🏅 Top traders

thanks brian 😁

My understanding is that this can resolve NO, @BTE, unless you had in mind future versions of GPT-4. https://cdn.openai.com/papers/gpt-4.pdf

GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks

@firstuserhere can't produce image output

GPT-4 LM leak: https://www.maxonlift.com/sites/default/files/2017-04/m9726_0.pdf

@mkualquiera woahhh this defies all my expectations

@mkualquiera So GPT4 created this manual or is this an elaborate joke?

@BTE It's a real truck lift gate which happens to be called GPT-4 LM and this is its manual :D

@BTE It's a GPT4-LM manual

ChatGPT is already vaguely capable of generating some simple images using textual formats like SVG. How would this resolve if, for example, they just update the ChatGPT interface to automatically render any SVG it generates inline?

@NLeseul I saw that. Doesn’t count. True multi-modality is what I am asking about.

Classic delusional @BTE

@Wobbles Microsoft has already leaked that GPT4 will do this apparently.

@Wobbles I agree with @BTE here, my best guess of what's going on after all the recent information is that GPT-4 is released this month and is a multimodal MoE (mixture of experts) model. I'm writing this instead of betting more because I've hit the % limit of my portfolio I'm willing to risk but all the relevant markets seem quite mispriced at the moment.

@BTE "We will introduce GPT-4 next week, there we will have multimodal models that will offer completely different possibilities – for example videos," Braun said. "

They have other models like kosmos 1 they may demo for multimodality.

@NamesAreHard Why do you think mixture of experts?

@NoaNabeshima This is what makes all the other information consistent (but here I'm including a set of rumours which in my view got more credence after the article). It is also what I guess the CTO in the article meant by "GPT4 ..., there we will have multimodal models" - one of the experts does text, one images, etc. Perhaps similar to this but I'm out of my depth here.

GPT-4 is released this month and is a multimodal MoE (mixture of experts) model

Now that the last piece of the puzzle is pretty much clear, totally called it ^^

@NamesAreHard Everything is an ensemble these days.

Suppose GPT-4 has an option for image generation. It uses the latest DALLE to generate an image conditioned on hidden text generated by GPT-4 and then GPT-4 can see the resulting image for further text generation. Does this resolve YES or NO?

@NoaNabeshima I think that is a bit in the weeds, this question is simply about whether or not it has the ability to generate multi-modal outputs. An example would be a textbook with graphics and diagrams inline with the text.

@BTE But what counts as GPT-4, then? If I take a text GPT-4 and combine it with DALLE to produce a website that creates illustrated stories, does this resolve YES?

Is it that it needs to be called GPT-4 and it needs to be on an OpenAI or Microsoft-controlled website?

What if it's not claimed that this is GPT-4, but rather a "picture-book experience produced using AI tools from OpenAI" but it clearly uses a text GPT-4. How would this resolve?

@NoaNabeshima Yes that is a good characterization of my thinking. I think OpenAI has already been very clear that will be the case. I heard Bill Gates on a podcast the other day say he has used GPT4 and intimated it would have some multi modal capability.

@NoaNabeshima We are waiting for GPT4 to see what it does. That is pretty simple. No tricks.

@BTE

@NoaNabeshima Like nothing but the capability of THE ONE AND ONLY GPT 4 counts.

@BTE Do you mean this interview?: https://www.futuretimeline.net/forum/viewtopic.php?p=33495

@NoaNabeshima Yeah it was great. He talked about giving it questions from an AP bio test and it blowing his mind and I took that to mean it generated a textbook and not just an essay. But he was cagey.