Menu

Posts tagged “ai”

Link roundup for November 8, 2023

It’s been a bit quiet on the blog lately, so I thought I’d bring back the link roundup thing I used to do quite a bit. Here’s some stuff I read and enjoyed recently outside the regular product/business topics I usually write about here…


Very good summary of The OpenAI Keynote by Ben Thompson. This bit stood out to me:

The fact of the matter is that a lot of people use ChatGPT for information despite the fact it has a well-documented flaw when it comes to the truth; that flaw is acceptable, because to the customer ease-of-use is worth the loss of accuracy.


I’ve been following Craig Mod’s work for over a decade and know what to expect, yet his reflections on “Aloneness” took my breath away.

The real shitter is that if you’ve inured yourself to living in this state of aloneness, it can be difficult to break the habits that have led to it. Aloneness as default becomes comforting, and habits built around aloneness feel palliative because they’re known, and we tend to repeat familiar actions, even if they hurt us.


Great essay by Anne Helen Peterson on how friendship changes over time—including a period she calls “The Friendship Dip”:

Right now, the way our society is organized, we have a prolonged stretch of adulthood that is not conducive to forging or sustaining friendship or community. In many cases, I’d say it’s actually hostile to it.

7MqYrBBr8gSDreEKTjraPC 1920 80

Everything about the new U2 show sounds amazing. So sad I don’t have tickets.

Zoo TV had predated reality TV, fake news, social media—all these things. Bono had heard about this new venue in Vegas with nearly 20,000 seats, custom sound and an incredible screen that was akin to the whole audience having a VR experience. In the post-Covid era, it was appealing not to have to travel every night


Every new house in Portland uses this font for the house number, and now I can’t get this article out of my head. “The gentrification font: how a sleek typeface became a neighborhood omen”:

As Neutraface house numbers have become too commonplace to ignore, some now associate them (along with gray paint jobs) with neighborhoods overtaken by construction and renovations.


Feels like spam is about to get a lot harder to detect… “Inside the Underground World of Black Market AI Chatbots”:

We’ve got folks who are building LLMs that are designed to write more convincing phishing email scams or allowing them to code new types of malware because they’re trained off the code from previously available malware.


And finally, for my fellow Northerners… “How to light the dark months” has some excellent advice—not just the normal stuff we’ve all read a thousand times.

Lighting winter is an art and a daily practice, an act of survival and a gesture of love. Here are 10 ideas for fighting the gloom in the dark half of the year.

Everything Looks Like A Nail

Ed Zitron’s newsletter is kind of a hate-read for me because his vitriol knows no end and it can be a lot… but I think he did pretty well in his response to Marc Andreessen’s latest essay:

This is Andreessen’s dream—a continual race to the bottom where the tech industry is incentivized not to solve problems, but to find ways to make already-solved problems cheaper to solve so that venture capitalists can make money.

That’s a good quote, but please don’t stop there. The whole essay is the best rebuttal I’ve seen so far.

Unbundling AI

This is a thoughtful, well-argued essay by Benedict Evans about where we’re at with LLMs.

Whenever we get a new tool, we start by forcing it to fit our existing ways of working, and then over time we change the work to fit the new tool. We try to treat ChatGPT as though it was Google or a database instead of asking what it is useful for. How can we change the work to take advantage of this?

Uncovering a new class of responsibilities with AI/LLM

Since I prefer reading over watching, I appreciate Dave Rupert’s summary of this video about AI/LLM responsibility in his post Uncovering a new class of responsibilities:

Three rules of technology outline their nearly hour long talk:

  1. When you invent a new technology, you uncover a new class of responsibilities
  2. If that tech confers power, it starts a race
  3. If you do not coordinate, the race ends in tragedy

It’s those last steps that are the concerning ones. If we fail to respond to #1, we end up with #3.

Google Search's Death by a Thousand Cuts

Matt Rickard reminds us that it’s worth considering the long-term effects that putting public APIs behind paywalls might have on search engines:

Large models are trained on public data scraped via API. Content-heavy sites are most likely to be disrupted by models trained on their own data. Naturally, they want to restrict access and either (1) sell the data or (2) train their own models. This restriction prevents (or complicates) Google’s automatic scraping of the data for Search (and probably for training models, too). Google will lose results, site by site—it will be Google Search’s death by a thousand cuts.

You're in the right place

Here’s some great advice from Robin Sloan on how to find good educational content on YouTube:

These days, when I’m investigating a subject, I tend to go straight to Low View Count Scholarly YouTube, which is of course the version of YouTube you get when you append the term “lecture” to your search. When you hit a tranche of videos between forty and ninety minutes long, with between 500 and 5000 views, you know you’re in the right place.

This Google experiment is interesting and kind of related:

To make it easier for people to learn about topics they’re interested in, we’re experimenting with AI-generated quizzes on the YouTube mobile app Home feed.

The Homework Apocalypse

The Homework Apocalypse is an interesting post by Ethan Mollick on how educators can prepare for (and, to a degree, embrace) the incoming prevalence of LLMs in schools:

Students will cheat with AI. But they also will begin to integrate AI into everything they do, raising new questions for educators. Students will want to understand why they are doing assignments that seem obsolete thanks to AI. They will want to use AI as a learning companion, a co-author, or a teammate. They will want to accomplish more than they did before, and also want answers about what AI means for their future learning paths. Schools will need to decide how to respond to this flood of questions.

Link roundup for July 2, 2023

Technology and product

Pledge To Executives →

Marty Cagan’s latest is all about the agreements between product teams and executive teams. This point about deadlines stood out for me:

Product teams ask that only the product team that will be responsible for delivering on a promise be the one to make that promise, and they not be asked to make a promise or deliver on a commitment where they don’t know what is involved and what would be required to succeed.

How will AI affect workers? Tech waves of the past show how unpredictable the path can be →

A good piece by Bhaskar Chakravorti, also discussing AI’s impact on DEI in the workplace:

For example, while the broad shift toward remote work could help promote diversity with more flexible hiring, I see the increasing use of AI as likely to have the opposite effect. Black and Hispanic workers are overrepresented in the 30 occupations with the highest exposure to automation and underrepresented in the 30 occupations with the lowest exposure. While AI might help workers get more done in less time, and this increased productivity could increase wages of those employed, it could lead to a severe loss of wages for those whose jobs are displaced. A 2021 paper found that wage inequality tended to increase the most in countries in which companies already relied a lot on robots and that were quick to adopt the latest robotic technologies.

Also worth noting this discrepancy, which we seem to hear about a lot these days:

A 2022 study showed improved efficiencies for remote work as companies and employees grew more comfortable with work-from-home arrangements, but according to a separate 2023 study, managers and employees disagree about the impact: The former believe that remote working reduces productivity, while employees believe the opposite.

SparkToro Year 3 Retrospective: Investor Payback, Systemic Challenges, and V2 on the Way →

I enjoyed Rand Fishkin’s extensive and transparent thoughts on how their business is doing. A couple of things especially stood out. First, this point about marketing attribution:

In businesses like ours, most top-of-funnel marketing happens months or years before conversions do. When someone buys SparkToro, we have no way to attribute it to the three videos they watched on LinkedIn or the word-of-mouth recommendation from an ex-colleague at their previous agency, or the podcast they heard Amanda on last month. This would drive a lot of CMOs and CFOs bananas, but if you can lean into the process of trusting your “vanity metrics” (views, likes, comments, shares, emails, I-heard-about-you-ons), you can build a marketing flywheel that’s almost entirely devoid of competition.

I had to read that last sentence a few times to make sure it’s not a typo. This may be the first time I’ve ever seen someone speak positively about vanity metrics. Definitely food for thought…

And then there’s this important point about market segmentation:

Great products aren’t enough, either. To be “great” is, in my opinion, not nearly as valuable as being irrelevant to 99% of people, but exactly perfect for the 1% who deeply care about the problem you solve. Extra bonus points: target your product at a group that’s well-connected to others in their field, and gets value from sharing new things. Nothing’s better than word of mouth marketing. Nothing.

Other interests

The customers might be human, but the audience is Google →

This is a really interesting exploration of how “the SEO arms race has left Google and the web drowning in garbage text, with customers and businesses flailing to find each other.” Some small businesses deal with by having two websites: one for humans and one for robots.

How Google Reader died — and why the web misses it more than ever →

This is a really good history and retrospective of Google Reader. Dang, I feel for this team. It was so much more than an RSS Reader, and they didn’t even like that name. It was the first true social media feed: curated content you care about.

In other words, Fusion was meant to be a social network. One based on content, on curation, on discussion. In retrospect, what Shellen and Wetherell proposed sounds more like Twitter or Instagram than an RSS reader. “We were trying to avoid saying ‘feed reader,’” Shellen says, “or reading at all. Because I think we built a social product.”

Why aren’t smart people happier? →

Really interesting exploration by Adam Mastroianni, and a history of how messed up our definition of “smart” has become:

My grandma does not know how to use the “input” button on her TV’s remote control, but she does know how to raise a family full of good people who love each other, how to carry on through a tragedy, and how to make the perfect pumpkin pie. We sometimes condescendingly refer to this kind of wisdom as “folksy” or “homespun,” as if answering multiple-choice questions is real intelligence, and living a good, full life is just some down-home, gee-whiz, cutesy thing that little old ladies do.

Hometown’s Finest →

I’ve always been interested in “sense of place”—finding the reasons why a town or a place exists, and why people are drawn to certain places. Anne Helen Petersen writes beautifully about this concept in an essay about her hometown:

Optimization and remodel culture robs spaces of that heart. I’m sure MOD Pizza, the latest upstart in the pizza world, makes a lot more money. It’s slicker, faster, easier. But it’s not a place, it’s a product—a profit center. You can always tell, can’t you, when a restaurant’s primary purpose is to make a bunch of people who’d probably never eat there a whole bunch of money.

The Reader in Mind Is Me →

John Warner writes about the passing of Cormac McCarthy as well as Elizabeth Gilbert’s decision to indefinitely postpone the publication of her novel following the appearance of over 500 negative reviews of the book on Goodreads (also see How Goodreads Reviews Can Tank a Book Before It’s Published). He makes some interesting observations about “parasocial relationships”:

My first reaction was that we were in the realm of the parasocial, the invention or a relationship with a celebrity who doesn’t know you exist. My most parasocial relationships are with my favorite Peloton instructors who are clearly encouraged to stoke this feeling in platform participants as a way to keep us invested and involved.

Another example is Taylor Swift’s recent relationship with some other recording artist with bad politics and questionable hygiene, something her fans could apparently not countenance, and perhaps drove her to break up with the dude.

How generative AI might change the product profession in the future

I promise this isn’t going to become an AI blog, but Marty Cagan’s latest on Preparing For The Future has some solid points on how generative AI might change our profession. Here he talks about the impact on QA:

The new generation of AI-based test automation tools has the promise to revolutionize our approach to ensuring the product is behaving properly. On the other hand, our current approach to quality is largely based on deterministic products.  This means that given a set of inputs we can predict what the appropriate output should be, and we can count on that being true indefinitely.  

Yet for many new products built on generative AI, our products are no longer deterministic, but rather probabilistic.  We can no longer count on the same inputs generating the same outputs.  For many contexts, this is not a problem.  But for other contexts, especially when safety is involved, this will require different approaches to ensure appropriate behavior.

Also, yes please:

For those product managers of empowered product teams, the time spent creating artifacts such as written narratives, roadmaps, PRD’s, and acceptance criteria, as just a few examples, should be significantly assisted by the new generation of tools, some of which are already starting to appear. Even at the very rudimentary level, if the new generation of tools can significantly reduce the time a product manager (or engineer) spends dealing with tools like Jira, that will be a substantial win.

Evaluating AI product opportunities by plotting them on a “Survival Curve”

Aniket Deosthali (Head of Product for Conversational Commerce at Walmart) provides a great framework for evaluating AI product opportunities in How to Build AI Products People Want:

The most efficient way to evaluate AI opportunities and unlock the advantages of AI is by using the Consideration x Context framework. Let’s start with some baseline definitions.

Y-Axis = Consideration: The amount of effort required to make a decision.

The more thought you put into a decision, the higher consideration it is. For example, choosing a dish detergent is “low consideration” for most shoppers, compared to buying a car which is “high consideration.” Consideration can be represented as a function of the number of compelling alternatives and the stakes - users’ tolerance for errors, for example.

X-Axis = Context: The volume of abstract concepts AI needs to know. 

Context refers to how many abstract concepts a model needs to know in order to provide a useful response. Does it only need to understand a small batch of data points (like a product catalog), or does it need to understand the entire internet (like ChatGPT)?

He goes on to explain how to plot different solutions on a “Survival Curve.” If you work on AI products this is one of the rare actually helpful articles for PMs in the area of AI and LLMs.