A new model of artificial intelligence competes Openai for less than $ 50 in an account

It has become increasingly clear that artificial intelligence language models are a commodity tool, as the sudden height of open source offers is like Dibsic It shows that it can be penetrated together with a relatively small budget. One of the new participants called the S1 again reinforces this idea, as researchers in Stanford and Washington University trained the “thinking” model using less than $ 50 in cloud account credits.

S1 is a direct competitor to Openai’s O1, which is called the thinking model because it produces answers to “thinking” demands through relevant questions that may help it verify its work. For example, if the form is asked to specify the amount of money that you may cost to replace all Uber cars on the road using the Waymo fleet, the question may be divided into multiple steps – such as checking the number of Uber on the road today, then how much the Waymo vehicle costs for manufacturing.

according to TechcrunchS1 is based on a language model on the cliff, which was taught by studying the questions and answers from the Google, Gemini 2.0 Thinking Thinking Treemerial. The Google model displays the thinking process behind each answer, allowing S1 developers to give their model a relatively small amount of training data – 1000 questions carefully sponsored, along with answers – and teaching them to imitate Gemini thinking.

Other interesting details are how researchers enabled the performance performance of the S1 using a simple creative method:

The researchers used an elegant trick to get the S1 to check its work and expand the time of “thinking”: they told her to wait. Adding the word “Wait” while thinking about the S1 the model helped to reach a little more accurate answers, for each paper.

This indicates that although fears that artificial intelligence models reach a wall in capacity, there is still a lot of low fruit. Some prominent improvements on the computer science branch decrease to evoke the correct spelling words.

According to Openai Cry About Chinese Deepseek team training on model outputs. The paradox is not lost on most people. Chatgpt and other major models have been trained in the data that were equipped from all over the web without permission, a problem that is still charged in courts like companies such as companies New York Times It seeks to protect their work from using it without compensation. Google also Technically Comiminos like S1 are prohibited from training on Gemini outputs.

Ultimately, the S1 performance is impressive, but it does not indicate that one can train a smaller model from scratch with just $ 50. Set the form mainly of all Gemini training, and get a cheating paper. Good measurement may be the pressure in the pictures. A bishful version of the artificial intelligence model with JPEG can be compared to the image. Good, but still loss. Language models still suffer from many accurate problems, especially general models on a large scale, which are looking for the entire web to produce answers. But a model like S1 can be useful in areas such as the device processing features such as Apple Intelligence.

There was a lot of discussion about what might mean the high open source models for the technology that is written large. Was Openai be sentenced if anyone can easily copy models? The company’s defenders say that the language models have been permanently estimated to be a commodity. Openai, along with Google and others, will succeed in building useful applications on top of the models. More than 300 million people use Chatgpt every week, and the product is synonymous with Chatbots and a new form of search. The interface is above models, such as Openai’s Operator The user’s web mobility, or a unique data collection such as XAI access to X (previously Twitter), is what will be the final difference.

Another thing to consider is that “reasoning” is expected to remain expensive. Inference is the actual processing of every user inquiry that is submitted to a model. When artificial intelligence models become cheaper and easier, thinking will affect artificial intelligence every aspect of our lives, which leads to an increase in demand for computing resources, not less. And Openai’s Servant Farm Project worth $ 500 billion It will not be a waste. This is as long as all this noise about artificial intelligence is not just a bubble.