If you follow the world of artificial intelligence, recently you may have seen the rumor about the project OpenAI Strawberry. Well, now it has materialized in the notice of the t1.
It is directly from a post i published on the blog of the official OpenAI on September 12, 2024, which teach you to launch a new model of IT, with the ability for the common good”. To put it simply, this is the first OpenAI trained to answer questions to the most advanced-and in a way, more quickly than in a man.
t1, which is associated with the t1-mini (a small and free, allows improvement of the largest in the writing of the code, and solve the problem with a lot of steps. Although there is a reference to the ability for the common good”? OpenAI, the head of the training by using the technique of learning the amplifier, the d.m.th. with the aim of teaching and learning for automatic use of a “chain of thought”, which is made up also of the rewards and sanctions, and to develop the questions.
These are technical issues, but at the core of this model is that HE is very different from that of the previous one, because you can study the issues, by reviewing the step-by-step, a little bit like us humans, we are taught to do. This is an innovation, which, in practice, it allows IT to, for example, to solve the most accurate tests in mathematics, and improve the situation in terms of the halucinacioneve (despite the fact that this is a problem that is internal to IT, and for this reason they are not fully solvable) .
An interesting comparison to the new model is the one with the GPT-4o, in relation to mathematics. Compared to t1, and the GPT-4o to the solution of a qualifying exam for the International Olympiad in Mathematics, and, in fact, GPT-4o solved correctly, only 13% of the problemave, while the t1 resolve to the correct 83%. t1 will also be able to solve it correctly enigmat of the text. However, given the training, t1 left GPT-4o, in some contexts, ranging from browsing the web to the processing of the files, and images.
In other words, t1 is not perfect, as at the moment it is only available to users of the ChatGPT Plus, and the Team hold t1-preview (there is also a t1-mini). Soon the pattern will reach the users in the Enterprise, and the Edu, but the game is still in the phase of the experimental. The model t1-the mini will arrive to a user to plan free of charge ChatGPT, but at the moment it is not clear to a date to be determined. It should be noted, on the other, that the access to the model via the API to the developers there is a significant cost: $ 15 for one million tokena of the product.
t1 can be seen from the OpenAI as a first step towards a new model of IT that is going to be implemented in the ChatGPT, so far as the name of “t1” and was created in order to show a “restoration of the numëruesit number 1”. Kind of a “start new”, in other words.
Discussion about this post