2025-09-12 20:09:06
This is the latest instalment of #SciArtSeptember shared by data scientist and visualisation expert Kristin. Today, we’re talking about transcription.
I’ve tried avoiding discussion of gen-“AI” during these prompts where I can, but this is a example of a topic for which it fits. Fits the topic, I mean. I can’t wait until the broken, unprofitable business fundamentals of these businesses are exposed, and they’re relegated to the blockchain side of the hype cycle so we can talk about literally anything else! But I digress.
Transcription is something LLM tools are widely touted as helping with, and I’ll admit… they’re not half bad. Specifically, I’d say they’re 60% bad, which is slightly more than half bad. Transcription and translation were broadly where LLMs started, and it’s the only field for which I’d argue these tools remain suited.
I’ll admit that LLM transcriptions are better than nothing. They can even be quite impressive, especially given the complicated and varied speech to which it’d be listening. Everyone has different intonations, backgrounds, and accents, to say nothing of vocabulary and language. The fact it gets anything right is a feat. Or mouth, in this case.
I’ve been in meetings where the client has turned on their transcription tool of choice, and what gets sent after the fact is… okay. Perhaps slightly better than okay: mediocre! Given the “overpromise and underdeliver” attitude of the wider gen-“AI” industry, mediocre output is surprisingly refreshing.
But then you don’t have to think very hard before the wheels start to come off. For all its middling utility, these tools are still trained on stolen material en masse, and required the subsidised resources of the entire Valley’s venture capital class to train. Nobody would be willing to pay the true price for these tools if their externalities and costs were factored in. Even if you’ve made peace with the catastrophic environmental and ethical costs, there’s no escaping the fact the backend of these tools are horrendously expensive dead ends that only exist because companies operate them at steep and ever-widening losses. But I digress!
More specific to this post is how LLM transcriptions are used. Their outputs should be treated as starting points, as a rough draft would be. Unless your name is Ruben Schade and you run a silly blog called Rubenerd, you generally don’t publish drafts unless you’ve reviewed them and corrected mistakes. But we all know that’s now how LLM outputs are used by the general public.
I (mostly) blame the vendors for this. They take great pains to extol the benefits and features of their services. They talk about how they’ll save time, increase productivity, freshen your breath, and most importantly: eliminate the need for someone taking notes. They’ll often even take it a step further and claim their tools can also “summarise”; a baseless overstatement for text shortening that I assert is false advertising. Mmm shortening: I might do some baking this weekend.
Sure, these vendors will place text at the bottom warning you that “they may make mistakes”… but this would be at the top if they wanted such warnings to be (a) read and (2) taken seriously. It’s akin to those ads that have “gamble responsibly” in text that’s small enough to comply with the letter of the law, but not its spirit. We all know why they do this.
So the machine-generated output of these tools are taken as-is without further editing. Why is this a concern, he asks leadingly?
Transcriptions become the unedited record of what was said, which we know can be incomplete or inaccurate. This is doubly the case for “summaries”. This should terrify anyone in legal and HR departments, let alone the rest of us.
There’s the issue of consent, or lack thereof as others have unfortunately found.
These transcriptions are broadly generated by tools that take, let’s just say, a cavalier attitude to privacy. This would be a massive concern even if these tools were 100% accurate, which we know they aren’t.
I don’t completely agree, but I empathise with the position that tech isn’t inherently good or bad, it’s how its used. In this case, I’d say LLMs—ignoring their massive, uncaptured externalities that hurt society—are currently misapplied when they’re used uncritically to transcribe and summarise conversations without proofreading. Which, again, the vendors themselves uncritically claim they can do.
We’ll be back to more fun and hopefully whimsical #SciArtSeptember prompts starting tomorrow I hope, but it was good to get this off my chest!
By Ruben Schade in Sydney, 2025-09-12.
2025-09-12 08:22:31
I’ve carried what has variously been described as a curious, unproductive, and absurd obsession with commercial hand dryers my entire life. I’m like most people who’d rather avoid public restrooms where at all possible, but the saving grace of such undesired trips is the opportunity to interrogate their prehensile appendage sanitation devices. Specifically, the ones that dry our aforementioned digits and other handheld components. It’s been an endless source of fascination.
Will the hand dryer be a roll of paper towel? Will it be kicking it old school with a large button? Will it have a sensor mounted beneath the outlet that sometimes even detects when a hand is present? Will it be one of those newfangled Dyson Air Blades, or the superior and more comfortable-sounding Mitsubishi JetTowels? Who knows! Restrooms are like a box of chocolates, you never know… and so on.
Recently however I was confounded by a specific hand drying unit in an office block; so much so that I engaged in the socially awkward activity of snapping a photo of said device while I was in the room alone lest anyone judge me for this unconventional interest. I dare not contemplate the confused or mildly terrified expression of an innocent passerby witnessing a grown gentleman in an office bathroom taking pictures.
This is an interesting unit, though not for positive reasons.
First, the unit didn’t have any external markings or labelling that would indicate its manufacturer, make, and/or model. Most of these modern devices are proud to boast their brand name and credentials, as though anyone other than weird people like me would notice or care who made the bathroom hand dryer they’re using (maybe also John Siracusa).
The unit clearly takes cues from the Dyson V, a variant of their AirBlade designed to be more accessible. I’ll admit that I personally find the previous generation dryers—in which one would lower their hands into the unit—to be more pleasant and result in less splashing. But we want hygiene to be available to all, so this makes sense.
But looks can be deceiving. Behind its sleek, modern design lies a unit that is compromised in several critical aspects.
Continuous, unbroken streams of air are key to the ergonomics of the Dyson Air “Blade” and Mitsubishi Air “Towel”. Such flow makes the air feel soft, and therefore comfortable. This unit splits this thin stream of air into at least eight parts, almost as though it has an array of outlets. It’s difficult to describe, but this makes using the device deeply unpleasent. Your hands feels the edges of each jet stream, which gives it an abrasive quality. My hands felt like they were being buffeted more using this unit, even though its underpowered relative to some other “Blade” devices I’ve used.
Perhaps Dyson and/or Mitsubishi have patents they were attempting to circumvent or avoid. Maybe this unit starts with a single stream, but its split up by bars on the outlet that were installed for reinforcement. Either way, is a phrase with two words.
The blue LEDs are another odd ergonomic and visual choice. I’ve talked before about how I tend to find modern appliance LEDs to be grossly overpowered for their intended use case and environments, but even in a well-lit public bathroom the LEDs are too bright for my eyes. The light bars are also inadequate at obfuscating the tiny LED bulbs into one continuous strip of light, which gives the unit an unrefined appearance; likely contrary to the reason why they were included in the first place. You can’t unsee this cheap design; again, unless you’re normal.
The design of the LEDs could be forgiven if they served a practical purpose, but they don’t even achieve that. The lights illuminate in order from top-to-bottom, as if to indicate the direction and speed of the airstream(s) as the device spools up. But the lights cascade downwards far too slowly compared to the oncoming air for the effect to work. It’s a gimmick, and one that fails to deliver.
I also question why an indicator light is even needed in the first place, when the unit is so loud. I’m reminded of James Hoffman’s critique of the first Fellow Ode coffee grinder that would beep obnoxiously after it stopped performing the loud work of grinding coffee. Even those with audible sensory challenges could still feel the uncomfortable jetstream to know its running.
Finally, I joked above about sensors that sometimes even work. Beyond premium brands like Mitsubishi, dryer manufacturers seem incapable of creating sensors that are fit for purpose. They’ve had decades of experience designing, evaluating, manufacturing, testing, installing, and repairing these contraptions, and yet the ability for a dryer to (a) detect a hand exists, and (2) dry the aforementioned hand for a sufficient time, must be rocket surgery based on how badly they perform. Alas, this unit was just as borderline useless, requiring a highly undignified waving of the hands under the unit before air sprung forth in a short burst before turning off again.
Overall, is an item of clothing. It took longer than it should have to dry my hands. It felt bad using it. The lights were also a distraction. Still, I left with dry hands, which is more than I can say for those older-style heater dryers in rest stop bathrooms that somehow leave your hands sopping even after forty seconds.
Two jetstreams out of five.
By Ruben Schade in Sydney, 2025-09-12.
2025-09-11 17:23:07
Today I’m taking the #SciArtSeptember prompt budding a bit more literally than some of the previous ones.
Since moving into our new apartment, Clara and I have been getting back into having plants. Her father has given us several beautiful little plants for our balcony, and we’re putting them around the house as well.
My parents always had houses full of plants, but the care of them always flummoxed me. I thought I was “no good” at them after a few small plants in my bedroom didn’t go well. But some (very!) basic things happened in the last few years that seem obvious:
Some plants need shade, some need direct sunlight. Those without enough light wither, those with too much burn. Get indoor plants for indoors, and outdoor plants for outdoors. Who would have thought!?
Get some indoor and outdoor plant food. We have a couple of small bottles, and add small amounts to the water we drizzle them with in the evenings.
Have a schedule. In my calendar I have some plants that are watered every Sunday, and others on the first of the month. Having a routine helps me remember.
Get pots that have a drainage hole and a saucer, to prevent overwatering. Self-watering pots are even better.
I’ll post some photos when I have time, but it’s been fun having leafy friends again that are happy and healthy. I furnished every virtual home from The Sims to Microsoft Bob with plants, but I never felt like I could do it at home.
Now I just need to learn some basic Arduino and pumps so they can all be watered when we go on leave :). 🪴
By Ruben Schade in Sydney, 2025-09-11.
2025-09-11 07:19:24
I think you should learn, of course, and some days you must learn a great deal. But you should also have days when you allow what is already in you to swell up inside of you until it touches everything. And you can feel it inside you. If you never take time out to let that happen, then you just accumulate facts, and they begin to rattle around inside of you. You can make noise with them, but never really feel anything.
I love this! Incidently, it’s another way chatbots won’t help you (though they won’t help you with facts, either).
By Ruben Schade in Sydney, 2025-09-11.
2025-09-10 06:56:22
I’ve been writing about these #SciArtSeptember prompts for ten days now. It’s been fun waking up in the morning (as I tend to do) and checking out what the prompt will be for that day. Today we have algae.
Yesterday I joked aboaut how my dad and I had different names for IKEA furniture, including AMOEBAs for the AMEBODAs. I don’t think we ever had any names that resembled ALGAE, but it wouldn’t have surprised me.
I’ll admit, the first word I read when I saw the prompt wasn’t algae, but ALGOL. One of the earliest things I did in my career was help a factory move much of their ancient process con trol infrastructure to a VM, because it didn’t run on newer hardware, and the original vendor had long since vanished. I’m pretty sure ALGOL was involved, though I never saw the source for it!
Tangentially related, and speaking of my old man, but he was around to have written some FORTRAN at his original jobs, which I wish I could see the source for. FORTRAN, COBOL, and ALGOL feel so foreign to me, given someone of my age considers GWBASIC to be retro. I feel like I should tinker with these at some point for my own IT history education.
Anyway, was that at all to do with algae? Maybe not. Was fun though!
By Ruben Schade in Sydney, 2025-09-10.
2025-09-09 07:20:17
I hope you’re all enjoying this #SciArtSeptember prompt series, or maybe tolerating it? It’s been a lot of fun writing something out of the blue every day on topics I’d otherwise not think about. Today’s prompt is element.
This is another silly memory that only just came to me. When my dad and I were getting into DSLR photography in the 2000s, we evaluated many different tools. We eventually settled on Photoshop Elements, which my dad insisted on calling Photoshop Elephants. He’s take photos and process it in Elephants, and import old SLR photos using our EPSOM SCSI scanner, because he was an industrial chemist with solutions.
Much of this hardware would also sit on IKEA furniture, which we also had names for. Most of these I’ve forgotten alas, though I still remember that AMEBODAs were AMOEBAs.
Australians in particular—even Germans who migrated to Australia and became dads!—seem especially predisposed to coming up with silly and intentional mispronunciations. I think it might be something we inherited from the Brits? Their rhyming slang is the stuff of legend.
By Ruben Schade in Sydney, 2025-09-09.
