The new model from Google "clicks" huge amounts of data like nuts.
The world of artificial intelligence is undergoing amazing changes, and the latest news from Google once again confirms this. Just two months after the launch of the Gemini 1.0 Pro language model that replaced Bard , and a week after the launch of the advanced Gemini 1.0 Ultra, the company announced Gemini 1.5 Pro, a new version of the proprietary LLM model that is designed to further strengthen the tech giant's position in the AI market.
Gemini 1.5 Pro is reportedly a step-by-step change in the company's approach, based on research and engineering innovations in almost all areas of model development and related infrastructure.
In terms of performance, Gemini 1.5 Pro is roughly comparable to the current flagship Gemini 1.0 Ultra, but other innovations in the model are enough to distinguish it in a separate row.
The main innovation is the "Mixture of Experts" (MoE) technology, which allows you to process requests more efficiently and economically. One large neural network will now be divided into many small "expert" neural networks that will be able to give more accurate answers and develop independently of each other, which will simplify the process of their further training.
However, the greatest interest is caused by a significant expansion of the context window of the new model to 1 million tokens, which is several times higher than the capabilities of competitors. GPT-4 Turbo, for example, boasts only 128 thousand tokens, and Claude 2.1 — 200 thousand.
Such a wide context window allows you to process queries with a much larger amount of information, better understand the context, analyze and summarize data. All this opens up new horizons for the use of AI in business and entertainment, allowing you to work with huge amounts of information, including long video files or complex financial reports.
However, the context window of one million tokens will only be available to some developers and corporate clients for the first time, and all other users will have to settle for 128,000 tokens, which is also quite a lot. But this is only a temporary limitation.
The company also said that it is already testing a context window of 10 million tokens. Perhaps this is already some groundwork for Gemini 1.5 Ultra. However, Google's position is clear — the company clearly wants to create the most powerful and efficient AI model on the market.
The Corporation of good does not forget about the safety and ethical aspects of using the new model, especially in light of its expanded capabilities. Google is proud to say that it has conducted many tests and made enough adjustments to make the model truly secure.
All the presented innovations emphasize the relevance and importance of the development of AI technologies not only for business, but also for everyday life. While competition in the AI market is gaining momentum, Google is confidently demonstrating its willingness to innovate, promising users a new level of interaction with artificial intelligence.
Sundar Pichai, the current head of the company, noted that over time, the development of AI will reach such a stage that users will simply stop paying attention to the performance of models, dry numbers and comparisons, and will simply enjoy the interaction experience. However, at the moment, the characteristics of models still play an important role. For how long — only time will tell.
The world of artificial intelligence is undergoing amazing changes, and the latest news from Google once again confirms this. Just two months after the launch of the Gemini 1.0 Pro language model that replaced Bard , and a week after the launch of the advanced Gemini 1.0 Ultra, the company announced Gemini 1.5 Pro, a new version of the proprietary LLM model that is designed to further strengthen the tech giant's position in the AI market.
Gemini 1.5 Pro is reportedly a step-by-step change in the company's approach, based on research and engineering innovations in almost all areas of model development and related infrastructure.
In terms of performance, Gemini 1.5 Pro is roughly comparable to the current flagship Gemini 1.0 Ultra, but other innovations in the model are enough to distinguish it in a separate row.
The main innovation is the "Mixture of Experts" (MoE) technology, which allows you to process requests more efficiently and economically. One large neural network will now be divided into many small "expert" neural networks that will be able to give more accurate answers and develop independently of each other, which will simplify the process of their further training.
However, the greatest interest is caused by a significant expansion of the context window of the new model to 1 million tokens, which is several times higher than the capabilities of competitors. GPT-4 Turbo, for example, boasts only 128 thousand tokens, and Claude 2.1 — 200 thousand.
Such a wide context window allows you to process queries with a much larger amount of information, better understand the context, analyze and summarize data. All this opens up new horizons for the use of AI in business and entertainment, allowing you to work with huge amounts of information, including long video files or complex financial reports.
However, the context window of one million tokens will only be available to some developers and corporate clients for the first time, and all other users will have to settle for 128,000 tokens, which is also quite a lot. But this is only a temporary limitation.
The company also said that it is already testing a context window of 10 million tokens. Perhaps this is already some groundwork for Gemini 1.5 Ultra. However, Google's position is clear — the company clearly wants to create the most powerful and efficient AI model on the market.
The Corporation of good does not forget about the safety and ethical aspects of using the new model, especially in light of its expanded capabilities. Google is proud to say that it has conducted many tests and made enough adjustments to make the model truly secure.
All the presented innovations emphasize the relevance and importance of the development of AI technologies not only for business, but also for everyday life. While competition in the AI market is gaining momentum, Google is confidently demonstrating its willingness to innovate, promising users a new level of interaction with artificial intelligence.
Sundar Pichai, the current head of the company, noted that over time, the development of AI will reach such a stage that users will simply stop paying attention to the performance of models, dry numbers and comparisons, and will simply enjoy the interaction experience. However, at the moment, the characteristics of models still play an important role. For how long — only time will tell.
