First look at the OpenAI o1 model; An artificial intelligence that is obsessive

OpenAI yesterday o1published his first model of artificial intelligence with the ability to reason. The remarkable thing about this model is its short pause for “to think» When providing the answer required by the users. Recently, the first looks at this hot and anticipated model have been published.

o1 compared to GPT-4o has its own advantages and disadvantages. OpenAI says o1 excels at reasoning and answering complex questions, but it’s roughly 4 times more expensive to use than GPT-4o. Also, the new OpenAI model lacks the tools (such as data analysis and file uploading), versatility and speed that made GPT-4o so popular.

Even the OpenAI website says “GPT-4o is still the best choice for most applications” and o1 is weak in simple tasks.

“Kian Ketanforosh”, Iranian-born CEO of Workera and professor of machine learning at Stanford University, is among the experts who spoke about o1 with TechCrunch:

“There is a lot of excitement in the AI ​​community. If you can train a reinforcement learning algorithm in combination with some of the language model techniques that OpenAI has, you can technically create step-by-step thinking and let the AI ​​model go backwards from the big ideas you want to work on. »

Some examples of experience working with artificial intelligence o1

In one example, o1 AI was asked to help a user plan a family Thanksgiving party and was asked questions; Including a question about whether 2 gas stoves are enough to cook dinner for 11 people and the need to rent a third stove.

After 12 seconds of ‘thinking’, o1 wrote a 750+ word reply telling the user that two stoves are enough and will save you money.

In another example, o1 was asked to plan a busy workday that included multiple meetings and travel. Artificial intelligence OpenAI has provided a very detailed program for this request, although its output is apparently very extensive.

In general, o1 seems to have a habit of generating long answers for even the simplest of questions. When the AI ​​was asked in which region of the United States a cypress tree can be found, o1 produced a more than 800-word answer that included all of the country’s cypress species and their scientific names.

So if you’re looking for short, direct answers, GPT-4o will do better. On the other hand, note that O1 is an expensive model; Therefore, users should be more careful in using it and not pay too much for simple questions. Also, some experts consider o1 to be less of a decision-maker and more of a tool to question your thinking about big decisions.