Google DeepMind’s Gemini Can Rival ChatGPT, Claims DeepMind

As the AI race heats up, Google-owned research lab DeepMind has claimed that its next AI chatbot can rival or surpass ChatGPT. DeepMind is developing its chatbot, Gemini, using techniques from its AI system AlphaGo. As it happens, AlphaGo was the first ever AI system to beat a professional human player at a board game.

At a high level, you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the large models.Demis Hassabis

DeepMind CEO Demis Hassabis confirmed that Gemini could analyze text and plan and solve problems, along with some pretty interesting" new innovations.

One of the biggest differences between GPT-4 and Gemini is that the latter doesn't require specialized models to handle any data or task.

Google's DeepMind announced Gemini soon after making headlines with PaLM 2, its large language model. It aims to make Gemini the next major breakthrough in AI technology, and there's a good chance it might succeed.

Gemini's multi-model will be more potent than GPT-4, claimed Demis Hassabis. Its training data don't restrict Gemini so that it can generate off-script responses. It's speculated that the chatbot will be capable of offering more creative responses once it's fully developed.

Although OpenAI's GPT-4 technology can now process images, audio, text, and video content, the responses are generated in only one format - text. Gemini, on the other hand, will be capable of generating image and audio outputs since its LLM is not restricted to a single modality. Thus, when handling advanced requests, Gemini will likely be better equipped.

DeepMind Applying Reinforcement Learning Techniques to Gemini

With Gemini, DeepMind aims to accomplish tasks that are too difficult for the current chatbots by leveraging reinforcement learning. This learning technique works exactly how the name indicates - the chatbot is rewarded for desired behaviors and punished for undesired ones. The ultimate goal is to teach the AI system how to act in a specific situation.

In a public statement, DeepMind CEO Hassabis said that most beneficial technology for humanity ever".

DeepMind has already gained a wealth of experience incorporating reinforcement learning, and AlphaGo is a perfect example. Considering the results obtained so far, it's no surprise that the company wants to apply the same techniques to Gemini.

One might also want to note that Gemini isn't the first language model developed by DeepMind. Last year, DeepMind introduced Sparrow - a chatbot that was supposedly less likely to give unsafe" or inappropriate" answers than other language models. Though Hassabis said that DeepMind would consider a private beta release for Sparrow this year, we aren't sure if the plans are still on track.

What Does the Future Hold?

One thing is sure - Gemini is among the most ambitious projects by DeepMind so far. The failure of Google's AI chatbot Bard to compete against ChatGPT has led to Google higher-ups rooting for Gemini.

While the future of generative AI remains to be seen, it appears that both Google and OpenAI will continue to push the limits, creating even more buzz in the future.

Hassabis admits that AI tech does come with its risks. He and several other leading figures in the AI domain, such as OpenAI CEO Sam Altman, signed an open statement last month saying that unless the tech is safeguarded correctly, it could pose an existential threat to humanity.

The post Google DeepMind's Gemini Can Rival ChatGPT, Claims DeepMind appeared first on The Tech Report.