Will OpenAI's GPT-4 API support image inputs in 2023?

Question

Resolves YES if:

The OpenAI API allows combined text + image inputs for some model labeled GPT-4, for any persons not affiliates of OpenAI or Microsoft.

Resolves PROB if:

This market does not resolve PROB.

Resolves NO if any of:

No verified access in 2023

OpenAI API is permanently taken down during 2023

GPT-4 is renamed to a different product for image input, even if the underlying model is the same, unless OpenAI explicitly states that the model is the same for both products within the 2023 year.

Resolves NA if:

This market does not resolve NA past 1 week after creation. Before 1 week passes, I can NA for any reason or no reason at all.

Definitions:

An "affiliate of X" is an employee, owner, investor, contractor, or vendor signing an NDA of X. Customers bound by a Terms of Service are not considered affiliates.

"Strong evidence" means a press release, journalist article, interview with an affiliate.

"Verify" means either personally using it, observing someone I believe to be a non-affiliate using it, or seeing a journalist report access.

"OpenAI API" means (API Reference - OpenAI API). Playground is usually released at the same time as general API, but Playground-only access would not count. It is not necessary that OpenAI provides any programming language bindings, as long as the endpoint is documented somewhere and it is possible to write code that uses the official API. If OpenAI renames their company(say to "Fortress AI") and it offers a text + image input substantially similar to the existing API, then it will be considered the same company and product. If OpenAI is acquired by Microsoft, but their existing API product remains online and is extended to support text + image inputs, then it will be considered the same API for this market.

"Image Inputs" means images sent via the API - likely the body of a POST request, or using the Chat API a turn like the "user", "assistant", and maybe "image" tags(though none of this is required) . I expect transmission of colored pixel data Encoding can be lossy(within reason) or lossless. The only requirement is: Text completion, where text is allowed to allude to an image; an image input, that can be alluded to by text instructions; and any specific text prompt containing instructions that need an image input, and an example output for which the model succeessful read the image and followed its instructions.

Description can be adjusted within 1 week after market creation. After that, terms can only be refined to have narrower meanings, or to have additional examples added.

Manifold Markets · Accepted Answer

Yes — resolved on Nov 24, 2023 by Manifold Markets prediction market.