I still feel weird about how prompt engineering is promoted as a new concept.
Prompt engineering is a technique or discipline in AI where the prompt (what you ask a model to do) contains a description of the task you want it to complete. That part is fine and features all the advantages of single/few shot learning. Great. But the rhetoric around it is full of hyperbole.
Prompt Engineering And Why It Matters To The AI Revolution
Is ‘Prompt Engineering’ the Future of AI?
Prompt Engineering: The Career of Future
The demise of blogging and rise of the prompt engineer
I think AI and data science have bigger problems to solve than whether prompt engineering is the way to sustain a career. Again, I have nothing against the concept itself, just how people are claiming its potential makes it a surefire hit for individuals and industries. A lot of the time, I see people making educated guesses and turning them into blog posts about “cool prompt engineering tips”. For people who aren’t directly in the field, that’s all we’re doing. If that’s a hot take, call me Luke Warm. Heh.
A lot of this discourse reminds me of those marketing “experts” who conduct studies on a handful of websites (1 million out of the 2 billion+ online) and claim that results show what Google uses as ranking factors. The key link is the idea that a governing body of sorts has a black box where all the algorithms live and only the clever few with knowledge in that area can tease answers out of it. Perhaps we’ll see these “gurus” (fwiw, I despite that term and wish people would stop using it unironically) pivot towards prompt engineering to make themselves look relevant and go Super Saiyan Expert 6.
Or maybe we could fix the models and stop pretending like AI models are magicians hats. Okay, that’s the last analogy I’m using. This stuff is already complicated to explain to laypeople as it is (myself included).