Google just launched Gemini – its most capable LLM

Alphabet Inc (NASDAQ: GOOGL) is in focus on Wednesday after it started the phased rollout of Gemini – its most capable large language model to date.

Gemini Ultra beats GPT-4 on MMLU benchmark

The LLM comes in three different sizes: the Gemini Ultra, Gemini Pro, and the Gemini Nano.

Google will make the “Pro” variant available to developers and enterprise customers on December 13^th but they’ll have to wait for the “Ultra” for another few months.

The tech behemoth also confirmed today that its Gemini Ultra is “not just more capable, it’s more efficient” as well. It scored 90% on the Massive Multitask Language Understanding (MMLU) benchmark versus OpenAI’s GPT-4 at 86.4%.

Google will soon introduce an all-new “Bard Advanced” that will use Gemini Ultra. For now, the chatbot will make do with the Gemini Pro, it added.

Google also launched a new AI chip today

Also on Wednesday, Alphabet Inc unveiled the TPU v5p – its next-gen tensor processing unit aimed specifically at training artificial intelligence models.

But the multinational did not reveal how it stacks up against the market-leading products of Nvidia Corp. On CNBC’s “Squawk Box”, Dan Flex – a Neuberger Berman analyst recently said:

Google is infusing more Gen AI capabilities. Search experience is changing. It’s creating value for advertisers and users. That’ll create a larger market that Google will have an attractive share in.

Note that McDonald’s Corp also partnered with Google today on AI solutions to improve operations as Invezz reported here.