In a recent interview with WIRED, Demis Hassabis, CEO of Google DeepMind, discussed the development and deployment of their latest AI model, Gemini, and the future of artificial general intelligence (AGI). Gemini, a “multimodal” AI model, has already been implemented in Google’s answer to OpenAI’s ChatGPT and various other Google products. With the release of Gemini Pro 1.5, which is more powerful and capable of analyzing vast amounts of text, video, and audio, DeepMind aims to compete directly with OpenAI.
Hassabis explained that Gemini Pro 1.5 allows users to input larger datasets, making it useful for applications such as finding specific information in lengthy lectures or films. He also highlighted the architecture behind Gemini, called mixture of experts, which was developed by DeepMind’s chief scientist, Jeff Dean. This architecture enables Gemini to perform at a similar level as the previous generation’s largest models.
When asked about the significance of increasing computer power in driving advancements in AI models, Hassabis emphasized the importance of scale. However, he noted that DeepMind’s approach differs from other organizations as they prioritize fundamental research. With a strong focus on invention and a team of senior research scientists, DeepMind aims to push the boundaries of existing techniques while also exploring new innovations.
Regarding the future of AI and AGI, Hassabis expressed the belief that several more innovations, in addition to increased scale, are necessary to achieve AGI. While scaling existing techniques is crucial, he also highlighted the importance of exploring new computational approaches. Hassabis mentioned that DeepMind is currently investing in AI systems that go beyond simple chatbots and focus on agent-like behavior, planning, and tool use. These advancements are expected to result in a significant step change in the capabilities of AI systems.
In conclusion, the interview with Demis Hassabis shed light on the advancements made by Google DeepMind in the field of AI models. With the release of Gemini Pro 1.5, DeepMind is poised to compete with OpenAI by offering a more powerful and flexible AI model. Moreover, Hassabis discussed the future of AGI, emphasizing the need for further innovations and exploring new computational approaches. DeepMind’s research efforts are concentrated on developing AI systems with agent-like behavior and planning capabilities, which will lead to significant advancements in the field.
FAQ:
1. What is Gemini?
Gemini is a “multimodal” AI model developed by Google DeepMind. It is capable of analyzing large amounts of text, video, and audio. It has already been implemented in Google’s answer to OpenAI’s ChatGPT and other Google products.
2. What is Gemini Pro 1.5?
Gemini Pro 1.5 is an updated version of Gemini that offers more power and the ability to process larger datasets. It is especially useful for applications that involve finding specific information in lengthy lectures or films.
3. How does the architecture of Gemini work?
Gemini uses an architecture called “mixture of experts,” which was developed by DeepMind’s chief scientist, Jeff Dean. This architecture allows Gemini to perform at a similar level as the previous generation’s largest models.
4. What is the importance of increasing computer power in AI advancements?
According to Demis Hassabis, increasing computer power is significant in driving advancements in AI models. However, DeepMind’s approach goes beyond scale and focuses on fundamental research, pushing the boundaries of existing techniques while exploring new innovations.
5. What is the future of AI and AGI?
Hassabis believes that achieving artificial general intelligence (AGI) requires more than just increased scale. DeepMind is investing in AI systems that go beyond simple chatbots and focus on agent-like behavior, planning, and tool use. These advancements will lead to a significant step change in the capabilities of AI systems.
Key Terms:
1. AGI: Artificial General Intelligence refers to highly autonomous systems that outperform humans at most economically valuable work.
2. AI Model: A program or algorithm that can perform tasks that would typically require human intelligence.
3. Multimodal: Refers to AI models that can analyze different types of data, such as text, video, and audio.
4. Mixture of Experts: An architecture used by Gemini that allows it to perform at a similar level as large-scale models of the previous generation.
Related Links:
1. DeepMind – Official website of Google DeepMind.
2. OpenAI – Official website of OpenAI, a competitor to Google DeepMind.
The source of the article is from the blog maestropasta.cz