Phil Simon's Blog, page 3

Battle of the LLMs

My friend Alan and I often test AI tools on their sports knowledge. We find it a fun way to see how advanced contemporary large language models have become. In today's post, I quiz a few popular ones on their knowledge of an insanely valuable sports franchise.

Photo by Tim Hart / UnsplashBeginning Prompt

Yes, I'm talking about the LA Lakers:

Initial Claude Prompt: | Click on the image to enlarge it.

Claude initially returned a bunch of details on the careers of Kareem, Magic, Kobe, and a few other legendary players. Yes, each is a member of the Naismith Memorial Basketball Hall of Fame, but I didn't want details. I quickly refined my prompt:

ChatGPT-5 Still Makes Egregious Errors

Like so many folks, I was curious about the launch of OpenAI's latest and greatest large language model.

In today's post, I'll explore how OpenAI's self-described expert-level tool performed on some simple image tweaks. Would it uncover major issues or pass with flying colors?

Let's find out.

Moving From WordPress to Ghost: The Verdict

A little over a month ago, this site moved from WordPress to Ghost. (Read the post about my motivations here.) Today, I'll offer some early thoughts on the Ghost experience so far. Long story short: It's been exceptionally positive.

Visual Editor

A few quirks aside, I've embraced Ghost's editor. It's clean and insanely fast. The ability to save snippets for future use saves a great deal of time. The blocks specific to emails and newsletters make sense. Conditional logic, baby.

Maintenance

Minimizing the amount of time I spent fixing bugs, messing with caches, and the like was a primary goal of moving to Ghost. Mission accomplished.

A More Integrated Reader Experience

Over the past few years, I've often visited an author's website and noticed something odd: it almost immediately encouraged me to go to Substack, Beehiiv, or someplace else.

Exhibit A: David Epstein—a truly gifted writer whose work I enjoy. Ditto for Dan Pink and scores of others. I wonder if I'll miss out on one of their articles because I don't subscribe to both their blogs and newsletters.

By using Ghost as a single distribution channel, I've eliminated that source of potential reader confusion. They need not wonder:

Where I publish different articles. How I delineate among blog posts, articles, or newsletters.If they're missing anything.

Put differently, if I publish something, it's here. Period. (In the future, I may create separate newsletters if I want to segment my audience.)

Photo by Sergey Kuznetsov / Unsplash

Two months ago, that wasn't the case. My content was fractured across different websites. Different people saw different things—and I couldn't easily determine which posts were popular.

The Move to a Subscription Model

I'll be honest: It's been an adjustment not just publishing content for all to freely read. I suspect that my decision to adopt a membership model has in part contributed to a slight decline in website traffic. No doubt, the explosion of genAI tools is also culpable. On the plus side, at least now I know who's consuming my content and becoming a true fan. (As an added benefit, tech companies cannot gobble up new posts to further train their AI models.)

I couldn't be happier with the resutls.

In the weeks after my site relaunch, a few people expressed confusion about the posts they could view. Hopefully, the new emojis next to each one make it clear. From my FAQ:

I see an emoji near a blog post. What does it signify?

🌍 = Public content that anyone can view without signing up for any plan.

🎁 = Reading the whole post requires creating a Free account & logging in to it.

➕ = Reading the whole post requires creating a Plus account & logging in to it.

⭐️ = Reading the whole post requires creating a Premium account & logging in to it.

Along these lines, the revamped Index page lets people see the posts associated with each tier.

Index Page: | Click on the image to enlarge it.

I'm still noodling with subscription tiers. This much I know: I haven't quite nailed the right price and features for each one. I suspect that I'll tinker with them over the next few months.

Ghost Support

In a word, it's fantastic. On a few occasions, I sent the folks an email and found their responses genuinely helpful. And I love Ghost's glorious alerts.

I Asked AI How My Writing Has Evolved Since 2009. Here’s What It Found.Claude weighs in on my nearly 1,500 blog my posts. Its findings were scary accurate.Phil SimonPhil SimonParting Thoughts

Ghost isn't the best alternative for every WordPress user. For me, though, it made sense to pull the trigger when I did. A little more than a month later, I'm ecstatic with my decisions to move to it and to hire Spectral Web Services.

I expected a few bumps during the migration. I'm still squashing the random bug, but the time and money spent were well worth it.

Finally, I've written more posts on my site in the last month than I have in the past year. The simple reason: Ghost has rekindled my love of blogging. I'm no longer battling a bloated WordPress theme and legacy tech. Less friction means more time getting my ideas out there.

🐜Finding any problems with my new website? The launch went very well, but there's no way that we caught all the bugs. Let me know if you find any issues or would like to provide any feedback.

Will AI Make Us All Line Judges?

🎤Excerpted from my keynote talk, Proceed With Caution.

In June of 1981, John McEnroe arrived at Wimbledon on a mission. The talented but volatile American tennis prodigy was hell-bent on avenging his epic, five-set finals loss from the previous year to his longtime rival, the stoic Swede Björn Borg.

The All England Club had announced the tournament’s draw. McEnroe—the second seed—would play his first-round match on Centre Court. He faced fellow American southpaw Tom Gullikson on June 22 of that year.

As it turns out, that date would live in tennis infamy.

Four Words That Would Define a Career

Early in the first set, McEnroe served from the deuce court. The lefty hit an apparent ace up the tee. Chair umpire Edward James waited a few seconds before ruling the ball out.

The questionable call infuriated McEnroe—and he was not shy about expressing his frustration. Tournament officials, fans, and local journos had seen this movie before. The British tabloids had previously dubbed him Superbrat for his vituperative on-court antics.

Mac didn't want to disappoint—or didn't care. He screamed at James, ultimately uttering perhaps the four most famous words in the history of the sport: You cannot be serious.

During his diatribe, Superbrat also called James “the pits of the world” and an “incompetent fool.” Before resuming play, he took some shots at tournament referee Fred Hoyles. McEnroe’s actions earned him a point penalty and a $1,500 fine for his unsportsmanlike behavior.

The Nvidia Arrow

On July 11, 2025, Nvidia became the first company to reach a $4 trillion valuation. It famously allows its employees to work from anywhere.

Put those two facts together, and you may be inclined to think that Nvidia's wealth allows it to adopt flexible work policies.

But what if the arrow went the other way?

The AI Goldilocks Test

An MIT study from June 2025 on the use of AI has caused quite the stir. TL;DR: Some evidence suggests that AI is making us stupid. I'm reminded here of Nicholas Carr's 2011 book The Shallows: What the Internet Is Doing to Our Brains.

So you should ignore AI altogether, right?

That's a dangerous proposition. Fortunately, just like with Goldilocks's three bowls of porridge, you can find a happy medium with AI. Let that sentence serve as the starting point for today's post.

Photo by Peter Herrmann / Unsplash

One AI Tool, One Simple Question, Two Different Answers

On July 3, 2025, I was watching Wimbledon when I noticed that Marin Čilić was scheduled to play in the second round. It turns out that the big-hitting Croatian hadn't hung up his racket yet. (Čilić had upset lefty Jack Draper, the fourth seed.)

I've known for a while that Čilić was on the back nine of his career. (Pardon my mixing of sports metaphors.) How much money, I wondered, had he earned on the court?

Let the Research Begin

Could I ask AI? Of course, but a simple query didn't interest me all that much. Readers of my site and books know how much I like geeking out on data visualizations. #foreshadowing

I'd like my answer to be correct, and I'm still wary of AI hallucinations. I'd feel better if I prompted AI with a trusted source of sports information—in this case, ESPN.

On Thermostats and AI Hallucinations

I was wrapping up my workout the other day when I struck up a conversation with the manager at my gym. Long story short: I threw out a quote, but I forgot its source.

That bothered me.

The manager knew the quote, but he didn't remember who said it either.

Since I'm a curious dude, you probably know where this story is going.

AI to the Rescue?

I returned home and asked Claude:

Claude Response #1 | Click on the image to enlarge it.

Seth is a smart dude whom I even interviewed a long time ago. I knew for certain, though, that the original quote didn't belong to him. I strongly suspect that Claude attributed the words to him for two reasons:

His blog is extremely popular.He has uttered those words many times.

The Forthcoming Wave of AI Mop-Up Jobs

The Irish playwright George Bernard Shaw once famously quipped, “Beware of false knowledge; it is more dangerous than ignorance.”

From the moment I read the quote, I loved it. It even adorns the dust jacket of Message Not Received.

Shaw was speaking in general terms—not about ChatGPT, Claude, Gemini, or Perplexity. Make no mistake, though: today's breed of generative AI tools makes people think they can accomplish things they just can't. In this post, I'll connect the dots. AI hallucinations will result in a tsunami of clean-up work for true subject matter experts.

Photo by Edi Libedinsky / UnsplashLooking in the Mirror

Let's start with a personal anecdote.

In July, I moved my website from WordPress to Ghost. I performed a good bit of the work myself, but I couldn't have done it alone. I hired a true Ghost guru.

The Harold Baines AI Hallucination