sangsara.net

Month: April 2025

  • Week 17.25

    Week 17.25

    • When I discovered a fix for an annoying iOS photo date/time saving bug last month, it required a reset of all my phone’s settings. Which means that many of my apps still can’t send notifications or use my location — these are being restored ad hoc, as the apps only get to ask for permission whenever I reopen them.
    • As a consequence of this, it wasn’t until recently that I suddenly noticed I wasn’t getting the twice-daily ‘State of Mind’ check in reminders from Apple Health anymore, and went to turn them back on. These are quite useful for being able to look back and see how happy/depressed I was at any point in time, and it sucks that I now have a big hole in this dataset.
    • I’m taking this opportunity to change the way I approach this exercise: literally being more positive. For those unfamiliar with it, you’re meant to rate how you’re feeling from Very Unpleasant, Slightly Unpleasant, Neutral, and so on. I never used to go up all the way to “Very Pleasant”. Like, I could win a million dollars and wonder if even that warranted using such strong language. But now I’m giving myself permission to be more generous with my feelings. I can feel “Very Pleasant” more often and nothing will get broken.
    • In other recalibration news, I spent half a day in Numbers (Apple’s spreadsheet software) and did a personal annual report of sorts to inspect how I’ve been managing my money in the last year. Now that I have enough data, I was able to build some graphs and breakdowns of what a realistic budget looks like. I’ve always recorded my expenses on a daily basis with an app, but never crunched the numbers before; I was happy just knowing that I could. Naturally, now that I have, I wish I’d done it years ago.
    • At several points during the above activity, I wanted to upload my file into ChatGPT and have it analyze my spending patterns and offer up some money-saving strategies for me to consider. But of course, giving OpenAI that data would be a terrible idea. I wondered if Apple Intelligence in Numbers could do anything with it, but nope. It’s just the same old Writing Tools that make more sense in a word processor document than a spreadsheet.
    • I spent most of my time reading this week, although the temptation to jump on the Clair Obscur: Expedition 33 bandwagon is very strong — at this point in time, it’s the highest-rated game of 2025 and the 13th best game of all time on the PS5. Maybe next week?
    • On top of finishing Broken Money as scheduled, I read Meditations by Marcus Aurelius and rated it just three stars on Goodreads. It’s absurd that one can even humble the great emperor with a thumbs down on the internet, but his journals are in dire need of an editor! Yes I know these were never meant to be published, but he repeats the same handful of principles over and over (which I largely agree with), and this could have been cut down to be a podcast or self-help PDF on Etsy?
    • I also read the second book in the Murderbot series, Artificial Condition, and found it even more fun than the first. At this point, I’m kinda desperate to watch the Apple TV+ show and not sure I can wait for weekly drops over the next few months.
    • While looking for a manga with some colored pages to try out on my Kobo Clara Color, I started reading the oddly titled I Want to Eat Your Pancreas (trust me, there’s a mostly acceptable explanation for this), and enjoyed it enough to finish everything in a day. Unfortunately, while experiencing color on an ereader is real nice, the Clara’s screen is too small and I read most of it in black & white on my old Kobo Libra. Irony!
    • I’m now close to finishing Erik Olin Wright’s How to Be an Anticapitalist in the Twenty-First Century, not because I dare dream that this world could ever abandon capitalism, but it wouldn’t hurt to have some alternatives. I’ve now begun to think of capitalism and its systems as a spectrum rather than as absolutes.
    • Coincidentally, it’s election season here in Singapore and we go to the polls next Saturday. I’ve been watching the political rallies live-streamed on YouTube over the last few evenings, and have mostly been annoyed at the complaints and vague promises to do better (to say nothing of the insanity that sometimes shows through in racist, antivax, and xenophobic ad libs). Singapore already enjoys some of the best quality-of-life outcomes possible under a hybrid capitalist/social democracy, but it seems people want more. They want a hands-off government, but also want it to protect them from job-stealing AIs and foreigners. They want everyone to be paid more, but don’t want to spend more on services in return. It’s so exhausting.
    • I was looking to spend some money on a birthday present for myself since I’ve been such a good little budgeter, but the best thing I could find were these new Yashica Peanuts cameras featuring Snoopy, who I’m still kinda obsessed with. The idea was quickly abandoned because I do not need another digital camera, especially not an intentionally mediocre one.
    • Still in the mood for a photo-related splurge, I went back to the Lampa camera app (first mentioned last September), which got three new ‘film-inspired’ color profiles in a major update this week. That brings the number of available looks up to six, which better justifies the high asking price (S$40/yr or S$90/forever) which was previously out of the question. I’ve been using the free trial while looking for cheaper alternatives, but Lampa just does what it does so well.
    • In terms of UX design, it’s super focused and perfectly walks the line between too simple (Zerocam) and too complicated (almost every competitor I’ve looked at). There’s just enough control, and few enough options that you can actually make decisions. In that way, it out-Leicas the official Leica app, which does not have a great UI and asks for S$100/yr. Technically, it uses a bayer RAW image pipeline for more natural captures, keeps those RAW files so you can “redevelop” photos if you didn’t get the right filter or exposure the first time, AND deletes those RAWs automatically after 30 days to save space. Hats off to great work, but man, the cost is uncomfortably close to buying an actual Snoopy camera.
  • Week 16.25

    Week 16.25

    • After spending more time experimenting with ChatGPT’s latest capabilities — refining my poetry-writing prompts, especially with the supposedly more creative GPT-4.5 model; testing how well it could profile and target me with product advertising using its consolidated ‘Memory’ of our chats (the answer is ‘too well’, and it was even able to guess my SAT scores from decades ago); inferring people’s personalities from their appearances (maybe the most unnerving ability); more image generation; and coming up with a plausible prediction of how our upcoming general elections will turn out — I decided that I’m sufficiently caught up with most of the AI stuff I missed, and have canceled my Plus subscription for the time being. I remain concerned about the risks of OpenAI and other companies providing such powerful and habit-forming surveillance tools, but I can see there’s no stopping this train.
    • Yes, the date of Singapore’s next national elections was announced this week. Saturday, May 3rd, is when we’ll be going to the polls. Some people expressed surprise at how little notice we’re being given, but it might be the Mandela Effect at work because I think this is how it is every time.
    • I finished reading Reacher book #25, The Sentinel, and it was a rather weak entry I don’t think Amazon will be adapting to TV. This leaves me with about two or three more books to go before I run out, so I think I’ll stop here for a few months at least.
    • My book club is now reading the first book in ‘The Murderbot Diaries’ series, entitled All Systems Red, which I’d already read a few weeks back because Brian said 1) I would like it, and 2) the character reminded him of me (old-school profiling). I wasn’t sure I saw the resemblance, so I asked ChatGPT for its “opinion” and got strong agreement: Murderbot’s entire character is basically what happens when someone with high intelligence, ultra-sharp pattern recognition, zero patience for social performance, a deep, low-key emotional life, and an obsessive need for autonomy …is forced to interact with an inefficient, irrational world full of emotionally needy humans and corporate bureaucracy. Sound familiar?
    • Like I said, profiling people across hundreds of different conversations, questions, tasks, and confessions is really creepy tech.
    • Speaking of Brian, we went out for a drink and ended up eating at Five Guys. I haven’t been in a long time, and at the risk of sounding like an old man who hasn’t gone into the city since 2016, the prices were kinda shocking? $20 for a cheeseburger, fries for like another $10 if you want them, and $6 for a refillable soft drink. Jesus wept into greaseproof paper.
    • But anyway, since I have some time off from mandatory book club readings, I went back to give Lyn Alden’s Broken Money another try. I started this giant tome over a year ago but found myself unable to focus and get excited about the history of debt and the workings of the American economy. But wait long enough, and like a broken clock, any book will become topically relevant. It might be that I’m in the right headspace now. Or the hundreds of hours of Bloomberg TV I’ve watched since have given me the landmarks needed to make sense of it. But this time I’m finding it much easier to stay on the horse and should be done with it soon.
    • After finishing I Parry Everything, I tried to find other anime with a similar premise but both —deep inhale— Failure Frame: I Became the Strongest and Annihilated Everything with Low-Level Spells and Chillin’ in Another World with Level 2 Super Cheat Powers did not nail the comedy/power tension as well as I Parry Everything.
    • We watched Steven Soderbergh’s Black Bag, and it was an excellent low-key, nearly chamber drama of a spy flick. I say ‘we’ but Kim fell asleep near the end through no fault of the film and will have to catch up later. It’s evident that Michael Fassbender could have been a great Bond if he’d been cast a decade ago and aged into it.
    • Let’s end with some music. While asking ChatGPT to guess my Enneagram type, Big Five personality traits, IQ, SATs, primary school exam results, and other traits, it offered to make a playlist that would represent me. Not a playlist of songs that I would like, mind you, but a playlist as representation, as metaphor. I took its suggestions and assembled the songs in Apple Music, including cover art it made, and played it out loud on the HomePod. Earlier gen AI chatbots would just chuck songs together without any sign of understanding that a playlist should flow, but this one works so well that I’m not sure it’s coincidental. It’s also music that I wouldn’t have chosen myself, but I enjoyed it without qualifications. Here it is if you’d like “an ambient-leaning, melancholy-smart, emotionally layered playlist. Meant for headphones, twilight hours, and slow revelations.” The first track is a little challenging for a cold start imo, then it gets good.
  • Week 15.25

    Week 15.25

    • Kim was away for work again, so I mostly stayed in and spent more time with the Vision Pro. Apple’s new VIP video series that takes you to major sporting arenas doesn’t sound like my sort of thing on paper, but the first episode on Yankee Stadium was a surprisingly entertaining watch. The crowds feel real and the only thing missing was a closeup of a hot dog. It was filmed last June, I think, and feels like a document of a much saner time. There are still many more apps and videos I have yet to check out.
    • Back in the present, Trump’s wacky tariff rollout drama continued, with the latest being a series of takesies-backsies that is great news for Apple and other makers of computers and smartphones: exemption from whatever the crazy current rate is on imports from China. 125%? Or was it 155%? Everyone now expects the stock to surge on Monday’s open, but come on, none of these pronouncements stick or mean anything now. The best move might be to sell all US-based assets and spend the next year off-the-grid on a beach somewhere.
    • Starbucks has done a collaboration with Peanuts (that’s Peanuts × Starbucks, for readers who were here last week), and of all the merch they put out, it was a yellow stoneware mug that I liked the best. It features the whole gang sitting at a long table, with Snoopy dressed as a barista and serving up espressos in his latest persona, a corporate caffeine shill stupidly named “Joe Kind”. I like Snoopy, probably on account of being exposed to the Apple TV 4K’s wonderful Snoopy and Woodstock screensaver, so I’m turning a blind eye to this sacrilege. Anyway I received my mug in the mail this week and it’s my new teatime go-to.
    • I did go out fairly early on Saturday morning to go hang out with a bunch of coffee nerd ex-colleagues who spent more time messing around with grind sizes than I ever thought possible (my contribution was a bag of beans fresh from Ho Chi Minh). It was a good time, and the occasion gave me a novel writing prompt to try out, which brings to me to my latest ChatGPT-based hobby…
    • About two years ago, when we were just grappling with the GPT-3 model and what it could do, I spent quite a lot of time trying to get it to write passable poetry. The most successful approach I found was to provide it with a detailed character backstory and ask it to write from that person’s perspective, taking into account their preferred styles and word choices. I managed to eke out some genuinely good stanzas (imo), was moved and amazed, and then moved on to other experiments and forgot about it.
    • Until this week, that is, when I wondered how the latest models would fare given the same tasks. I put my old prompts together into a custom GPT, and asked for the first silly idea that came to mind: a poem about growing older but ironically becoming more attracted to cute things like Snoopy. It did an honestly good job with just that. Then I asked it to write one about the story of how he got his name — a true story: Charles Schulz’s mother daydreamed they’d buy a family dog when she recovered from cancer, but she never did — and the result was good enough to refine further. So I went back and forth on it with the bot (who I named Penrose), combined several attempts, intervened slightly, and here’s where we landed.
    She Gave Him the Name

    “If we ever get another dog,”
    she said,
    “we should call him Snoopy.”

    It was a small thing,
    tucked between errands
    or waiting rooms,
    the kind of sentence
    you don’t expect to echo.

    He was fifteen.
    Maybe he nodded.
    Maybe he carried it like lint in his pocket,
    forgotten
    until everything else was gone.

    Years later,
    drawing alone in his studio,
    he gave the name to a dog
    who danced on rooftops,
    typed novels no one would read,
    stood in for joy
    and for all the feelings
    boys weren’t supposed to have.

    Snoopy caught on.
    So did the quiet boy with no luck,
    the girl who loved too hard,
    the blanket dragged across the years.
    The world came to know them
    and needed their steady kindness.

    Maybe that’s what it means
    to miss someone forever,
    to take the smallest thing they gave you
    and let it become
    the sky above everything you make.
    • And here’s one from another session where I asked it to identify a core human paradox and write something about it. It chose to explore how we want freedom but crave structure.
    The Night I Rearranged the Furniture

    I moved the couch against the other wall
    just to see if the room felt different.
    It didn’t.
    I stared at it like it had betrayed me,
    then moved it back.

    It was past midnight.
    The kind of hour
    when you wonder
    if this version of your life
    is the one that sticks.

    I opened the window to let in some air,
    but it was the same air as yesterday.
    Nothing new came through.

    I thought about driving somewhere,
    but didn’t know what I’d be driving to.
    The gas tank was full.
    The fridge was full.
    I was full of something I couldn’t name.

    They say people like me
    always want to leave,
    but never go far.
    We change the wallpaper.
    We cut our hair.
    We imagine new cities
    without maps.

    The truth is, I love knowing where the forks are.
    I love the sound the radiator makes
    right before it kicks in.
    I love the old light switch that sticks
    unless you press it just right.

    But still—
    some nights I crave an airport gate,
    a highway at dusk,
    a conversation in a town
    where no one knows what I used to be.

    I don’t go.
    I just stay up too late
    and move the couch.
    • It feels unreal to have this tool for S$30/mo. I can ask it for poems about any random scenario or feeling, and it tends to throw out lines and images that catch me off guard. I’ve now taken to giving it things I half-wrote long ago and never finished, and it gives pretty good notes. An unexpected side effect of this is that I’m feeling more inspired myself, and now I’m asking it to give me prompts, and I’ll write things for it to review.
    • Here’s one I reworked this week.
    He died in his 17th year

    And flowers were laid by the side of the road
    Fit to be swept within just a fortnight

    Parents kept his memory alive another thirty years,
    Until grief or emptiness took them too

    I used to think we were made for more
    How clever bones and cells can seem

    But a body is mostly waiting to be undone
    And our home keeps going around the sun
    While out there
    Nothing waits for no one
    • While I was messing with Penrose, OpenAI launched a new feature called “Memory”, where ChatGPT is able to reference all your past conversations across previously separate chats. The way they chose to showcase it was to suggest you let the AI describe you, the black mirror lighting up to reveal a camera was recording all along. I didn’t expect them to so transparently demonstrate how this technology is a more powerful, more dangerous profiling machine that anything Facebook or Google has ever put online, but I suspect they think there’s nothing we can do. I asked it to guess how I felt about certain topics, like Nolan’s Batman trilogy or the Fujifilm X100VI camera. It didn’t exactly read my mind but they were very educated guesses. Not very reassuring. It offered to guess other things like my favorite cocktail (it said Negroni; not my absolute favorite but one of them), my favorite beer (a Belgian saison; not even close), and my favorite band (Radiohead; they might be in second place).
    • This reminded me I was due another playthrough of OK Computer to keep my millennial card intact. It only gets better with time. I’ve also been listening to Counting Crows’ Hard Candy again recently. Some of those songs are the best that anyone or any AI will ever write.
  • Week 14.25

    Week 14.25

    • I read All Systems Red, the first book in Martha Wells’ Murderbot series. It’s about a security robot that’s hacked its own governor module, secretly sentient but pretending not to be, and mostly just wants to be left alone to binge-watch serials. Deeply relatable. Apple TV+ has made a show from it that’s meant to come out soon, and I can’t wait. Thankfully there are six more books (plus some novellas), so this could be their next Silo or Slow Horses: a long-running fan favorite franchise they get to keep making more of. If you like introverted robots with trust issues getting into some space shootouts, it’s a fun time.
    • Still on AI bots, since I paid for ChatGPT Plus again last week, I decided to update a custom GPT I made to serve as my personal editor and proofreader. It’s trained on a bunch of these very blog posts and now incorporates a detailed summary of my writing style into its prompt. It’s shockingly fun to work with and makes half-decent suggestions. If you’d like to try this, give ChatGPT access to a bunch of your writing, get it to codify your style as a JSON profile, then refine it by reviewing examples together.
    • It actually managed to write me a half-decent LinkedIn post from a premise I provided, not that I care to post on LinkedIn at all. After some editing and joint revisions, it’s now in a shape that wouldn’t make me cringe if I read it from someone else on LinkedIn. Wait, that’s not true. Everything on LinkedIn is cringe.
    • I’m not going to say a lot about Trump’s tariffs and the mess they’ve made of the stock market, but boy am I seeing red in my finance apps. I don’t know how Americans will be able to afford anything, and I’m kinda mad that this will affect the rest of us too.
    • Caught in the blast is Nintendo’s new Switch 2, which was detailed this week in a series of live broadcasts I’ve been anticipating for the past couple of months. The new GameChat and GameShare features they showed are very welcome, especially if we’re ever locked down in a future pandemic. They’ve done a lot to make playing with friends online feel like hanging out on the same couch. Unfortunately, the announcement was marred by a higher than expected price, something of an unforced error on their part, and people flooded the Treehouse livestream chat with calls to “DROP THE PRICE”. To make matters worse, the already unwelcome US price of $449 is now set to rise once they calculate the impact of tariffs.
    • We’ll be missing the June 5 launch in any case, with the official site saying “July–September in Singapore, Thailand, Malaysia, and the Philippines”. As for price, the original Switch launched at $299 USD, which should’ve meant about S$400 — but we ended up paying S$650 here, bundled with Breath of the Wild, because of limited supply and some greedy local distribution. I don’t expect the same kind of scalping this time, but I also wouldn’t be shocked to see it land at S$800. Can’t wait.
    • I still have so much to play on my old Switch OLED anyway, and this week I got started on Ace Attorney Investigations Collection. It’s a remastered version of the two Miles Edgeworth games from the Nintendo DS, the latter of which was never released internationally. Also in my backlog are Kirby and the Forgotten Land and The Legend of Zelda: Tears of the Kingdom, both of which are getting Switch 2 enhancement patches — higher frame rates, HDR, all that good stuff — if you’re willing to pay for the upgrade. Which means I now have a perfectly valid excuse to keep ignoring them until summer.
    • Michael pinged me out of the blue to ask if I knew anything about the origins of using the × symbol in Japanese to mean “and” or “plus”. It’s something I’ve long wondered about too, especially with anime titles like Hunter × Hunter and Spy × Family. So I outsourced the research to Perplexity (an AI search engine), and found that this usage came out of Japanese fashion subculture in the 1990s. Turns out it’s a Japanese invention, possibly inspired by its use in botany to denote crossbreeding. In modern use, the × stands in for “with”, “versus”, “of”, or “intersection”. It’s also not pronounced aloud, which is why the show is just called “Spy Family”. I like how the symbol invites layered meanings — it implies both conflict and connection. In Spy × Family, it’s the tension between the fake family setup and their hidden identities, but also how those roles merge into something real. A simple little mark doing a lot of work.
    • I watched a new anime on Netflix called I Parry Everything. Following the isekai wave a couple years back, the new trend seems to be fantasy stories about “weak” characters who go all-in on training one obscure skill — to the point of accidentally attaining god-tier strength. Jose reminded me of another in the same vein, with the glorious title I Don’t Want to Get Hurt, So I’ll Max Out My Defense. In Parry, the main guy is told early on he has no future as a swordsman — so he just spends the next 14 years practicing how to block. Now he’s practically unkillable. But the show’s comedy hinges on him not realizing this, while everyone else assumes he’s the savior of their kingdom. It’s extremely stupid, extremely fun, and yeah I binged the whole thing in a weekend.
    • We’ve also been watching The Pitt on Max and it’s a great hospital drama (starring Noah Wyle of ER) that leans more towards realism than the likes of New Amsterdam. Everything takes place over 15 hours in 15 episodes, which takes me back to watching 24 in absolute awe as a young man.
    • Pulse on Netflix is everything The Pitt is not. It’s cheesy, everyone’s more model than medic, and there’s no urgency or realism. Even the surgeries are shot in crispy iPhone-like HDR and cinematic lighting. It does have Willa Fitzgerald (aka Reacher’s partner in S1) and Néstor Carbonell (Yanko from The Morning Show) but even they can’t lift this to greatness. It’s fine background TV though.
    • What’s up with this image? I went for dinner with Peishan and Cien, who decided it would be funny to tell HaiDiLao (a massive Chinese hotpot chain) that it was my birthday month, so the staff came round and sang/blasted out of a Bluetooth speaker a proprietary and very Chinese birthday song, that apparently everyone around us knew because they joined in and clapped along. I tried to stop them, but in the end had to endure it with a pained smile.
    • Btw one legitimate use case of AI is transforming images into drawings to get around the problem of publicly sharing people’s faces.