What you need to know
- Google announces the arrival of its "next generation" AI model, Gemini 1.5, which will arrive in "early testing" fashion.
- To begin testing, Gemini 1.5 Pro will be made available to developers and enterprise customers today with an 87% better performance than the 1.0 Pro model.
- Google states it has also increased the "context window" of 1.5 Pro, meaning developers will have more tokens to use to process more data.
In Google's push for better artificial intelligence, the company has announced the launch of its Gemini 1.5 model. The blog post states the latest model "delivers dramatically enhanced performance" and "represents a step change in their approach." To accomplish this, Google is bringing Gemini 1.5 Pro to those in its early testing environment.
This is said to be a "mid-size multimodal" model capable of handling a "wide range" of tasks. Google states its performance can be compared to Gemini's original Ultra model. Tested for code, text, images, audio, and video, Google adds 1.5 Pro performances 87% better than its 1.0 Pro iteration, which arrived in December 2023.
The tech giant also hopes to bring better context to the user with its 1.5 Pro model. The Keyword post states that an AI's "context window" comprises tokens—which are essentially building blocks for processing data. Gemini 1.5 Pro has been increased to 1 million tokens, furthering its capabilities to output "relevant and useful" information.
The company says that Gemini 1.5 Pro can analyze, classify, and summarize large portions of text. Google gives the example of the 1.5 Pro model handling Apollo 11's 402-page transcription, stating it can understand events, bring up worthwhile details, and more.
Other iterative improvements arrive for users engaging with videos. Google states the new Pro model can analyze plot points, events, and "reason about" small details in videos or movies with length. Additionally, 1.5 Pro can better handle lengthy blocks of code.
The model has been tested with more than 100,000 lines of code and has proven successful as it offers suggestions to correct errors alongside explanations.
Security-wise, Google begins by stating its work done with Gemini 1.5 Pro falls within its "AI Principles." The model is running through several tests for ethics and safety purposes, as well.
Google informs that Gemini 1.5 Pro will arrive today as a "limited preview" for developer and enterprise customers through AI Studio and Vertex AI. Early testers can experiment with the model with a 1 million token cap "at no extra cost," however, there may be an increase in latency as this is a test.
The model will arrive with a context window of 128,000 tokens upon a "wider release." Moreover, Google plans to expand this window to that initial 1 million token mark as more work is completed and testing proves positive.
