ChatGPT, Canva and the hype problem
On expectations versus reality
Anyone with a ChatGPT Pro account has now got access to a Canva plugin. Canva has become the go-to tool for design-work-when-you-don’t-want-to-pay-a-graphic-designer. So a team-up with ChatGPT sounds great. And if an algorithm has even a sniff that you’re interested in generative AI you have probably heard about it this week. A TikTok asks me, “are graphic designers in trouble?”, YouTube informs that “ChatGPT can now design anything”, a tweet reveals that the plugin “makes graphic design 10x easier”.
There are two things of note here, the new plugin and the hype. First, the plugin. It’s available for all ChatGPT Pro users (other users are on a waitlist for all plugin access). You need to enable plugins in your settings and then hover on the GPT-4 logo to access the plugin store.
Once you have Canva installed any request for a visual result should offer elements from Canva. It’s one of those generative AI things, there are lots, that look very impressive on the first search. Mine was for a standalone image for this edition of the newsletter. Look at all those options!
But once you dig in it becomes clear that none of your prompts on color or wording are followed and all the results are templates that need to be edited in Canva. That’s not a bad thing (Canva is great!), but it doesn’t change what Canva can do for you. It will leave you exactly as good, or bad, at Canva as you were before. If you value keeping open tabs to a minimum, then I guess the plugin does have some value.
That goes for many of the 918, and counting, plugins in GPT Pro. It’s easy to imagine a future iteration where all your work is done within a chat tool and all your designs, or spreadsheets, or graphs, get amended through text prompts, with the chat tool figuring out the tricky part of how all these different tools actually work. But, so far, many plugins just paint a picture of the future, without delivering it now. And in fairness, neither OpenAI or Canva have given the plugin much publicity since it was added, it’s very much in beta, as is the whole plugin offering.
And a lot of this is based on what people think a large language model is capable of, not what it is capable of. ChatGPT remains a great way to get a chunk of text that can be researched and amended by humans until it’s presentable. It’s important to remember these are reasoning machines, not generating machines. But that’s not the pitch we’re frequently getting when reading about LLMs.
Which leads us to the hype. Anyone who has ever had the misfortune to follow an Apple launch knows that tech and hype go hand in hand. And that there’s a lucrative media ecosystem based on declaring every new thing to be a game-changer. But it’s gone up a level with the generative AI era. When you’ve put all your chips on AI the temptation to hype everything is real. I’ve read three different, very excitable, newsletter segments about the Canva plugin this morning alone.
Most users of generative AI tools really did have a ‘wow’ moment or two when first diving in over the past 12 months. But then most of us have also been underwhelmed by plenty on offer. Or discovered the really cool stuff comes after a tonne of work and learning. So Canva and ChatGPT, and many other plugins; nothing special now, another to watch for the future.
Small Bits #1: When is AI not AI
Great investigation by 404 Media this week. They uncovered a company promising AI-powered 2D to 3D conversions, but using human workers for much of the work. Since the story was published the Kaedim company website was rebranded to make it more clear that human input was used. There’s a strong chance this will not be the last company found to have misled about the extent of its AI input. So do you have to just rely on investigative journalism to assess any tools you’re thinking of using? No, simply Google the tool name and Reddit, it’s almost always the first place these tech red flags are easily found.
Small Bits #2: AI and policy
The News Media Alliance, an American advocacy group for news and magazines, published its ‘Global Principles on Artificial Intelligence’ this week. There’s been a necessary scrambling for some level of coherent thinking on using AI in publishing and, well, every industry this year. It’s worth a read if you’re trying to sketch out something for your own industry (or you could just ask us at Explainable). Under intellectual property the line “publishers are entitled to negotiate for and receive adequate remuneration for use of their IP” could apply to any creators.
Small Bits #3: Just can’t quit the Musk fanart
I might start an AI oddities corner segment for the newsletter. I promise not to make it all Musk-focused, but there is something endearingly/infuriatingly odd about the Musk and AI fandom crossover. Who do they think this guy is?!