Google’s Gemini would love to be AI’s evil twin

Google was once famed for its do-good motto, “Don’t be evil”, which was even enshrined in its code of conduct. It quietly changed the motto in 2015 when it was restructured as a conglomerate called Alphabet, and dropped it from the preface to its code of conduct.

Clearly, it was preparing for the era of generative artificial intelligence (AI), ushered in a year ago by OpenAI’s ChatGPT, and variously heralded as both the saviour and destroyer of mankind. The problem, for Google, is that it has played second fiddle to both OpenAI and Microsoft in a field that it has long sought to lead. No one sees Google as the epitome of AI evil, which is where one should want to be positioned as AI market leader.

Last week, it attempted to wrest back that leadership. On Wednesday, it unveiled Gemini, its long-awaited large language model (LLM) – an AI platform that can be used to generate new content.  It comprises Gemini Nano that will run on Android devices; Gemini Pro that went live as part of Google Bard this week; and Gemini Ultra designed for enterprise applications.

“Gemini is the result of large-scale collaborative efforts by teams across Google, including our colleagues at Google Research,” said Demis Hassabis, CEO and co-founder of the company’s AI division, Google DeepMind. “It was built from the ground up to be multimodal, which means it can generalise and seamlessly understand, operate across and combine different types of information including text, code, audio, image and video.”

That sounds exactly like the capabilities already announced for the next version of ChatGPT.

“Gemini Ultra’s performance exceeds current state-of-the-art results on 30 of the 32 widely-used academic benchmarks used in large language model (LLM) research and development.”

In other words, Google’s own hype points to it outperforming ChatGPT and its current GPT 3.5 LLM,  rather than doing something dramatically different from ChatGPT.

But wait, there’s more: “Early next year, we’ll also launch Bard Advanced, a new, cutting-edge AI experience that gives you access to our best models and capabilities, starting with Gemini Ultra.”

By the time Avanced arrives, chances are ChatGPT will have moved from GPT 3.5 to GPT 4, which makes the current comparison irrelevant.

It will, however, help Google catch up with OpenAI and Microsoft’s Bing AI. Not forgetting a dark horse in the race, Anthropic, with an LLM called Claude. It was trained on a high-quality dataset and curated to minimise harmful responses. Google seems to have learned from them how to pretend not to be evil.

Said Hassabis: “For Gemini Ultra, we’re currently completing extensive trust and safety checks, including red-teaming by trusted external parties, and further refining the model using fine-tuning and reinforcement learning from human feedback before making it broadly available.”

Translated, that sounds like a combination of existing strategies and functionalities in the generative AI world.

I asked Google Bard what made Gemini better than its rivals, and it gave me a comprehensive and fairly convincing response, but the most telling comment was this: “Gemini demonstrates ‘novel capabilities’ not present in other LLMs. This suggests potential for future advancements and applications that are currently beyond the reach of GPT-4 and Claude.”

When I asked what these capabilities were, it said: “Unfortunately, Google hasn’t fully disclosed the details of Gemini’s ‘novel capabilities’ yet. They likely want to maintain a competitive edge and prevent other companies from replicating these unique features.”

By saying nothing, Bard said everything. If the novel capabilities will be so easy to replicate that they cannot be revealed a couple of months in advance, they are unlikely to provide the evil edge Google needs.

* Arthur Goldstuck is founder of World Wide Worx and editor-in-chief of Gadget.co.za. Follow him on Twitter and Instagram on @art2gee. His new book, “The Hitchhiker’s Guide to AI”, was released this week.