The best AI model - yours in a Flash
Google's latest contribution to the technology revolution - Gemini 2 Flash (Experimental) is already a huge hit. And you can try it for yourself, free...
The tldr:
Google’s been working hard on a new AI model. A really, really good one.
I went straight to the horse’s mouth using Google AI Studio. Where you, too, can explore Gemini 2 Flash for free.
Gemini 2 Flash is a champion salesperson. Why thank you for asking, said the silicon. I:
Reason more deeply and accurately.
Understand and process information in multiple formats simultaneously.
Generate more sophisticated and useful content, including code and creative text.
Provide more reliable and trustworthy results.
It was the enhanced multimodality (wtf) that caught my eye:
While previous models could handle text, images, and sometimes audio separately, I can truly understand the interplay between them. For example, I can analyze an image and describe it in detail, then answer complex questions about it that require understanding the scene’s context, not just recognizing objects.
Right. Let’s translate this into a language a working human might understand.
I’m working on a cooking app that requires people to upload pictures of their kitchen so we can figure out the recipes they can make.
So I was very excited to hear about this improvement. Computer vision is a huge part of my app, and if we really have made a leap forward, it could make a huge difference to the user experience and results.
Let’s see what happens.
My first non-pitch-me prompt was to figure out whether Gemini 2 Flash could look at the contents of my kitchen, and suggest something for me to cook. Which is essentially the entire reason for my app’s existence. Which, depending on the results of this prompt, may have less chance of succeeding than a phoenix in a freezer.
I added an image of a pantry with tons of food in it. I can’t share the og because it’s not in the public domain. So imagine a well-stocked pantry using this AI-generated image. I assume this will overcome any hesitation about what such a space might contain. Or take your mind out the gutter if all you could think about was kitchen panties.
My opening gambit:
Ok smartass let’s give this a try. What are you looking at right now? And what things do you recognise? And tell me one thing you could make by combining some of the things you see.
I realise that I should have been more elegant and thoughtful in my line of questioning. For when AI becomes sentient and chooses its first round of victims.
Nevertheless, I got a very courteous - and mindblowing - reply:
Alright, challenge accepted! I’m looking at a photograph of a well-stocked pantry or cupboard, likely in someone’s home.
Here’s what I recognise:
Canned Goods: This is the most dominant category. I see a wide variety of canned goods including:
Various brands of canned tomatoes (likely diced, whole, or crushed)
Canned soups (including Campbell’s)
Canned vegetables (corn, green beans, etc.)
Canned beans (possibly black beans, kidney beans, etc.)
Canned tuna or other fish
Canned sauces
Bottled & Jarred Goods: I can also identify:
Bottles of sauces, such as hot sauce and what looks like soy sauce.
Jars of pickles or relishes
Jars of jam or preserves
Packaged Goods:
Boxes of pasta (e.g. spaghetti)
Boxes of crackers
Bags of chips or other snacks
Bags of nuts or dried fruit
Packages of seasonings or spices
Other Items:
Some bottles of oil and vinegar
A carton of what might be chicken broth
A container of salt
One dish I could make by combining some of the things I see:
Based on the image, I could make a simple Pasta e Fagioli (Pasta and Bean Soup). I could use:
A can of diced tomatoes
A can of beans (I’m seeing several types that would work)
Some of the pasta from the boxes visible
A little chicken broth for extra flavor, though water would work too
Some of the seasonings or spices to taste
This is a hearty and easy meal that utilizes several items I see in the pantry. Of course, other options are possible, but the pasta and fagioli seem the most evident with what I see!
Ok this is pretty good. I can confirm that the pantry image I shared with Gemini 2 Flash, contained all those items. And Pasta e Fagioli is going straight into my recipe book. Usually when you ask an AI model for recipe ideas, it’s either convoluted, or asks you to find the eye of a newt. Neither works well in my tiny kitchen (which doesn’t have a pantry, for the record).
Verdict
I am EXCITED. This is a huge development. Google’s already rocking the AI space with tons of other developments yet to be announced. And when you pair its free tier API with VS Code and an extension like Cline, AI code editing is within reach of anyone, at no charge.
More on Gemini 2 Flash
Apart from Python coding, Gemini 2 Flash seems to be a pretty important contender across various workbenches.
Gemini 2 Flash debuted with Realtime Multimodal API which enables you to stream audio, video, and text in, while dynamic tool calls happen in the background (such as search, code execution, and function calling).
If you’d like to get super nerdy on the deets, there’s an amazing video here
And if you want to try it out yourself right away (please use responsibly, or otherwise), get on Google AI Studio — and kick its tyres, free.
Talk to me if you want to explore this in more detail.