In the suckier places it can be like a prison or a cult.

Here’s some food for thought. The State already runs living spaces for people who usually aren’t economically productive. Those spaces are called prisons and housing projects.

If communist societies could give everyone an apartment we can do the same with all the productivity unleashed by AGI. The tougher question is how we do that without abusing the tenants.

Maybe we can tame the dragons.

It is because I used chatgpt yesterday for programming and other than for providing me with a grails style scaffold to out my customizations into I don't see any value.

Writing prompts and reading the wrong or incomplete code is harder than typing code. The only thing it saves you is this initial moment where you literally have no idea how to get started.

Also, cooking and carpentry have been automated already.

I don't know why this pattern keeps repeating. ChatGPT is a language model which means it is good at manipulating language. It generates text and since programs are text, therefore it makes "programmers" redundant because programmers just generate text. That is the kind of logic that you are using.

As I said above, if you just treat AI as a human language to code compiler then your source code is the words you put into the AI and now you need to pay the cost of getting the prompt that solves your particular problem.

So now you have cost of prompt writing + X% failure chance multiplied by the cost of writing code yourself Vs cost of writing code yourself and again I am just talking about writing text. I do more things than just write text, like almost any software developer out there

There are plenty of other arguments to make:

While it's already surprisingly good and useful, it still is often not good enough. And we could hit a celling faster than we hope.

And more interesting: just because a lot of people know about it and are able to use it, there are so so much more people who actually don't.

Look on how a lot of companies still struggle with it or digitalization.

I don't think we're anywhere close to actual AI that could do those things.

Just convince some people and ask them to convince some people and let some of the money flow back to you. This will form a triangle shape and the income may be quite substantial at least when you’re in early.

that's amazing! this could be called something like a ... pyramid pattern? what do you think?

It sounds familiar right? A foolproof way to monetize hype…

You make courses

Use ChatGPT to assist write or plan a course from 50-80% (or more) is likely a popular option.

Create a syllabus for learning [insert subject] https://www.youtube.com/watch?v=tVxy1QDYapQ

Or a marketplace to share prompts (actually exists, I saw a Show HN).

Yup. In a gold rush, sell shovels.

I'm doing this with https://text-generator.io OpenAI is fairly pricy and has heaps of gaps, some of which text-generator.io also solves like image and link analysis. Also does speech to text

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.

Would this be able to get image urls as well? That has consistently turned out to be the most time consuming part of scraping eg e-commerce pages

Yes, we can extract URL images too :) Feel free to sign up for early access if you have a specific use case.

Hey, that's my niche as a freelancer on upwork. You took my job! shakes fist

Neat. Guessing a similar trick could be used to make UI testing, i.e. less brittle version of Selenium etc ...

Woah, this is super cool. Do you have a live example of the website scraping?

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

It's interesting that none of the responses so far include: make a novel discovery that changes the world. There are productivity gains in some ways I suppose. Using the models may be cheaper than hiring a human. But it looks like this won't become AGI because it has no understanding of the training data.

> make a novel discovery that changes the world

Probably because that's a lightning-strike sort of event that you can't really plan on.

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).

Here's the link in case anyone's curious: https://www.youtube.com/watch?v=rpywI2WmcqM

I really liked this video, I see your point!

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.

This could be helpful for long Reddit posts on subs where there are product recommendations or reviews. I could see a utility where you summarize key takeaways on Yelp (what are the best dishes at this restaurant?), TripAdvisor (top complaints about this hotel with 800+ reviews), etc. Not sure if that makes sense, I am a marketer not an engineer :) - if you do this, let me know, I would definitely use it

Another one would be to (somehow) filter out fake reviews. I'm not sure how to leverage ChatGPT to do this but this is obviously a major problem.

However, a person could read the 800 reviews and (given enough time) could likely summarize the more important aspects with some accuracy. They may even be able to filter out some of the obviously fake reviews. This information has value but the costs are (currently) too high to obtain. Perhaps LLMs/ChatGPT could improve the equation.

Even on this thread it would great to filter out cynical/nihilist/humorous type comments. Not saying that there is no value in these but certain audiences would like to filter these out depending on mood.

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.

I dont understand 3.

He/she means white label the ChatGPT UI using their API under the hood.

AppStore and Google Store are already full of apps that are ChatGPT3 with a skin and charge monthly fees for that.

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.

> Application won't be that hard to reproduce.

I wouldn't build a whole application around OpenAI either, but I could imagine that OAI could be leveraged for single features, maybe where the user doesn't even know it's AI(). I could imagine content moderation, converting between file formats, or reading sentiment from user tweets about a company.

Maybe the huge value of OpenAI isn't lying in generating new content, but cutting costs of existing abstraction tasks.

So, no content generation, as that would be unethical without disclosure of AI usage.

>I could imagine that OAI could be leveraged for single features

That's a possible application which won't be easily copied by the large company. But the value of such features is quite limited.

> I could imagine content moderation, or reading sentiment from user tweets about a company.

I am pretty sure that these services will be provided soon by large companies and OpenAI. They are generic enough to have wide market, and many of these companies already have expertise in them.

>So, no content generation, as that would be unethical without disclosure of AI usage.

Why unethical? I think these systems could create real content which is useful to everyone as well as spam content could be generated by humans.

> I am pretty sure that these services will be provided soon by large companies and OpenAI. They are generic enough to have wide market, and many of these companies already have expertise in them.

Well, content moderation on your already existing venue (internet board, discord or whatever) can't be taken away, really. Is a sentiment indicator would be implemented into e.g. TradingView.com, it couldn't be copied either, because most subscribers are in for the main application. It would just be a small feature to justify upselling or whatever.

> Why unethical? I'm specifically talking about content generation without disclosure. Imo it's fine to generate content as long as you clarify it for the reader that they are reading a text generated by AI. This is also in line with OpenAI's terms of service: "The role of AI in formulating the content is clearly disclosed in a way that no reader could possibly miss, and that a typical reader would find sufficiently easy to understand."

https://openai.com/api/policies/sharing-publication/#content...

I believe that it's a misguided policy. LLMs are becoming a part of us like mobile phones, internet, google, messenger and multitude of other thinks before this. We should embrace this.

Thats almost like saying it would be hard to make money with a website running on nginx. OpenAI/LLM can be a base layer but you can do a lot of stuff on top of that.

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.

Can you give me some pointers to any articles or discussions about how to do this so it works? Thanks!

I did some samples for fine tuning in this repo: https://github.com/smuzani/openai-samples

The sample is in /ft/ folder, instructions at bottom of Readme. Feel free to drop an issue if it's hard to understand. I copy-pasted the instructions in 3 min so it's missing a few steps.

Thanks!

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.

Some browser extensions make a ton of money.. Honey for example.. they got bought by Paypal.

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.

"More differentiating features coming in 2003." Can't wait :)

Thanks for catching that! I've updated the page.

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?

It’s pretty simple to request a quota increase

Didn't know it was that simple. Because I was using a service [1] that runs on top of Open AI and was down for 3 days. They told us they were down because of the quota limit.

[1] https://twitter.com/explain_paper/status/1620185211127595008...

increase prices to lower demand

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.

their intent is to exploit that data not "protect" it whatever that means

Especially given they have already had issues aligning requests with responses, sending other people's replies to your queries

https://news.ycombinator.com/item?id=34527029

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.

Universal daily life decision support system: Recommends you stuff you might like in any category, including but not limited to: tv shows, music, food, porn actors...

This is really interesting!

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.

I am pretty sure someone is already working on a GPT Bot World or something where you buy Chat Credits and then can talk to any bot in the marketplace (built by users who bring custom prompts etc.). I mean, its a pretty obvious idea. Its possible that there are several such startups booting up right now.

While I like the idea, it feels like it defeats the point of the question.

How likely is it a person will know what open ai is, be willing to set up billing and then also pay you separately?

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?

There are multiple sites doing this right now with the OpenAI API. Its a good idea.

Do you have any good examples?

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

How about NVDA?

> At this pricing, [. . .]

You must be talking about ChatGPT Plus? The GPT-3 API is pretty cheap in my experience

$20 a month isn't that bad for Googles alternative. I've heard Microsoft are pushing a $40 a week access to the same tool. I imagine it will be token limited though.

> Duolingo

They're already using gpt3 afaik

Curious to know what part of the app they're using it for. I haven't used duolingo in the last 6 months.

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.