GPT4 was announced as multimodal, but the ability to process images has not been publicaly released yet. Will that functionality be publically available this year?
This market resolves YES if I can submit images into GPT4 prompts by the end of the year. Otherwise this market will resolve NO.
I am paying for GPT4, and will sign up for a waiting list if one is released.
Related questions
๐ Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | แน360 | |
| 2 | แน323 | |
| 3 | แน77 | |
| 4 | แน25 | |
| 5 | แน24 |
@SneakySly Wait, what!!! It's a different model! I assumed it didn't count otherwise I would have bet yes.
Bing is much worse than the stock GPT-4 and acts a lot different due to Microsoft's tuning. I don't think it shouldn't count. Can we resolve N/A now?
@ShadowyZephyr Fair, that's a good point. re-reading my initial description I'll stick with not counting Bing.
@SneakySly Bing is a variant of GPT-4 along with a complex frontend that keeps an "inner monologue", runs multiple queries to the underlying model, and does web searches and other tools. The upcoming Bing image input support should be using the same model for vision as the original GPT-4 demo Greg Brockman did - that part isn't finetuned. So it would be reasonable to count it(since it's GPT-4, and the vision part is unmodified), and also reasonable not to(since it's run through a complex frontend).
ChatGPT is also a GPT-4 variant with a frontend that does rolling summarization, has access to tools, but doesn't have any control flow beyond calling out to the model. It seems like it should count, but then Bing has many of the same properties.
If there was more trading activity in this market, this change in clarification would've been problematic. But fortunately, there was only one small sell, so I can exit pretty easily. Otherwise, I may have wanted it to either count or NA.
I've reversed my bet since I think neither ChatGPT nor the API will support image inputs this year.
@Mira I think N/A is fair now, we clearly disagreed on what counts. It's called GPT-4 Prometheus btw.
By "a variant of GPT-4" Microsoft means "we made GPT-4 so much worse that we are only using the name for marketing purposes and it doesn't reflect anything about the original GPT-4." The fact that they did CoT optimization, make multiple requests, and the model still performs this badly, honestly makes me think the model it's calling is worse than GPT-3.5. Calling this "Prometheus" GPT-4 is a stretch. (Maybe the vision part will be unmodified, but I doubt it'll stay that way for long if it's too expensive for Microsoft.)
ChatGPT does nothing to stock GPT-4 except summarization. (And some of the plugins allow it to call the model multiple times maybe? I don't have plugins)