Faced with the joint encirclement of traffic black holes such as social media, short videos, and generative AI, Google, the former internet hegemon of the search era, came up with its own “killer weapon.”
On February 8, Google announced that it will officially change its AI chatbot Bard to Gemini. At the same time, a new subscription plan was announced. Users can pay $19.99 a month to access its “most powerful model” Gemini Ultra 1.0, which directly competes with competitor OpenAI's GPT-4.
Google called Gemini “the most powerful model ever.” In the evaluation, Gemini did not live up to expectations, sharing the same color with GPT-4, making it the only large model currently comparable to Open AI's most advanced model.
Google CEO “Pichai” (Pichai) was also recently interviewed by the tech media Wired and revealed many thoughts on Gemini and its possible future development path.
“Innate” polymodality
“Multi-modality” is one of the features of the Gemini AI model that Google claims distinguishes it from OpenAI's ChatGPT and Microsoft's CoPilot AI Assistant (also driven by OpenAI technology)'s internal mechanisms.
This means Gemini has been trained on data in many formats — not just text, but images, audio, and code. As a result, the finished model can also use these modes fluently, and can respond with text or voice, or interact by taking and sharing photos.
“This is how the human mind works; you're constantly searching for things and really craving to connect with the world you see,” said Ge Chai excitedly. He said he has long sought to add this ability to Google's technology. “That's why we added multiple searches to Google Search, and that's why we made Google Lens (for visual search). Gemini is inherently multi-modal, so you can put in images and start asking questions. This is where it really makes a difference.”
Is it free or paid?
Google has also conducted parallel experiments on reshaping its core search interface using AI to launch a generative search experience that provides chatbot-like answers before familiar ads and link lists.
Pi Chai Ge said that Google is currently working to make generative AI have a better experience, and he is open to choosing the “free ad or paid subscription” business model. He declined to say whether the paid Gemini products would be completely ad-free, but cited another of its products, YouTube, as an example.
YouTube began experimenting with a paid, ad-free subscription service a few years ago. “Ads allow us to make products available to more people, but there are subscription cases where people get a different experience.” He added, “I can imagine the same user switching back and forth between a free search and a Gemini subscription.” In other words, generative search is no longer an add-on menu to a search, but a main menu item — although it can be more expensive.
It's not an “illusion,” it's an “imagination”
However, PiChai Ge admits that even the advanced version of Google Gemini still has the risk of being “illusory” like Bard or other generative AI apps do. “We want people to be aware of this Chaikko saying. “I think this technology will be useful to many people. But it has to be used the right way, and I'm still concerned that people rely on it.”
Ge Chai said that Google is trying to reduce the phenomenon of models getting out of control. But he also cautioned that the word “illusion” should be used with caution, and implied that illusion is both a feature and a bug, which is an interesting rebranding of misinformation. He thinks the technology should be factual, but if you lower it too low, your chatbot will get really boring and quickly boring.
Ge Chai said that generative AI experiences should be “imaginative.” “It's like a child imagining something without knowing what the limits are.”
Editor/Somer