Gemini: Almost instant ChatGPT4 abilities for a billion+ users (this is the real news)

Note: Image was generated by ChatGPT4 with the prompt: an image of Google's Gemini announcement.

Today, Google released its much-awaited Gemini model (well, they released part of it and announced what is coming very soon; no, it’s not all available today in Bard and it won’t all be free, so I say “almost...”).

What everyone is excited about is that the most advanced version of Gemini (which is not available yet..) beats ChatGPT4 on 30 of 32 benchmarks, and it beats them significantly on a couple of them. The focus of everyone’s excitement is not surprising given everyone’s obsession with competition (maybe someday all the big AI companies will also own pro football teams) and with “AGI,” though nothing announced today brings us materially closer to AGI.

The gains are somewhat significant in that they contribute to Gemini being “the first model to outperform human experts on MMLU (massive multitask language understanding), which uses a combination of 57 subjects such as math, physics, history, law, medicine and ethics for testing both world knowledge and problem-solving abilities.” That’s something especially significant for educators. And Bard will now use Gemini Pro, making it better at understanding and summarizing, reasoning, brainstorming, writing and planning.

That said, I don’t think any of these gains are monumental (you can review the benchmarks yourself), but I do think this is significant for reasons that most people aren’t commenting on.

Another billion users. Deep Mind’s “Pro” version has the approximate capabilities for ChatGPT3.5 and is being integrated into Googles bard (bard.google.com) and is also free. Microsoft (supported by OpenAI/ “ChatGPT” has an estimated billion users. So does Google. There is obviously some overlap between their user bases, but everyone doesn’t use ChatGPT and/or Microsoft products, so now another (approximately) 1 billion people are getting exposure and access to the world’s most advanced AI technology. This is not technology that can be controlled (school AI bans, please RIP). This is now being integrated into products you already use, making it even easier to use AI and even harder to avoid AI if you want to.

Speed. People often talk about how quickly AI is developing and how we will see more technological change in the next 5 years than we’ve seen in the last 100 years.

Why? There are many reasons (the development of the technology is now exponential), but also because we already have the distribution tools and lines. I’m typing on one of them, and you’re reading this on one of them.

The first public demonstration of electric lighting in the UK was at the International Health Exhibition in London in 1881. It took decades for it to become widely used in homes because the establishment of electrical infrastructure, such as power stations and distribution networks, was necessary for electricity to become accessible to households. This process was slow and required significant investment. Now, AI development requires significant investment, but AI distribution and usage require almost no investment (imagine you lived in 1881 and all you had to do was pay $20/month (approximately 35 cents per month at the time) and make no other personal investment)).

Mobile/distributed. Hardly any attention is being given to Google’s announcement that a version of the model can run on mobile devices. While computer scientists have long been able to use AI models locally on their computers, the recent announcement of being able to run Llama2 on your own device, with some being able to get it working on their iPhones, was significant. Now it will be even easier for everyone to use AI locally. This has implications for privacy, security, and even the favorite topic of many schools: cheating! ‘Just wait for all these kids to have the entire AP curriculum on their phones and they can ask the models questions about it!

AI is fascinating. I didn’t want to bother trying to figure out the equivalent of $20 in 1881, so ChatGPT4 calculated it for me, and when it couldn’t figure it out, it tried again without any additional prompting (see below)!

AI is changing the world and it will change it dramatically in many ways. Are you ready? Are your students? Are you trying?

Less impressively, my sister asked me if AI could generate quiz questions from her Power Point. Of course!