May 15, 2024

Gemini AI updates. Has ChatGPT won?

On its Google I/O conference, Google announced a couple of AI updates. Will they be enough to overshadow ChatGPT?

In the landscape of AI and software development, updates are not uncommon. But when all eyes are on you after your top competitor has just led a successful conference, you have to be on your best game. Can Google show us something to challenge the ChatGPT-4o dominance?

Gemini 1.5 Updates

Gemini 1.5 Pro has undergone a range of enhancements to improve its performance in various areas such as translation, coding, and reasoning, among others. These updates have been implemented in the model and are now available, enabling users to tackle even more extensive and complex tasks.

Gemini logo and Gemma logo
  • Introducing Gemini 1.5 Flash, a new addition to the Gemini lineup. This model is specifically designed to excel in narrower or high-frequency tasks where the speed of the model's response time is crucial. This is a lightweight model, albeit weaker than the flagship. Both Gemini 1.5 Pro and Gemini 1.5 Flash are currently available in preview mode in over 200 countries and territories. They are expected to be generally available by June. One notable feature of both models is their native multimodal capability with a long context window of 1 million tokens. This allows users to input text, images, audio, and video simultaneously.
  • Gemini context window has been increased to 2 million tokens. One caveat - to access it now, you'll have to enter a waitlist at one of Google's AI Services
  • Gemma 2 is rolled out - it's a 27B parameter lightning-fast model, stronger than other models twice its size. It's also optimized for TPUs and GPUs.
  • PaliGemma, a pre-trained Gemma variant will be rolled out as Google's first vision-language open model

For the review of all announcements from Google I/O check out our coverage article.

ChatGPT-4o Key Features

  • The model has gone truly multimodal, beating other models on audio-, video-, and text-based benchmarks.
  • It's quick, beating ChatGPT-4 Turbo
  • 128K token window
  • Novel functionality like spatial recognition, ability to generate proper-looking letters, and an impressive human-like voice generation.
  • Available now with our API. So go for it if you'd like to test it against other models and decide for yourself!

Google's secret weapon

One thing that gives a certain edge to Google's presentation is this: It has many more models in development.

  • Project Astra: an app that allows AI to use your phone camera directly to analyze the world. Will use speech generation and recognition to communicate.
  • Imagen 3: high-quality image generation.
  • Music AI Sandbox: AI Music Generation.
  • Veo: Sora-like video generation.

Moreover, they've developed a watermarking system SynthID that should theoretically mark generated content for further detection.

This tech is possibly game-changing, and the sheer volume of it at the very least levels the playing field between OpenAI and Google. We'll just have to get the release dates.

Predictions for the Future of Gemini and ChatGPT

Looking ahead, one can expect both Gemini and ChatGPT to continue evolving in their respective domains. Gemini, with its focus on facilitating developer tasks, might see updates geared towards more seamless integration with existing software development workflows, making it an indispensable tool for programmers.

ChatGPT-4o, on the other hand, is the balanced leader of powerful and quick models, making it an invaluable asset for a huge batch of AI needs. It might also expand into other languages, making it a truly global AI tool.

The competition between these two AI models will likely continue, spurring both to develop more advanced capabilities. However, the real winners of this competition will be the users who stand to benefit from the expanded possibilities that these improved AI models bring.

Author: Sergey Nuzhnyy.

Get API Key