Some sounded the alarm about Alphabet (GOOG 5.34%) (GOOGL 5.32%) after the company reported a slowing growth rate for its Google Cloud business in the third quarter. There were serious concerns that Google simply wasn’t keeping up with Microsoft and its partner, ChatGPT creator OpenAI.
Those worries might have been premature. On Wednesday, Alphabet CEO Sundar Pichai revealed that Google launched its most capable artificial intelligence (AI) model yet. Could it even be a ChatGPT killer?
Pichai stated in a blog post earlier this week that Google is nearly eight years into our journey as an AI-first company. He said that the rate of progress is picking up, with people across the world using generative AI to perform tasks they couldn’t do just one year ago.
Those comments prefaced Pichai’s announcement that Google is now taking the next step in our journey with Gemini, our most capable and general model yet. Gemini version 1.0 will be available in three sizes:
1. Gemini Nano: For mobile and edge devices.
2. Gemini Pro: For developers and researchers.
3. Gemini Ultra: For building large-scale models.
According to Pichai, Gemini represents one of the biggest science and engineering efforts we’ve undertaken as a company.
Demis Hassabis, CEO and co-founder of Google DeepMind, added that the new generative AI model was built from the ground up to be multimodal, which means it can generalize and seamlessly understand, operate across, and combine different types of information, including text, code, audio, image, and video.
Google also announced on Thursday an update to its Bard AI tool that uses Gemini Pro. The company said that the new version of Bard is its biggest upgrade yet, with more capabilities to understand, summarize, reason, code, and plan.
Hassabis revealed that Gemini Ultra beat the current state-of-the-art AI technology on 30 of 32 key benchmarks. OpenAI’s GPT-4 was the state-of-the-art comparison on most of those benchmarks. Gemini Ultra especially outperformed GPT-4 in Python code generation, optical character recognition on natural images, infographic understanding, and video question answering.
He said that Gemini Ultra scored 90% on MMLU (massive multitask language understanding), a test that encompasses 57 subjects including ethics, history, law, math, medicine, and physics. Google’s AI system is the first model to beat human experts, according to Hassabis.
All this means that Gemini Ultra appears to be better across the board than GPT-4. However, it’s too soon to proclaim that Google’s new model will be a ChatGPT killer. OpenAI isn’t resting on its laurels. CEO Sam Altman confirmed in an interview last month that GPT-5 is in early development.
Of course, Google won’t stop with Gemini version 1.0, either. The company said that it is working hard to further extend its capabilities for future versions. Likely upgrades on the way include advances in planning and memory, as well as expanding the AI system’s context window to enable it to provide better responses.
Some investors could be disappointed that Google won’t roll out Gemini Ultra until early 2024. The company said that it’s completing extensive trust and safety checks before a general release of the model.
However, the introduction of Gemini was good news overall for Google. Developers considering switching from Google Cloud to another cloud platform will probably want to think long and hard before doing so.
They don’t have to wait idly for Gemini Ultra, though. Google AI Studio and Google Cloud Vertex AI will enable access to Gemini Pro beginning on Dec. 13. Android developers will also be able to build apps using Gemini Nano.
In addition, Google announced Cloud TPU v5p, its most powerful and scalable Tensor Processing Unit (TPU) system so far. This new TPU platform will help developers train large-scale generative AI models more quickly.
The battle for AI supremacy continues. Google’s unveiling of Gemini shows that it intends to mount a strong challenge.
