Just last week, Google had some major news, rebranding Google Bard to Gemini, unveiling a Gemini app, known as Gemini Advanced, and revealing a new premium AI plan. Continuing its hot streak of news, Google has announced yet another AI development — a new AI model.
On Thursday, Google unveiled its next-generation model, Gemini 1.5. Even though Gemini 1.0 just launched in December, the new model boasts massive upgrades from its predecessor, including a longer context window, better understanding, and an overall enhanced performance.
Also: Nvidia’s new AI chatbot runs locally on your PC, and it’s free
The model is so advanced that Google CEO Sundar Pichai said that 1.5 Pro, the first Gemini 1.5 model that Google is releasing for early testing, achieves comparable quality to 1.0 Ultra, the company’s most advanced large language model (LLM), which was announced last week, while using less compute.
“Longer context windows show us the promise of what is possible,” Pichai added. “They will enable entirely new capabilities and help developers build much more useful models and applications.”
To achieve this enhanced performance, Gemini 1.5 has been built on a new version of Mixture-of-Experts (MoE) architecture, which allows the model to learn and selectively activate the most relevant pathways in its neural network, increasing efficiency, according to the press release.
Also: Microsoft and OpenAI detect and disrupt nation-state cyber threats that use AI, report shows
Google claims that Gemini 1.5 Pro can run up to one million tokens in production, a massive increase from the original 32,00 tokens for Gemini 1.0. This rise is noteworthy because the model context window, the amount of information it can take in, is made up of tokens. Therefore, the more tokens a model can take in, the more likely its responses are to be better and more informed.
Google says 1.5 Pro can process vast amounts of information in one sitting, “including up to 1 hour of video, 11 hours of audio, and codebases with over 30,000 lines of code or over 700,000 words.”
Also: The best AI image generators to try right now
In a demo, Google provided 1.5 Pro with a 44-minute silent Buster Keaton movie, which the model could quickly process and then answer all sorts of questions, including multimodal queries, as seen in the video below.
The model also performed impressively against benchmarks. It outperformed 1.0 Pro in 87% of the benchmarks Google uses to develop its LLMs. Gemini 1.5 Pro also performed stellarly in the Needle In A Haystack (NIAH) evaluation and Machine Translation from One Book (MTOB) benchmarks, which test the model’s acuity and learning abilities.
Also: Slack’s AI features are finally here, including channel recaps, thread summaries, and more
To reassure users about the safety of Gemini 1.5 Pro, Google says that it has conducted extensive evaluations to ensure the safe and responsible deployment of this advanced model.
Google is releasing 1.5 Pro with a one million token context window in a limited preview to developers and enterprise customers, via AI Studio and Vertex AI, at no cost. Once the model is ready for wider release, Google plans to introduce 1.5 Pro with pricing tiers that start at the standard 128,000 token context window and that go up to one million tokens.
