Although AI techniques could be used to generate images for years, it wasn't until 2022 that AI image generation became popular. I was surprised by how popular.
Many applications are disqualified because they are already popular, like translation, transcription and facial recognition. Other applications are disqualified because their popularity in 2023 wouldn't be surprising (at least to me), like short video generation.
If there is a lot of ambiguity in how this should resolve and the market has at least 5k in total volume, I'll make a poll and use it's results to resolve.
1,000🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ24 | |
| 2 | Ṁ19 | |
| 3 | Ṁ18 | |
| 4 | Ṁ14 | |
| 5 | Ṁ10 |
People are also trading
@Lily I think most people have been surprised by the popularity of LLMs. An LLM isn't an application, but it enables many applications, like writing essays, helping with coding, chatting, and answering questions, which have all become far more common. Also, I've been surprised by where image generation went in 2023. Hard to get good stats, but I'd guess Stable Diffusion models were used to generate tens of millions of pictures of anime girls. I think at least one of those applications would count as having become surprisingly popular in 2023.
1) GPT
I couldn't find any official stats, but nothing I've seen contradicts the following story:
In November of last year, ChatGPT was released. It got popular, fast. In 2023, usage exploded. According to one traffic estimator, ChatGPT and GPT-4 are on track to get >10 billion pageviews in 2023. Yes, usage may have declined over the summer, and yes that number could be off by a lot, given how unreliable these estimators can be. But I think the order of magnitude is correct.
In terms of applications, I've been surprised by how many programmers (myself included) are using GPT-4 to help assist with programming work. I've also been surprised (although perhaps I shouldn't have been) by how many students are using ChatGPT to cheat.
I've also been using GPT-4 to get answers to questions a quick google search doesn't help with. I'm not sure if this is a common use case, but even if it isn't...
2) Bing
I'd guess that Bing Chat has been used to answer questions maybe 100 to 200 million times so far.
3) Copilot
Microsoft Copilot is being tested with 600 enterprise customers. I don't know what counts as an enterprise customer, so that could be a lot of users, or very few. Could be surprising, or not. I was planning on waiting to see, but I don't think that's necessary given the other numbers.
4) CharacterAI
CharacterAI seems to have had a large drop-off in June, but I think it'll see at least 1 billion visits in 2023. I'm not clear on what these characters are being used for - helping with homework, chatting with simulated celebs, having an AI girlfriend, etc... But it's a lot of something.
I mean, I think this should resolve YES: Students using ChatGPT to help with homework/do homework for them. Became popular in 2023, wasn't popular before that.
Even the people in OpenAI working on ChatGPT said they expected it wouldn't be very popular when they demoed it.
@ShadowyZephyr I agree students using LLMs was rare in 2022. But I and many others expected GPT-4 to come out in 2023 and be enough of an improvement over GPT-3 that usage would increase. Yes, the popularity of ChatGPT was a surprise to most everyone, but I'm looking at AI applications (LLMs for writing essays, image generation) not specific AI products (ChatGPT, Stable Diffusion). Sorry if that wasn't clear.
That said, if the usage implied by the Stanford or Best Colleges survey is the norm, then I will resolve this YES. I wasn't expecting that much use. But I'd like to wait until I have a bit more certainty.
There are also other ways I might resolve this YES - I'm also just waiting on better usage data there (e.g. I wasn't expecting Microsoft to integrate their everything with GPT-4 in 2023).
@AmitAmin A surprise to who? Sure students in CS fields probably knew it was going to happen, but the average person had no idea. The market never clarified who it has to be surprising to…
Good point. I meant surprising to me and other Manifold users. That was somewhat implied. For example the market's description says, "Other applications are disqualified because their popularity in 2023 wouldn't be surprising (at least to me), like short video generation." Short video generation in 2023 would be surprising to the average person, but I said it was disqualified. I said that because it wouldn't be surprising to me (and possibly to some of the Manifold users following AI advancements).
The point of this question was to help me improve my AI expectations by being informed by Manifold users. In this context I didn't (and still don't) care what Manifold has to say about what the average person was thinking.
If I might resolve this NO (which is looking unlikely) then I guess we'll have to debate this more then...
@AmitAmin Honestly I think short video generation wouldn't be surprising to that many people, as long as they heard about deepfakes. On the other hand, the kind of mainstream attention chatbots are getting this quickly, was not predicted. We knew it would happen eventually, but not 2023 specifically.
If you mean Manifold users, it's still debatable, but I think it's more likely that would not count. However, there's not really any empirical evidence, like a poll from before 2023 about AI applications Manifold thinks is likely, so it's still fairly subjective.
@na_pewno If GPT-4’s integrations with Microsoft are successful, I’d be inclined to resolve yes. Microsoft’s search engine and office tools are popular, and I wouldn’t have expected last year that they would deploy integrations with both in 2023. I’m not sure how to judge ‘successful’ though.
@firstuserhere If that kind of thing becomes popular, I think that would count. They weren’t on my radar to play with until just now…