AI In Your Toolbox - Issue #11
Conscious artificial intelligence is a dream and a nightmare. But before that becomes a reality, you might want to put some AI onto your tool belt.
Long conversations about the future often turn into conversations about AI. The rise of AI seems inevitable, especially with the recent advancements made in this space. Machine learning gets more approachable for engineers with toolsets like TensorFlow and PyTorch (ask your engineer friend about these tools, if they're actively tech-curious, chances are they have heard of them). And even more approachable AIs for the rest of the people like GPT-3 and DALL-E.
To see how far AI has come, you don't have to look that far into the past. You might remember DeepDream from Google. The results looked pretty much like something that is nightmare fuel.
From the picture above, you could've predicted that there's a long road ahead for AI. It would be harder to expect that only six years later, we'd see such an advancement that some image-generating AIs came with.
This week I got access to DALL-E, a tool that takes your written input and generates an image. The results are astounding because even if you ask for something bizarre, it does a decent job generating it (for as long as you're specific enough in your written input).
But beyond having fun with AI-generated pictures of silly things, what's also emerging are real-life uses of these tools. I particularly enjoyed this thread from Charlotta Blunárová that brings closer to the attention why this might matter way more than many people realize. She used DALL-E to generate a set of stylized icons.
The tools shape creations. DALL-E, in particular, can give tools to people with a bit of creativity (you still need some creativity to come up with good text prompts) to do many things that they might not be able to do themselves easily (in this case, 3D modeling). While these are impressive advancements, to get a result like this, you still need to do a lot of work yourself. You need to come up with good phrases that'd precisely describe what you seek to get from the AI tool. And the result might even require some editing after its generation to fit within your creation (color adjustments, visual edits, etc.).
Artificial intelligence that generates images easily spreads over social media. The results are very visually interesting, so people share them. What might not be as glamorous are other uses for AI that are no less impressive. A remarkable AI tool that crossed my path was Github Copilot, a tool for code editors that uses GPT-3 for making suggestions while you're writing code.
For the longest time, the vast majority of code editors have something called IntelliSense. It's a tool that analyzes your codebase, maps the code, and lists for you the available variables and methods. It can also show you parts of the code documentation that the developer wrote. Though, it's not intelligent. It doesn't tell you which method you might want to call; it just shows you what methods you can call.
On the other hand, Github Copilot can read your previous code and start guessing what you want to code next. The reason this is revolutionary is that this removes friction between the ideas in your mind and their execution. In my view, that's a huge advancement.
Many people thought that when AI comes in actual useful forms, it'll become challenging to get a job in the field that given AI is taking over. What's fascinating about the AIs I wrote about here is that neither of them completely removes the need for specialists. Instead, it helps specialists become more productive.
You still need to have a good eye, creativity, and some expertise to turn DALL-E into a tool that can help you with graphic design or art. And GitHub Copilot won't turn a junior engineer into a senior, nor a mid-level engineer. And neither of the tools can actually facilitate any creative process from beginning to end. They can only be within that creative process as a tool. Even for this letter, I had to curate the generated images & code and leave out times when it didn't work well. Experts aren't going anywhere anytime soon. But in my opinion, the ones that start embracing tools like this will benefit heavily.
I hope all this sparked your interest in the emerging forms of AI. I'm particularly intrigued by these advancements. I believe this is still just the beginning, yet I'm already benefitting daily from the AI tools available now. In the end, it's a lot of fun playing around with these tools regardless. So let me finish this letter with an attempt to make GitHub Copilot generate a joke.
As a bonus for this issue, feel free to respond to this email with prompts for DALL-E and I will generate them! It could be anything, but the more specific, the better.
Have a wonderful Sunday!
PS: I skipped an issue last week as I've been on a family vacation. By the time I was back to the keyboard, there wasn't enough time to sit down and deliver at least a somewhat quality post. Consistency is good, but I'd rather deliver something good rather than something just for the sake of it. Take care :)
PPS: By the end of writing this article, I also realized I'm using Grammarly to help me write this. While I'm not sure how much AI that tool uses precisely, it is a form of artificial intelligence nonetheless. And an indisposable tool in my toolbox in everyday life.
PPPS: While on the topic of AI, I own Google Nest Mini. It's not very smart, though, to be honest. But it's a nice little speaker. Here I modeled it in Blender when I was making a 3D model of my room.