I rejoinedSABR a couple of years ago and focused my volunteer work on fact-checking articles for the Games Project. These accounts of past MLB games are notable in some context of the author’s choice. They could be historically impactful, meaningful within a player’s career, highlighted by a rare event, or any other such factors that make an otherwise mundane day in baseball history something worth remembering.
Last month I decided to try writing one of these articles. I trawled through the archives of Minnesota Twins history for interesting seasons and landed on an early game in 2009 that defined the year for Jason Kubel.
I initially drafted this story as part of a broader writing effort related to The Last Question. All the idiocy happening in the US government and in the world of large corporations encouraged me to finish it.
I gave Claude (3.7 Sonnet) the same prompt I provided ChatGPT two years ago: Write a short article in the satirical style of The Onion, titled “Optimistic AI Just Happy to Be Here”.
I also went back to ChatGPT to see how it has improved.
Claude’s attempt.ChatGPT’s attempt.
This test encapsulates why I’ve stuck with Claude as the LLM I choose to pay for a few months at a time. It has an uncanny sense of style that ChatGPT entirely fails to capture. I found Claude’s article entirely believable and legitimately funny. It addresses my complaints from two years ago—it attributes quotes to individuals rather than nameless people with a title, and doesn’t hit us over the head with any of the jokes. It plays it straight, shares the story, and makes us laugh.
ChatGPT has slightly refined itself but still suffers from being too on the nose with its humor. SunnyBot-9000 is obviously an optimistic robot. That’s a case of nominative determinism if I’ve ever seen it. Compare that to Claude’s creation of IRIS-7, where the running gag is that nobody understands how this bot’s mien came to be. ChatGPT also leans heavily into LLMs’ potential inaccuracies, making it feel like it was written two years ago. Hallucinations still occur, I’m sure, but it’s not what it was upon release. Claude focuses on the tasks’ relative drudgery, which I find more compelling. The happy idiot of ChatGPT’s creation could be refined, but on a first pass, it’s rather dull.
Interestingly, they both used the same rough setup about deleting/resetting the LLM, which it then took in stride. While the punchlines were different, they both landed.
I talked about the characteristic ending of articles from The Onion in my post two years ago, and ChatGPT failed spectacularly yet again. It at least puts the sentiment in a direct quote from the bot, similar to Claude, but hoo-boy, is it objectively not a funny ending. It has the trappings of something that could be funny, but there’s no flair or subversion of expectations. It keeps pounding the same nail of “Live, Laugh, Love” with no unique edge.
ChatGPT has certainly improved compared to two years ago, but it’s impressive what Claude was able to churn out. Heck, Claude even got the rough formatting of “CAPITALIZED CITY NAME—” that begins every Onion article. ChatGPT couldn’t be bothered.
These are two essays by a couple of “guys on the Internet” whose work I enjoy. John Gruber created Markdown and now works in the Apple/tech media space. Merlin Mann used to be Merlin Mann, one of the first modern productivity gurus. Now, he’s essentially a comedic personality. Both are tremendous writers, and these two essays are supremely affecting and have unique styles that show the authors flexing their muscles.
Crankingby Merlin Mann, posted April 22, 2011. (About parenting and priorities.)
How It Went by John Gruber, posted November 8, 2024. (About the election, kind of.)
Well, I did it. I was on the ropes a few times but always found the time, energy, and creative hook to keep my story, It’s Like Jazz, moving along and hitting the requisite 50,000 words.
This has been a great experience so far. I had a great start, followed by a few rough days after a five-day trip to Denver. I caught up this past weekend and am back on track, though I’m looking to keep pushing hard because going home for Thanksgiving will only complicate my attempts to write.
Even with that said, I’m proud of what I’ve done so far. Compared to 2019, I’ve started building a better story. I’m more thoughtful about creating characters and pushing a plot forward. I certainly won’t have told a complete story by the end of the 50,000 words, but that’s not necessarily the goal.
This has been a good experience halfway through the month, and I hope to bring the same drive and enthusiasm to the rest of the challenge.
I plan to return to the basics for the first time since my initial attempt in 2019. I’ll be sitting at any number of devices—I have half a mind to resurrect my old ThinkPad that runs Pop!_OS and make it a dedicated writing computer—and writing a new novel wholecloth. I feel both intimidated and energized by the prospect. I’ve been working hard on other side projects, including writing blog posts in advance, to make sure I have the space to give this a good effort.
In light of the NaNoWriMo controversy and the near-dissolution of the official website among those who care, I’ll track my daily word count in a spreadsheet, which I’ll make available on the NaNoWriMo page of my website. In addition, I’ll link to a PDF file in Dropbox that I’ll update at least daily, as I’ve done in the past.
NaNoWriMo is about the perfect mixture of commitment and carelessness: You need the perseverance and motivation to keep pushing each day but the naivety not to second-guess yourself the whole way through. I’m intrigued to see what comes out of my head this time.