Just two months after introducing Gemini, Google’s ambitious large language model aimed at dominating the AI industry, the company is already unveiling its next version. Today, Google is releasing Gemini 1.5 to developers and enterprise users, gearing up for a broader consumer release soon. Google is fully committed to Gemini as a versatile business tool and personal assistant, and it’s doubling down on its efforts to make it a success.
Gemini 1.5 brings a host of improvements. Gemini 1.5 Pro, Google’s general-purpose model, appears to match the performance of the recently launched high-end Gemini Ultra. It outperformed Gemini 1.0 Pro in 87 percent of benchmark tests. A notable advancement is its use of the “Mixture of Experts” (MoE) technique, where only relevant parts of the model are activated for a given query, rather than running the entire model continuously. This approach aims to enhance both user speed and Google’s operational efficiency.
But there’s one exciting new feature in Gemini 1.5 that’s generating a lot of buzz across the company, starting from CEO Sundar Pichai himself: an incredibly large context window. This means the model can handle much bigger queries and process a lot more information simultaneously. The window size is staggering, at 1 million tokens, compared to 128,000 for OpenAI’s GPT-4 and 32,000 for the current Gemini Pro. To put it simply, Pichai explains it as “about 10 or 11 hours of video, tens of thousands of lines of code.” With this expanded context window, users can ask the AI about all that content in one go.
ALSO READ: Top Web Browsers for Windows in 2023
(Pichai also mentions that Google’s researchers are experimenting with a 10 million token context window — that’s like having the entire Game of Thrones series available all at once.)
As Pichai is telling me about this, he casually mentions that you could squeeze the entire Lord of the Rings trilogy into that context window. Finding this oddly specific, I inquire further: Hasn’t someone at Google already done this? Are they testing Gemini to catch any continuity errors, decipher the intricate lineage of Middle-earth, and perhaps make sense of Tom Bombadil with AI? Pichai chuckles and responds, “I’m sure it has happened, or will happen — one of the two.”
Pichai also believes that the expanded context window will be incredibly valuable for businesses. “This opens up possibilities where you can incorporate a lot of personal context and information right when making a query,” he explains. “Think of it as significantly broadening the scope of each query.” He envisions scenarios where filmmakers could upload their entire movie and ask Gemini for reviewer insights, or companies could analyze vast amounts of financial records using Gemini’s capabilities. “I see it as one of the major breakthroughs we’ve achieved,” he adds.
For the time being, Gemini 1.5 will solely be accessible to business users and developers via Google’s Vertex AI and AI Studio. In due course, it will supplant Gemini 1.0, and the standard edition of Gemini Pro — the one accessible to all at gemini.google.com and through the company’s apps — will become 1.5 Pro with a 128,000-token context window. Accessing the million-token window will require an additional fee. Google is also conducting tests to ensure the model’s safety and ethical boundaries, especially concerning the expanded context window.
Google is currently in a fast-paced race to develop the top AI tool, while businesses worldwide are grappling with crafting their own AI strategies and deciding whether to partner with OpenAI, Google, or another provider. Just this week, OpenAI introduced “memory” for ChatGPT, hinting at a potential expansion into web search. While Gemini appears to be quite impressive, particularly for those already integrated into Google’s ecosystem, there’s still a substantial amount of work to be done on all fronts.
