‘Zeter’s Zeter’s’ lets go AI

Researchers to A group of antibala They have made a way to reduce the value of the price and complexity of a machine to explore more, resolve the importance of engines trading APIS everywhere.

The way, called “Zerisch“Allowing major languages ​​(llms) to have the ability to have a psychiatric paste in the process of being connected to pipe learning engines.

To study (rl) teaching regularly, which is achieved about thousands of stages, which spend a lot of money and make the skin, “write the researchers who are A paper published on Arxiv this week. “Consignance, we cause the shrubs of the rolls, the frame of the force that makes up the possibility of llms do not connect to real engines.”

Alback just dropped Zesers on the face

Put the keywords of llms do not search Pic.twitter.com/qfnijno3lh

– AK (@_Khamiteq) May 8, 2025

How Zerych Tsins Ai to check without search engines

The Problem Zerisch Dissolve it’s worth it. Arms that make up a sign of Ai Assistant who can search for information in the faces of two difficulties: Unexpected documents that have been returned with thousand engines.

The Abeba’s way begins with a lighter way that looks for the best way to change LLM in a removal and unnecessary text. Finally, the training process, the system uses the research projects calling “Training Study that gradually reduces.

“Our discernment is that LLM found in the worldwide llm in the world.

To take out Google on food

A balance enough to pass The 7-question question answersZenaker not only the same but often only exceeds the functioning of the work trained by real engines. So wonderful, a 7b-Parameter Redrapet To achieve the same functionality with Google Search, while a 14b-Parateter Even though they were abused.

Products are calories. According to research analysis, teaching with about 64,000 questions by using Google Search via Serpapi It can be damaged about $ 586.70, I’m using the 14b-part of A100 on A100 GPUS $ 70.80 – decreases 88%.

“This indicates the ability to use the User Trained as engines reactions to search,” paper records.

That means the best of AI development

The switch changes in the sophistication of what ai machine would be taught. Zerisch It indicates that Ai can change without being inferior to external equipment as search engines.

Difficulties may be a sort of ai. To date, higher education, AI sisters requires API expenses to work in large collections. Ziesis also changes equation by allowing Ai to change search instead of using real search engines.

For the smaller companies of Aes and the beginning and minimum money, the process can be part of a play. The growing value of API has become a major obstacle to Ai. To reduce the amount of the tree is about 90%, Zerheserch teaches most of the most.

Apart from more money, this method gives the production of officers in the training. When using real research engines, the type of recovered documents are not valid. With the test search, the applers can control the information of the AI ​​views during the training time.

Machines work in several families, including QWen-2.5 with Llama-3.2And with all the lower-case basis. The researchers have made their code, Datases, and the instruction available Grub with The rugged faceTo allow other researchers and other companies to achieve this process.

As the largest languages ​​continue, skills like Zerisch Explain the future that Noi will cause a great potential through the process of defensive activity instead of relying on the development of Ai development and improve the larger tower.

This looks better: Training Ai hunting without engines, while pregnants, which made technology that makes the bills be important to Ai. As the actual acts are very nutritious, technical format may seem very different in a few years.