What is Google Gemini 1.5? Google is excited about its latest enhancements to Google Gemini AI platform called Gemini 1.5, which comes hot on the heels of the launch of the Gemini platform back in December.
If you’re unfamiliar with Google Gemini, the new name for the initial forays with under the Bard monicker, there’s a really good explainer here from our computing editor Adam Speight.
There are plenty of different Gemini iterations, including Nano (the most efficient model for on-device tasks), Pro (the best model for scaling across a wide range of tasks) and Ultra (the largest and most capable model for highly complex tasks). Today Google is allowing select groups to test Gemini 1.5 Pro ahead of a consumer launch at a later date.
After the 1.0 version left no doubt Google was on track to reel in the early lead taken by OpenAI and Microsoft with ChatGPT-powered tools, the 1.5 update announced by Google today dramatically enhances that performance.
So here’s what Google is really jazzed about.
Kasa (KS105) smart plug is just £8.99
The TP-Link Kasa smart plug is tiny in size and now tiny in price. You can get these handy Wi-Fi connected plugs for 55% off.
- Amazon
- Was £19.99
- Now £8.99
What is Google Gemini 1.5?
Google says Gemini 1.5 “dramatically enhanced performance, with a breakthrough in long-context understanding across modalities.” Overall, Gemini 1.5 Pro outperforms 1.0 Pro by 87% and performs about as well as 1.0 Ultra in benchmarking tests.
The key is what Google calls a Mixture-of-Experts (MoE) architecture that means the entire model isn’t consulted for every query. Google says Gemini 1.5 can “selectively activate only the most relevant expert pathways in its neural network” meaning faster and more efficient performance.
“Our latest innovations in model architecture allow Gemini 1.5 to learn complex tasks more quickly and maintain quality, while being more efficient to train and serve,” Google says.
The company is also pretty pleased about the longer “context window” available within Gemini 1.5 and that’s most evident within the amount of tokens (bites of words, images, videos, audio or code) the model can handle per request.
Whereas Gemini 1.0 Pro has a capacity of 32,000 tokens, Gemini 1.5 Pro can now handle a million at a time.
What does this mean? Well vast amounts on information can now be processed per prompt. Google says one million tokens equates to “1 hour of video, 11 hours of audio, codebases with over 30,000 lines of code or over 700,000 words.”
Here’s a demo of this “long context understanding” based upon the Apollo 11 moon mission’s 402-page transcript.
