#AI#GenerativeAI#LLMs#PromptEngineering#TechnicalWriting#SoftwareDocumentation#Productivity: "Many tech writers have a constant fear that AI will take our jobs. I often think, what I’m doing isn’t rocket science. Any person with some education can do it. And yet, just as engineers struggle to write, tech writers frequently struggle with AI tools. They don’t understand how to use them effectively. Even though “prompt engineering” is often a ridiculed term online, again and again I hear feedback from TWs about AI not being useful to them, or they simply don’t have interest in AI, as if it’s irrelevant to their work. This blows me away. When I can ramp up on a product in an hour and write a user guide in a couple of days, and code a doc publishing script that automates even more tasks, how can AI not be useful? How can it not be essential?
An often repeated saying is that AI tools won’t replace us, we’ll be replaced by those who know how to use AI tools. I feel like this is more and more true. Consider this scenario: You hire a roofer to install a new roof, which mainly involves removing the old shingles and installing new ones. One roofer arrives with a hammer. It will take this roofer 2 weeks to do the job. Another roofer arrives with a pneumatic roofing nailer power tool. It will take this roofer 3 days to do the job. The cost of the first roofer is 4 times that of the second. The output is pretty much the same. Which roofer do you hire?
It’s the same with tech writers. Suppose you have a large project. One tech writer can create the documentation using AI tools in a quarter of the time, while the other will take 75% longer. Which tech writer do you hire?
Fortunately, I think tech writers can learn how to use AI tools as power tools. Especially with more awareness and knowledge about effective prompting techniques, tech writers can become much more productive using AI." https://idratherbewriting.com/blog/ai-is-accelerating-me
I just realized that #AI “prompt engineering” is just SEO (“#SearchEngine optimization”) in reverse.
#SEO spams unique language into specific content in hopes of raising the latter’s rank when querying a corpus indexed by an opaque non-deterministic algorithm.
#PromptEngineering spams unique language into a query against a corpus indexed by an opaque non-deterministic algorithm in hopes of returning specific content.
I also just realized a striking similarity between the total absorption of certain current programmers’ time with #AI#PromptEngineering and previous generations’ occasional obsessions with #CellularAutomata models like Conway’s Game of Life.
Both activities involve long hours playing Aristotelian Prime Mover, tweaking a system’s initial inputs while inducing larger conclusions about its emergent behavior. And both groups Will Not Shut Up about it.
#AI#GenerativeAI#LLMs#PromptEngineering: "There is an alternative to the trial-and-error-style prompt engineering that yielded such inconsistent results: Ask the language model to devise its own optimal prompt. Recently, new tools have been developed to automate this process. Given a few examples and a quantitative success metric, these tools will iteratively find the optimal phrase to feed into the LLM. Battle and his collaborators found that in almost every case, this automatically generated prompt did better than the best prompt found through trial-and-error. And, the process was much faster, a couple of hours rather than several days of searching.
The optimal prompts the algorithm spit out were so bizarre, no human is likely to have ever come up with them. “I literally could not believe some of the stuff that it generated,” Battle says. In one instance, the prompt was just an extended Star Trek reference: “Command, we need you to plot a course through this turbulence and locate the source of the anomaly. Use all available data and your expertise to guide us through this challenging situation.” Apparently, thinking it was Captain Kirk helped this particular LLM do better on grade-school math questions.
Battle says that optimizing the prompts algorithmically fundamentally makes sense given what language models really are—models. “A lot of people anthropomorphize these things because they ‘speak English.’ No, they don’t,” Battle says. “It doesn’t speak English. It does a lot of math.”
In fact, in light of his team’s results, Battle says no human should manually optimize prompts ever again.
“You’re just sitting there trying to figure out what special magic combination of words will give you the best possible performance for your task,” Battle says, “But that’s where hopefully this research will come in and say ‘don’t bother.’ Just develop a scoring metric so that the system itself can tell whether one prompt is better than another...”" https://spectrum.ieee.org/prompt-engineering-is-dead
“However, new research suggests that prompt engineering is best done by the model itself, and not by a human engineer. This has cast doubt on prompt engineering’s future—and increased suspicions that a fair portion of prompt-engineering jobs may be a passing fad, at least as the field is currently imagined.”
Conventional wisdom holds AI can't write anything personal or original. In today's post (Magnifying Ideas and Expanding Text with AI) I do just that, showing you how to get AI to write something both personal and original. https://sadlynothavocdinosaur.com/posts/magnify/
Most AI tools have limitations on the amount of text they can process at once.
Large inputs can hurt accuracy, cause memory issues, or lead to the AI getting “lost” in the complexity.
Proceeding in smaller chunks allows you to course-correct issues along the way for a better end result.
Precision: Smaller prompts give you more control over the AI’s output. You can provide tailored instructions for each step, directing the final result with greater accuracy.
Understanding how to break down complex tasks is essential for interacting effectively with AI tools. Even outside of prompt engineering, we do this naturally in many settings. Writing documentation often starts with a simple outline, expanding into greater detail over time. Similarly, a complex codebase is built through incremental steps, adding components over time. Learning to identify those smaller steps within a complex problem is fundamental to working with AI tools.
In this tutorial, you’ll learn to create a complex tree diagram (in ASCII text) from a JSON block. Breaking this task down into smaller steps will make it easier for the AI to process and will ultimately yield a more accurate and visually appealing diagram."
"Aftermarket" fixes applied after training, like injecting diversity terms into prompts, don't fix the underlying model and can even exacerbate harmful fabrications. If the training set is biased—and the Internet is—it's really hard to correct that after the fact.
As a teacher, I've come to understand the utility of new tech isn't always obvious until it becomes ubiquitous, until you have a chance to truly play with it. "Why on earth would I want a video projector in my classroom?" I thought at the turn of the century. Now, I couldn't live w/o it because it turned the "teacher's computer" into a "classroom computer." That's why I'm playing with LLMs, to discover the possible. You should join me. https://sadlynothavocdinosaur.com/posts/tldr-post/
Next week I'll be starting a pretty ambitious project—50 Days of LIT Prompts. Every weekday for 10 weeks, I'll be sharing prompt patterns along with my thoughts and readings relating to Large Language Models like those behind #ChatGPT. Follow the link below, and this thread, for updates: https://sadlynothavocdinosaur.com/posts/50-days-of-lit-prompts/
"...everyone's been handed a telescope [LLMs], and they keep looking through the wrong end... Instead of writing prompts with 5 words and expecting 500, more folks should be providing 500 and asking for 5."
What if I you could select a word or phrase, click a button, and get a definition, be it for a word, idiom, or initialism without leaving the page you're on? Well, I'm happy to say today's prompt template does just that.
It also starts us down the path of understanding how a neural net works!
Sometimes, I imagine the HAL 9000 answering emails on my behalf simply by saying, "I'm Sorry, Dave can't do that." So, today's title was a no brainier.
I'm Sorry, Dave Can't Do That: [Have AI] Use the text of an email to draft a polite reply declining any request(s)¹
I figured generating a headline is kind of the apotheosis of this week's prompts. I mean, ideally, it feels like a headline is a distillation of a text's essence.