Gemini: "Create a cartoony banner image of two hands. One should show a thumbs-up, the other a thumbs down. Keep is simple with no frills, no blemishes, and no weird faces."
*sigh*
While I'm still on the fence about the wide use of generative artificial intelligence ("GenAI") across consumer products, I know it's here to stay, so I'm definitely leaning in to see how I can make the best of it. Ultimately I think GenAI is a great new tool, but like all great tools, it only goes as far as the operator knows how to use it. A few examples:
The Good Tool: asking questions to validate what you already know
- ... and, this is fine! Most days ChatGPT is just a glorified search engine anyway
The Bad Tool: asking for medical advice (I can't believe how many people are doing this based on anecdotal evidence)
- ... but, uploading a picture of that mysterious lump on your leg misplaces the importance of situation-specific expertise that doctors provide
The Ugly Tool: outsourcing critical thinking
- ... FFS, asking questions and blindly accepting the answer is dangerous to personal and professional growth
You know what? Maybe my disappointment isn't the tool itself, but how people are using it.
What Works...
In the last few months of my tinkering, I have found a few great uses of GenAI:
- Simplifying topics: while some context gets lost on occasion, GenAI does a spectacular job of helping me understand the basics of just about any text I throw at it; this really helps with coding too - what the hell did I write six months ago?
- Proofreading: being able to get instant feedback on my writing has really changed how I create content for work and conferences; 100% of my conference abstracts this year have been proofread by GenAI and are better because of it
- Brainstorming: a prompt like "What are some possible topics I could write regarding my disallusionment of GenAI?" provides great alternative ideas when writing certain types of blog posts (although this one is actually 100% original)
- Visual understanding: when compared to traditional ML-based image identification, GenAI seems to be really good at telling me what's in a picture and why it matters; for example: I've discovered that most of my backyard is honeysuckle and poison ivy
- Drafting images: I'm not a great artist, but I can quickly create all kinds of images and graphics with a few simple words, which let's me focus on everything else that I'm good at (like writing) - although I'm not sure why GenAI likes to add faces to things (see banner image)
...And What Doesn't
On the other hand, GenAI quickly falls apart for me in a lot of ways:
- Writing longform content - writing a coherent five-paragraph article from a simple prompt sure is neat, but do I really want to read something written by a machine? When I'm dedicating time to read long-form content, I'm looking for thoughtful opinions and impactful perspectives that comes from a real person
- The forever pollyanna: GenAI has an optimism issue - "what a great topic!" is the type of response I get for so many prompts; I want answers, not opinions (I talk to people for opinions)
- Summarizing lots of content: in theory, summarizing long meetings or lots of content seems like it should be a great use of GenAI, but too often important details get left out or context is missing
- Double-searching: when I've tried to use GenAI as a search tool, I find myself having to double-check every response due to potentially incorrect information (so generously labeled "hallucinations"); this means every GenAI query ends up with another Google search anyway - now I'm just doing twice the work and saving no time
The Too Easy Button?
I think GenAI impresses a lot of people because it lets them exceed their own skills with a simple prompt, which is either high praise for how far GenAI has come or a critism of a lacking skilled populus ("I can't code, but now I'm a developer!").
My biggest fear with today's GenAI is pretty simple: that it will be deemed "good enough" because "most people" are blown away by the gee-whiz factor of watching a computer do the hard things they cannot - and then not challenging the results when they feel off.
We're supposed to be on the cusp of artificial general intelligence (AGI) any day now, but I suspect this is going to be our white whale for the foreseeable future. GenAI is really good at creating content based on its understanding of existing patterns, but I have yet to see AI have a thought of its own - and that's the real magic that everyone should be waiting for.
I'm not holding my breath for AGI, so for now I am learning to make the most of what GenAI can do while being very cautious of the things people think it can do.