GPT-4: 32k and Image recognition

Anyone’s got news on this? Did the public release of these get cancelled or should we expected a similar roll out to gpt-4 API - with early developer access - later this year?

Any news on this is appreciated it! Also, what would you use it for? 32k is easy to imagine, but the image recognition? so many different ways to go about it

1 Like

32k will be rolled out, but it is super heavy on resources so it will be a gradual one as more compute gets put online, image input will (I imagine) follow a similar rollout to GPT-4 in that respect. Slow initial alpha with a slow beta after that and then a full general release.

100% speculation but just following from prior offerings.

What can you do with an AI computer vision system?

Bulk CAPTCHA solver. My theory why you might not see it for a while.

It just means we need a new method of determining who is human, but existing CAPTCHAs have been broken for a while now by anyone moderately sophisticated and properly motivated.


There is a lot of cool and scary things we can do with GPT Vision. CAPTCHA solver is on the low hanging fruit rung. How about build out software from a design mockup which is only one or two rungs above a CAPTCHA solver

1 Like

Reading schematics and PCB layouts and designing electronics is one of my “would like to try” items


I mean… I’m guessing half of those would get you banned from from all OpenAI services in a digital heartbeat.


Now this… This is interesting…

I know machine learning has been used in chip design, but it would be a boon to makers everywhere if there were a readily-available, affordable system that could from an image of a PCB, redesign it to be more compact, use fewer parts, etc…

I think this is one of the best ideas I’ve heard for this yet.


If GPT-4 includes vision, does this mean it will be possible to do face recognition, maybe by doing “facial data embedding” or “facial data fine tuning”?

1 Like

Anything is speculation at this point regarding this, we need it in our hands to test it out. Facial recognition is already achievable with other AI methods, so I don’t see why it would be a problem. Fine tuning it would probably get it to be more accurate, but given that we don’t have GPT-4 fine tuning out, you can expect for different solutions around this to pop up, such as using databases in production and variables/lists/arrays in a experimental/developmental environment.

Further to this, facial recognition is a trivial task with mature technology, it could be that it will be part of some subsystem, but the point of image ingestion is to augment the models information input ability, it may be able to identify if the human in the image is wearing a shirt and tie or a t-shirt, but it won’t know the persons name.

here’s some good ones I can think of:

  • Locate cancer cell(s) and destroy it (virtual simulation done on unreal engine)
  • Video game npc village, with real time reaction to events
  • Based on medical exams, propose a diagnosis
  • Fully autonomous surgery with multiple agents on different focal view points (microscopic precision with macro perspective range of understanding)
  • Take out the garbage bot lol

Also, careful with sharing harmful ideas, it’s always best to not propagate them in my opinion

1 Like

One thing I think will be interesting is to include it in other image pipeline tasks.

For instance, take Meta’s Segment Anything model. A natural next step in the pipeline could be to meaningfully label the segments with GPT-4.

Then, I’m immediately imagining a content-aware segmentation model. Where knowing what the objects are it will be better able to determine segments, particularly partially obscured or occluded objects. Then using GPT-4 to label those segments.

Perhaps looping iteratively to either fine-tune the segmentation map, or possibly creating nested segmentations.

1 Like

I found information on the 32k model, no word on the image recognition functionality:

API Access

On July 6, 2023 , we gave all API users who have a history of successful payments access to the GPT-4 API (8k). We plan to open up access to new developers by the end of July 2023, and then start raising rate-limits after that depending on compute availability.

We are not currently granting access to GPT-4-32K API at this time, but it will be made available at a later date.


I need image recognition for some of my use cases. It’s advertised in OpenAI’s web site and apparently available through Bing, but it is still not available through the API. I found a video some time ago with someone from OpenAI teaching how to use it, but I tried and received an error message back. In summary, you guys from OpenAI created the high expectation, now please give us a chance to test it even if in beta.

how can to swich ChatGPT version 3.5 to version 4 ?

Bing Chat has been occasionally been rotating the availability of GPT-4-powered machine vision to some users.

Those who get access first: those who put $10 billion into OpenAI. Microsoft. Consider they also have the AI power to recognize and blur faces before submitting for AI answering analysis.

Hi. You have found the wrong forum category and posted to the wrong topic. However I can answer that ChatGPT with GPT-4 enabled is a feature that is available to plus subscribers at $20 per month, an upgrade available within the chatgpt web interface at