As the race for AI dominance intensifies, the CEO of Google DeepMind — an artificial intelligence research company — claims that its new AI network Gemini will be more capable than the algorithm powering ChatGPT.
Leveraging techniques used in AlphaGo, a DeepMind program that uses machine learning to master the complex board game ‘Go,' Gemini seeks to one-up other chatbots by offering unique problem-solving capabilities and creating novel output that exists outside of its training set.
While Gemini is still in development, it's set to succeed PaLM 2, Google's language model that powers its chatbot Bard, and will likely push forward the frontiers of AI if its rollout goes according to plan. Here's what we know so far.
DeepMind's Gemini Will Top Other AI Models, CEO Claims
After making a splash with its large language model (LLM) PaLM 2, Google's AI research company DeepMind has announced Gemini — a sophisticated LLM that is vying to be the next big breakthrough in artificial technology.
While Gemini is yet to be released, DeepMind CEO Demis Hassabis has already claimed the multimodel model will be more powerful than GPT-4, the technology behind the popular AI chatbot, ChatGPT.
“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, CEO of DeepMind
According to Hassabis, engineers behind Gemini are using many techniques deployed with AlphaGo, a program developed by DeepMind which was designed to master the ancient Chinese board game Go.
While the applications of these tools may seem disparate, both models rely on deep neural networks to execute advanced problem-solving. But superior problem solving isn't the only arena that Gemini is set to excel.
Gemini vs GPT-4: How Do the Models Compare?
Unlike its rival GPT-4, DeepMind's Gemini will be able to handle any data or task without requiring specialized models. Once fully developed, Gemini will have the potential to offer more creative responses too, by generating ‘off-script' content that isn't restricted by its training data, and that's based on the structures it learned during training instead.
Gemini is also expected to gain a competitive advantage by acting as somewhat of an AI Swiss army knife. Since the LLM isn't limited to a single modality, it will be able to create outputs in a range of different formats, including text, images, and audio.
While GPT-4 technology can currently process images, audio, text, and video, it's only capable of generating text, which will make it less equipped at handling advanced requests than Gemini.
But this isn't to say that other chatbots like ChatGPT don't stand a chance. Since GPT-4 is already available to the public, its output will be more accurate and reliable than prompts generated by Gemini, making it a better choice for those using the tool for educational purposes or content creation.
What Does the Future of Generative AI Look Like?
As OpenAI and Google continue to push the boundaries of artificial intelligence, both Gemini and ChatGPT will only create more buzz going forwards. But while DeepMind CEO Hassabis has publically claimed that AI will be the “most beneficial technology for humanity ever” he still admits it's not without its risks.
“If this technology goes wrong, it can go quite wrong.” – OpenAI CEO Sam Altman to Congress
Just last month, Hassabis and several leading tech figures like OpenAI CEO Sam Altman signed an open statement warning that the technology could pose an existential threat to humanity akin to nuclear war and a pandemic if not safeguarded correctly.
To mitigate these risks, the statement calls for companies and governments to regulate the development of these technologies, and to be mindful of their potential effects on wider society. As the race for AI hurdles forward at a breakneck speed, we just hope the CEOs take heed of their own advice.