Thought you had enough LLMs to keep track of? Here’s another.
The Information is reporting that Google will soon launch its long awaited LLM called Gemini to rival Open AI’s chatbot, ChatGPT.
Google’s been pretty hush-hush about it but here’s what we know so far:
After OpenAI’s immense success with ChatGPT, Google leaped into action.
They first tried with Bard but it didn’t hold up. Google’s own employees reportedly urged leadership to delay its release, citing ethical concerns in the product, but they forged ahead.
After that disappointing launch, two of Google’s AI research team – Google Brain and DeepMind – joined forces to develop an LLM that could compete with Open AI’s and potentially surpass it. (Alexa, play “Eye of the Tiger.”)
Google execs say Gemini will be multimodal, meaning it can process multiple modalities like text, images, videos, audio. In contrast, ChatGPT was unimodal for a long time, only recently becoming multimodal with GPT-4, which accepts image inputs.
It’s also expected to generate higher quality code for developers, and have planning and memory capabilities.
One edge Google has over OpenAI is access to millions of data points from its consumer products, like YouTube. Having proprietary data to train and improve its models is a huge advantage, as many publishers are blocking developers from using their content.
Now, after months of anticipation, Google has finally given select companies access to the LLM, which signals that it’s near a broad release.
Wait, what about PaLM 2? And PaLM? And LaMDA?
LaMDA (Language Model for Dialogue Applications) is a general-purpose LLM that was released in 2021 and powered Bard.
It was later replaced by PaLM (Pathways Language Model), which featured upgrades in language, reasoning and coding capabilities.
Then came PaLM 2: Google’s most recent language model released earlier this year. It’s trained on multiple languages and powers numerous AI features in Google products, like Google Docs’ “Help Me Write.”
Unlike LaMDA, these models are designed for large-scale generation tasks that require creating large amounts of text quickly.
Google execs say Gemini will be even more powerful, more versatile and offer better integration with tools and APIs.
It’s expected to power Bard – along with other Google products – and generate revenue through paid access to its chatbot and application programming interfaces (APIs). Open AI has followed this model to generate nearly $1 billion in revenue.
Google has not confirmed its pricing strategy but one report says that they may offer several pricing tiers, allowing developers to choose based on their needs.
What does this rivalry between Google and Microsoft (OpenAI’s parent company) mean for the rest of us? More options, better products.