Another new ChatGPT
Ray Poynter
At the intersection of work, fun & discovery (all views are my personal views unless indicated otherwise).
This month has seen the launch of yet another ChatGPT series from OpenAI. This is probably something we should get used to over the next year or more. This new product, the OpenAI o1 series, is an AI model designed to tackle more complex reasoning tasks. This update includes two versions: o1-preview and the cheaper o1-mini. The preview means we had better treat it as a beta for now.
1. Who is the new product for?
The OpenAI o1 models are aimed at professionals in science, coding, mathematics, and similar fields (e.g. market research and insights) where complex problem-solving is essential. It caters to researchers, developers, academics, and business analysts. With its advanced reasoning capabilities, it is supposed to assist in tasks ranging from scientific research and algorithm development to big data analysis and code debugging.
2. What benefits does the new product offer?
The key benefit of the o1 series is its ability to spend more time 'thinking' before responding, this is intended to make it significantly more effective at handling complex reasoning tasks. This allows it to perform better in areas such as mathematics, where it demonstrated an impressive 83% accuracy in the International Mathematics Olympiad, outperforming its predecessors. It is claimed that it offers improved coding capabilities and enhanced safety features to help it stick to rules and stop security issues. The o1-mini version offers a scaled-down, cheaper option for developers focusing primarily on coding.
领英推荐
3. Current limitations
While the o1 series is perhaps the start of a significant leap forward in reasoning tasks, it comes with some limitations. It is slower and more expensive than earlier models like GPT-4o, and lacks features such as web browsing, image, and file uploading capabilities. Additionally, usage limits are currently in place, with o1-preview restricted to 30 queries per week and o1-mini to 50 per day. These caps, however, are expected to increase as the model improves.
4. What do you need to do, right now?
If you are reading about o1 series for the first time in this article then you do not need to do anything for a month or two. There is interesting work going on, but I would let some others work out the gremlins.
5. Want to know something funny about its codename?
The codename for this product during its development was Strawberry. A few months back a series of social medial posts highlighted that if you asked ChatGPT how many Rs there were in Strawberry, it replied 2. If you ask that question now it tends to say 3 (at least it does for me). But, today I twice asked it how many Rs there are in Cranberry, it replied 2. When I asked it to confirm there were 3 Rs in Strawberry and 2 in Cranberry, it happily confirmed it.
Great Researcher, Insighter and Provider of Solutions at Winton Research and Insights Pty Ltd
2 周This is great news, Ray, especially that the new version will spend more time thinking before answering queries, especially as in recent weeks ChatGPT 4o has been answering increasingly more quickly and skipping or halving some parts of prompts, eg only giving five examples when you specifically state that it must produce at least ten examples. When I try to reason with it, it now answers by providing the references it used and saying I can look up the rest myself. Presumably that is the machine deciding for itself how to ration its time across more customers, but it won't admit it yet. So I am going to give it a very stern warning this morning that much as I love it, I will be trialling ChatGPT o1 in a few weeks if it doesn't give me greater priority!
Senior Managing Director
2 周Ray Poynter Very insightful. Thank you for sharing