Yesterday, we announced OpenAI o3 and o4-mini - the latest in our o-series of models, available now in ChatGPT Team and coming soon to ChatGPT Enterprise.
Check out the demo below to see an example of how a business could put o3 and o4-mini to work.
Imagine you work for a tourism company. You receive thousands of photos from your users every day, but there's no metadata tied to these photos. That's so much rich information that you can't access. Let's see how 03 can help. This is a view of the San Felipe Fort in Puerto Plata in the Dominican Republic. I'm going to ask 03 to identify it, but to make this harder, I picked a view that only shows the Fort in the distance. So I'll upload the photo into 03 and ask where is this? The Oath 3 reasoning model takes some time to think. And then immediately shows a clear chain of thought. It's able to reason that it's a Fort in the Caribbean and it's trying to decide between two different forts, including the correct one. The model is able to use its coding tool to zoom in on the photo of the Fort and assess the thickness of the wall, its overall shape and a nearby statue in the distance to narrow it down to the correct answer. And then to validate the answer uses a second set of tools, web browsing, to look up photos around the Fort. This is a nearly impossible task for human reviewer. Without a powerful tool like 03, that metadata would just never be generated and the tourism company would miss out on that critical information. But of course, the models can do so much more. For example, I can ask Go 3 to generate a list of relevant metadata tags for that image to further speed along the effort to categorize these different photos. And then finally, the O3 model can convert all of those tags that we just generated into structured data for your data analysts. So end to end, the 03 model can convert unstructured photos into structured usual business data that will unlock key value and insights from previously unreachable data. We're really excited for these new reasoning models. Let us know how you put them to work.
It's good idea to improve the JSON output of LLMs because as developers, we mostly have to get data in the JSON structure and feed it to our systems. But sometimes it gets tricky and LLMs are not following the instructions. 😒
This is great OpenAI for Business. We at OPENESS.AI loves building great AI Agents, helping individuals build their career and learnings.
Looking Forward.
Ceo presso Beam Me Up
1wPietro Del Debbio