Google has joined Silicon Valley’s new fashion, artificial intelligence will be responsible for generating the code of most of its products and services. Mountain View have just presented one of their most powerful AI agents.
Google Deepmind has not presented a chatbot or a chatgpt -style image generator, but an AI capable of solving complex computer and mathematics problems. Alphaevolve already surpasses the brightest minds of this field.
This tool is much more than a simple code generator, it is a continuous learning system. The AI of the technological giant is no longer limited to imitating, it is able to create content from scratch.
Alphaevolve combines the best of Google AI
The new agent of the Mountain View is the perfect combination of all the models that Google has designed so far. Alphaevolve mixes Gemini Flash’s creativity to generate thousands of responses to simple problems with the power of Gemini Pro to relate ideas.
Google has gone further, its new AI Add automated evaluation models to filter proposals and a continuous learning system. The operation is simple: the best evaluated solutions according to their own metrics are used as a basis for generating new ideas.
Alphaevolve repeats this process thousands of times until obtaining algorithms. The result? In some cases they already exceed the creations of the best programmers and mathematicians of this field.
Google’s AI is revolutionizing the industry
Google has demonstrated Alphaevolve’s potential with real world tests. Its new AI has improved the efficiency of some data centers recovering 0.7% of computational resources.
Ia models training is also more efficient. Alphaevolve has reduced Gemini training time by 1% Accelerating in 23% a critical kernel in architecture, huge savings taking into account the scale of this model.
Google’s artificial intelligence has begun to solve complex mathematical problems such as the Matrices Multiplication Paradigm exceeding Strasen theory in force since 1969. In addition, this model 75% of the most advanced algorithms have rediscovered by itself and has improved 20% of them.
The key to this system compared to other similar alternatives is precisely in the verification of the results, thus avoiding hallucinations of the own of these LLM. Alphaevolve is not limited to writing isolated code or patches, he is able to build it and evaluate his own answers to improve algorithms.
Know How we work in NoticiasVE.
Tags: Artificial intelligence, programming