Wisereads Vol. 48 — The O'Reilly guide to Prompt Engineering for Generative AI, Charles Féval's work journal, and more
Last week, we shared the entirety of A System for Writing, Bob Doto's guide to capturing ideas and fueling writing with a zettelkasten. This week, we're sharing a preview of James Phoenix and Mike Taylor's new O'Reilly guide, Prompt Engineering for Generative AI.
Keep reading to add to your Reader account below 👇
Most highlighted Articles of the week
Use A Work Journal To Recover Focus Faster And Clarify Your Thoughts
After his initial hesitation, Microsoft principal engineer Charles Féval started a journal to help refocus after inevitable workplace distractions and IM pings. "I’ve just been too lazy to ever do it. Or not necessarily lazy, but more: I didn’t trust the tool enough to think it was a good use of my time, and instead just mash on the keyboard till it works. After all, I’m writing pages of text, of which I will never read more than a fraction. But that’s not the point. The point is structure, and the point is caching."
Robert Putnam Knows Why You’re Lonely
Robert Putnam, author of Bowling Alone, underscores how joining a club can create social capital and support democracy by connecting dissimilar people. "Ties that link you to people like yourself are called bonding social capital… And bridging social capital is your ties to people unlike yourself. So my ties to people of a different generation or a different gender or a different religion or a different politic or whatever, that’s my bridging social capital. I’m not saying 'bridging good, bonding bad,' because if you get sick, the people who bring you chicken soup are likely to reflect your bonding social capital. But I am saying that in a diverse society like ours, we need a lot of bridging social capital."
Panic! at the Job Market
Matt Stancliff rages at the convoluted world of tech hiring and coding interviews, hoping his experience speaks for itself as unemployment rises. "What used to be as simple as 'i good wit commputr. u giv jorb?' is now a synthetic convoluted social status driven hierarchy of mind games just to get an initial interview then you are treated as a blank slate having to prove you can even read and write and speak from first principles."
Most highlighted YouTube Video of the week
AI Art Explained: How AI Generates Images (Stable Diffusion, Midjourney, and DALLE)
Researcher Jay Alammar provides a layman-friendly explanation of image generation, including Stable Diffusion and training methods. "What we did here is give it an example of an image and its own caption, but we can also give it another caption of another image and force it to say that these are not similar. We need both positive examples and negative examples—that's an idea called contrastive learning which is a very important idea in machine learning."
Most highlighted Twitter Thread of the week
WARNING: this is a long ass post
After selling Morning Brew for $75 million, co-founder Alex Lieberman is well-equipped to teach the ropes of building a newsletter business. His steps include developing a content strategy with an audience in mind: "Who is the one person you’re writing this newsletter for? Describe them in excruciating detail," and creating a team: "There are three legs to the newsletter stool: Content, Growth, Monetization. There needs to be an owner for each leg of the stool (but the same person can own multiple legs)."
Most highlighted PDF of the week
SpreadsheetLLM: Encoding Spreadsheets for Large Language Models
Because the complex grids in spreadsheets don't lend themselves to the sequential processing of LLMs, Microsoft researchers to developed an encoding method called SheetCompressor. "This framework effectively addresses the challenges posed by the size, diversity, and complexity inherent in spreadsheets. It achieves a substantial reduction in token usage and computational costs, enabling practical applications on large datasets."
Hand-picked book of the week
Prompt Engineering for Generative AI
While naive questions work for one-off interactions with LLMs, regular or programmatic inquiries are more cost-effective and reliable in response to precise prompts. Enter: prompt engineering.
In their new O'Reilly guide, Prompt Engineering for Generative AI, data engineer James Phoenix and technical marketer Mike Taylor go into depth on the conventions of prompt engineering, including giving direction, specifying format, evaluating quality, dividing labor, and providing examples.
"When briefing a colleague or training a junior employee on a new task, it’s only natural that you’d include examples of times that task had previously been done well. Working with AI is the same, and the strength of a prompt often comes down to the examples used. Providing examples can sometimes be easier than trying to explain exactly what it is about those examples you like, so this technique is most effective when you are not a domain expert in the subject area of the task you are attempting to complete."
James and Mike are generously sharing the first chapter of their guide, packed with examples and use cases, with Wisereads readers. If you want to level up your LLM usage, we highly recommend purchasing a full copy to support them here. Plus, it's perfect timing to put your improved prompting skills to use with Reader's new custom prompts via Ghostreader V2.
Handpicked RSS feed of the week
Nedra Nuggets
Author and therapist Nedra Glover Tawwab shares tidbits on mental health and relationships in her Substack, Nedra Nuggets. From Owning Our Mistakes: "There is something about being seen as wrong, unknowing, or not having it all together that makes us come undone. It makes us refuse to accept accountability even when something is our fault. This knee jerk reaction of 'I didn’t do it,' is reminiscent of the childhood reflex when we thought we were going to get in trouble. This programming from our youth seems to sometimes bleed into adulthood, but we need to let it go."