Do you remember when your domestic ISP – Internet Service Provider – used to be an Internet Services Provider? They
were only sometimes actually called that, but what I mean is: when ISPs provided more than one Internet service? Not just connectivity, but… more.
One of the first ISPs I subscribed to had a “standard services” list longer than most modern ISPs complete services list!
ISPs twenty years ago
It used to just be expected that your ISP would provide you with not only an Internet connection, but also some or all of:
I don’t remember which of my early ISPs gave me a free license for HoTMetaL Pro, but I was very appreciative of it at the time.
ISPs today
The ISP I hinted at above doesn’t exist any more, after being bought out and bought out and bought out by a series of owners. But I checked the Website of the current owner to see what
their “standard services” are, and discovered that they are:
Optional 4G backup connectivity (for an extra fee)
A voucher for 3 months access to a streaming service3
The connection is faster, which is something, but we’re still talking about the “baseline” for home Internet access then-versus-now. Which feels a bit galling, considering that (a)
you’re clearly, objectively, getting fewer services, and (b) you’re paying more for them – a cheap basic home Internet subscription today, after accounting
for inflation, seems to cost about 25% more than it did in 2000.4
Are we getting a bum deal?
Not every BBS nor ISP would ever come to support the blazing speeds of a 33.6kbps modem… but when you heard the distinctive scream of its negotiation at close to the Shannon Limit of
the piece of copper dangling outside your house… it felt like you were living in the future.
Would you even want those services?
Some of them were great conveniences at the time, but perhaps not-so-much now: a caching server, FTP site, or IRC node in the building right at the end of my
dial-up connection? That’s a speed boost that was welcome over a slow connection to an unencrypted service, but is redundant and ineffectual today. And if you’re still using a
fax-to-email service for any purpose, then I think you have bigger problems than your ISP’s feature list!
Some of them were things I wouldn’t have recommend that you depend on, even then: tying your email and Web hosting to your connectivity provider traded
one set of problems for another. A particular joy of an email address, as opposed to a postal address (or, back in the day, a phone number), is that it isn’t tied to where
you live. You can move to a different town or even to a different country and still have the same email address, and that’s a great thing! But it’s not something you can
guarantee if your email address is tied to the company you dial-up to from the family computer at home. A similar issue applies to Web hosting, although for a true traditional “personal
home page”: a little information about yourself, and your bookmarks, it would be fine.
But some of them were things that were actually useful and I miss: honestly, it’s a pain to have to use a third-party service for newsgroup
access, which used to be so-commonplace that you’d turn your nose up at an ISP that didn’t offer it as standard. A static IP being non-standard on fixed connections is a sad reminder
that the ‘net continues to become less-participatory, more-centralised, and just generally more watered-down and shit: instead of your connection making you “part of” the Internet,
nowadays it lets you “connect to” the Internet, which is a very different experience.5
A page like this used to be absolutely standard on the Website6
of any ISP worth its salt.
Yeah, sure, you can set up a static site (unencumbered by any opinionated stack) for free on Github Pages, Neocities, or wherever, but the barrier to entry has been raised
by just enough that, doubtless, there are literally millions of people who would have taken that first step… but didn’t.
And that makes me sad.
Footnotes
1 ISP-provided shared FTP servers would also frequently provide locally-available copies
of Internet software essentials for a variety of platforms. This wasn’t just a time-saver – downloading Netscape Navigator from your ISP rather than from half-way across the world was
much faster! – it was also a way to discover new software, curated by people like you: a smidgen of the feel of a well-managed BBS, from the comfort of your local ISP!
2 ISP-provided routers are, in my experience, pretty crap 50% of the time… although
they’ve been improving over the last decade as consumers have started demanding that their WiFi works well, rather than just works.
3 These streaming services vouchers are probably just a loss-leader for the streaming
service, who know that you’ll likely renew at full price afterwards.
4 Okay, in 2000 you’d have also have had to pay per-minute for the price of the
dial-up call… but that money went to BT (or perhaps Mercury or KCOM), not to your ISP. But my point still stands: in a world where technology has in general gotten cheaper
and backhaul capacity has become underutilised, why has the basic domestic Internet connection gotten less feature-rich and more-expensive? And often with worse
customer service, to boot.
5 The problem of your connection not making you “part of” the Internet is multiplied if
you suffer behind carrier-grade NAT, of course. But it feels like if we actually cared enough to commit to rolling out IPv6 everywhere we could obviate the need for that particular
turd entirely. And yet… I’ll bet that the ISPs who currently use it will continue to do so, even as the offer IPv6 addresses as-standard, because they buy into their own idea that
it’s what their customers want.
6 I think we can all be glad that we no longer write “Web Site” as two separate words, but
you’ll note that I still usually correctly capitalise Web (it’s a proper noun: it’s the Web, innit!).
Some time in the last 25 years, ISPs stopped saying they made you “part of” the Internet, just that they’d help you “connect to” the Internet.
Most people don’t need a static IP, sure. But when ISPs stopped offering FTP and WWW hosting as a standard feature (shit though it often was), they became part of the tragic process by
which the Internet became centralised, and commoditised, and corporate, and just generally watered-down.
The amount of effort to “put something online” didn’t increase by a lot, but it increased by enough that millions probably missed-out on the opportunity to create
their first homepage.
Fellow geek, Nightline veteran, and general volunteering hero James Buller wrote a wonderful retrospective on his experience with Surrey Nightline, National Nightline, and the Nightline
Association over most of the last three decades:
…
In 1997 I left a note in the Surrey Nightline pigeon-hole to volunteer and eventually become the Coordinator
In 1998 I emailed the leaders of National Nightline with a plea for support.
In 2000 I launched the first National Nightline website and email list
In 2003 I added the bulletin board online forum
In 2006 I led governance reform and the registration project that led to the Nightline Association charity
In 2007 I set up Google Apps for the recently established nightline.ac.uk domain
In 2008 We sent news via an email broadcast system for the first time
In 2025 All the user accounts and the charity were shut down.
So here’s my last post on volunteering with the confidential mental health helplines run ‘by students for students’ at universities, then the overarching association body.
…
I began volunteering with Aberystwyth Nightline in 1999, and I remember the 2000 launch of the National Nightline mailing list and website. It felt like a moment of coalescence and
unity. We Nightline volunteers at the turn of the millennium were young, and tech-savvy, and in that window between the gradual decline of Usenet and the 2004-onwards explosion in
centralised social networking, mailing lists and forums were The Hotness.
Nightlines (and Nightliners) disagreed with one another on almost everything, but the Internet-based connectivity that James put into place for National Nightline was enormously
impactful. It made Nightline feel bigger than it had been before: it was an accessible and persistent reminder that you were part of a wider movement. It facilitated year-round
discussions that might previously have been seen only at annual conferences. It brought communities together.
(Individuals too: when my friends Kit and Fiona met and got
together back in 2003 (and, later, married), it probably wouldn’t have happened without
the National Nightline forum.)
Gosh, I spent an inordinate amount of time on this site, back in the day.
But while I praise James’ work in community-building and technology provision, his experience with Nightlines doesn’t stop there: he was an important force in the establishment of the
Nightline Association, the registered charity that took over National Nightline’s work and promised to advance it even further with moves towards accreditation and representation.
As his story continues, James talks about one of his final roles for the Association: spreading the word about the party to “see it off”. Sadly, the Nightline Association folded last
month, leaving a gap that today’s Nightlines, I fear, will struggle to fill, but this was at least the excuse for one last get-together (actually, three, but owing to schedule conflicts
I was only able to travel up to the one in Manchester):
…
I had done a lot of the leg work to track down and invite former volunteers to the farewell celebrations. I’d gotten a real buzz from it, which despite a lot of other volunteering
I’ve not felt since I was immersed in the Nightline world in the 2000’s. I felt all warm and fuzzy with nostalgia for the culture, comradeship and perhaps dolefully sense of youth
too!
I was delighted that so many people answered the call (should have expected nothing less of great Nightliners!). Their reminiscing felt like a wave of love for the movement we’d
all been a part of and had consumed such a huge part of our lives for so long. It clearly left an indelible mark on us all and has positively affected so many others through us.
…
Many people played their part in the story of the Nightline Association.
I got to hang out with some current and former Nightline volunteers in Manchester, the smallest of the ‘Goodbye NLA’ parties.
Volunteering in charity technical work is a force multiplier: instead of working on the front lines, you get to facilitate many times your individual impact for the people who
do! Volunteering with Three Rings for the last 23 years has helped me experience that, and James’ experience of this kind of
volunteering goes even further than mine. And yet he feels his impact most-strongly in a close and interpersonal story that’s humbling and beautiful:
…
I was recently asked by a researcher, ‘What is the best thing you have done as a volunteer in terms of impact?’. I was proud to reply that I’d been told someone had not killed
themselves because of a call with me at Surrey Nightline.
…
I’d recommend going and reading the full post by James, right up to the final inspiring words.
(Incidentally: if you’re looking for a volunteering opportunity that continues to help Nightlines, in the absence of the Nightline Association, Three Rings can make use of you…)
Highlights of yesterday’s Goodbye Nightline Association party in Manchester:
👨💻 Responded to Three Rings user query in real time by implementing new Directory property while at the event (pictured)
🤝 Met a handful of Nightliners past and present; swapped war stories of fights with students unions, battles for funding, etc. (also got some insights into how they’re using various
tech tools!)
✍️ Did hilariously awful job of drawing ‘Condom Man’, Aberystwyth Nightline’s mascot circa 2000
🤞 Possibly recruited a couple of new Three Rings volunteers
Low points:
😢 It’s a shame NLA’s dying, but I’m optimistic that Nightlines will survive
Sure, it’s gaudy, but it’s got a few things going for it, too.
Let’s put aside for the moment that you can already send my website back into “90s mode” and dive into this take on how I could
present myself in a particularly old-school way. There’s a few things I particularly love:
It’s actually quite lightweight: ignore all the animated GIFs (which are small anyway) and you’ll see that, compared to my current homepage, there are very few
images. I’ve been thinking about going in a direction of less images on the homepage anyway, so it’s interesting to see how it comes together in this unusual context.
The page sections are solidly distinct: they’re a mishmash of different widths, some of which exhibit a horrendous lack of responsivity, but it’s pretty clear where
the “recent articles” ends and the “other recent stuff” begins.
The post kinds are very visible: putting the “kind” of a post in its own column makes it really clear whether you’re looking at an article, note, checkin, etc., much
more-so than my current blocks do.
Maybe there’s something we can learn from old-style web design? No, I’m serious. Stop laughing.
90s web design was very-much characterised by:
performance – nobody’s going to wait for your digital photos to download on narrowband connections, so you hide them behind descriptive links or tiny thumbnails, and
pushing the boundaries – the pre-CSS era of the Web had limited tools, but creators worked hard to experiment with the creativity that was possible within those
limits.
Those actually… aren’t bad values to have today. Sure, we’ve probably learned that animated backgrounds, tables for layout, and mystery meat navigation were horrible for
usability and accessibility, but that doesn’t mean that there isn’t still innovation to be done. What comes next for the usable Web, I wonder?
As soon as you run a second or third website through the tool, its mechanisms for action become somewhat clear and sites start to look “samey”, which is the opposite of what
made 90s Geocities great.
The only thing I can fault it on is that it assumes that I’d favour Netscape Navigator: in fact, I was a die-hard Opera-head for most of the
nineties and much of the early naughties, finally switching my daily driver to Firefox in 2005.
I certainly used plenty of Netscape and IE at various points, though, but I wasn’t a fan of the divisions resulting from the browser wars. Back in the day, I always backed
the ideals of the “Viewable With Any Browser” movement.
I’ve been playing Sean O’Connor’s Slay for around 30 years (!), but somehow it took until today, on the Android version,
before I tried my hand at “rewilding” the game world.
The rules of the game make trees… a bad thing: you earn no income from hexes with them. But by the time I was winning this map anyway, I figured that encouraging growback would be a
pleasant way to finish the round.
Play your videogames any damn way you want. Don’t let anybody tell you there’s a right or wrong way to enjoy a single-player game. Today I took a strategy wargame and grew a forest. How
will you play?
My love of the yesterweb forced me to teach myself just-enough Blender to make an animation for a stupid thing: an 88×31 button representing “me” (and, I suppose, my blog, whenever I
next end up redesigning its theme).
Back in the 1980s and early 1990s, I had a collection of 5¼” and later 3½” floppy disks1 on which were stored a variety of games and utilities that I’d
collected over the years2.
I had lots of floppy disks that looked almost-exactly like this: a scrawled label of their contents and notes on how to make use of them that would perhaps only make sense to me.
I remember that at some point I acquired a program called INSULTS.COM. When executed, this tool would spoof a basic terminal prompt and then, when the user pressed any key,
output a randomly-generated assortment of crude insults.
Do you feel thoroughly insulted yet?
As far as prank programs go, it was far from sophisticated. I strongly suspect that the software, which was released for free in 1983, was intended to be primarily a vehicle to promote
sales of a more-complex set of tools called PRANKS, which was advertised within.
In any case: as a pre-pubescent programmer I remember being very interested in the mechanism by which INSULTS.COM was generating its output.
I partially-reverse-engineered the permutations by polling the output and looking for parts I hadn’t seen before, and tallying them up. Mostly in an effort to validate the program’s
claim that it’s capable of generating “more than 22 million insults”3.
Of course, nowadays I understand reverse-engineering better than I did as a child. So I downloaded a copy of INSULTS.COM from this Internet Archive image, ran it through Strings, and pulled out the data.
Easy!
Wait for it, and you can be be insulted all over again!
Why did I do this? Why do I do anything? Reimplementing a 42-year-old piece of DOS software that nobody remembers is even stranger than that time I reimplemented a 16-year old Flash advertisement! But I hope it gave you a moment’s joy to be told that you’re… an annoying load of
festering parrot droppings, or whatever.
Footnotes
1 Also some 3″ floppy disks – a weird and rare format – but that’s another story.
2 My family’s Amstrad PC1512
had two 5¼” disk drives, which made disk-to-disk copying much easier than it was on computers with a single disk drive, on which you’d have to copy as much data as possible
to RAM, swap disks to write what had been copied so far, swap disks back again, and repeat. This made it less-laborious for me to clone media than it was for most other folks I knew.
3 Assuming the random number generator is capable of generating a sufficient diversity of
seed values, the claim is correct: by my calculation, INSULTS.COM can generate 22,491,833 permutations of insults.
Of all the discussions I’ve ever been involved with on the subject of religion, the one I’m proudest of was perhaps also one of the earliest.
Let me tell you about a time that, as an infant, I got sent out of my classroom because I wouldn’t stop questioning the theological ramifications of our school nativity play.
I’m aware that I’ve got readers from around the world, and Christmas traditions vary, so let’s start with a primer. Here in the UK, it’s common1
at the end of the school term before Christmas for primary schools to put on a “nativity play”. A group of infant pupils act out an interpretation of the biblical story of the birth of
Jesus: a handful of 5/6-year-olds playing the key parts of, for example, Mary, Joseph, an innkeeper, some angels, maybe a donkey, some wise men, some shepherds, and what-have-you.
Maybe they’re just higher-budget nowadays, or maybe I grew up in a more-deprived area, but I’m pretty sure than when I was a child a costume consisted mostly of a bedsheet if you were
an angel, a tea-towel secured with an elastic band if you were a shepherd, a cardboard crown if you were a king, and so on. Photo courtesy Ian Turk.
As with all theatre performed by young children, a nativity play straddles the line between adorable and unbearable. Somehow, the innkeeper – who only has one line – forgets to
say “there is no room at the inn” and so it looks like Mary and Joseph just elect to stay in the barn, one of the angels wets herself in the middle of a chorus, and Mary, bored
of sitting in the background having run out of things to do, idly swings the saviour of mankind round and around, holding him by his toe. It’s beautiful2.
I was definitely in a couple of different nativity plays as a young child, but one in particular stands out in my memory.
“Let us go now to Bethlehem. The son of God is born today.”
In order to put a different spin on the story of the first Christmas3, one
year my school decided to tell a different, adjacent story. Here’s a summary of the key beats of the plot, as I remember it:
God is going to send His only son to Earth and wants to advertise His coming.
“What kind of marker can he put in the sky to lead people to the holy infant’s birthplace?”, He wonders.
So He auditions a series of different natural phenomena:
The first candidate is a cloud, but its pitch is rejected because… I don’t remember: it’ll blow away or something.
Another candidate was a rainbow, but it was clearly derivative of an earlier story, perhaps.
After a few options, eventually God settles on a star. Hurrah!
Some angels go put the star in the right place, shepherds and wise men go visit Mary and her family, and all that jazz.
So far, totally on-brand for a primary school nativity play but with 50% more imagination than the average. Nice.
What the Meteor Strike of Bethlehem lacked in longevity, it made up for in earth-shattering destruction.
I was cast as Adviser #1, and that’s where things started to go wrong.
The part of God was played by my friend Daniel, but clearly our teacher figured that he wouldn’t be able to remember all of his lines4 and expanded his role into three: God, Adviser #1, and
Adviser #2. After each natural phenomenon explained why it would be the best, Adviser #1 and Adviser #2 would each say a few words about the candidate’s pros and cons,
providing God with the information He needed to make a decision.
To my young brain, this seemed theologically absurd. Why would God need an adviser?5
“If He’s supposed to be omniscient, why does God need an adviser, let alone two?” I asked my teacher6.
The answer was, of course, that while God might be capable of anything… if the kid playing Him managed to remember all of his lines then that’d really be a miracle. But I’d
interrupted rehearsals for my question and my teacher Mrs. Doyle clearly didn’t want to explain that in front of the class.
But I wouldn’t let it go:
“But Miss, are we saying that God could make mistakes?”
“Couldn’t God try out the cloud and the rainbow and just go back in time when He knows which one works?”
“Why does God send an angel to tell the shepherds where to go but won’t do that for the kings?”
“Miss, don’t the stars move across the sky each night? Wouldn’t everybody be asking questions about the bright one that doesn’t?”
“Hang on, what’s supposed to have happened to the Star of Bethlehem after God was done with it? Did it have planets? Did those planets… have life?”
In the end I had to be thrown out of class. I spent the rest of that rehearsal standing in the corridor.
And it was totally worth it for this anecdote.
Footnotes
1 I looked around to see if the primary school nativity play was still common, or if the
continuing practice at my kids’ school shows that I’m living in a bubble, but the only source I could find was a 2007 news story that claims that nativity plays are “under threat”… by The Telegraph,
who I’d expect to write such a story after, I don’t know, the editor’s kids decided to put on a slightly-more-secular play one year. Let’s just continue to say that the
school nativity play is common in the UK, because I can’t find any reliable evidence to the contrary.
2 I’ve worked onstage and backstage on a variety of productions, and I have nothing but
respect for any teacher who, on top of their regular workload and despite being unjustifiably underpaid, volunteers to put on a nativity play. I genuinely believe that the kids get a
huge amount out of it, but man it looks like a monumental amount of work.
3 And, presumably, spare the poor parents who by now had potentially seen children’s
amateur dramatics interpretations of the same story several times already.
5 In hindsight, my objection to this scripting decision might actually have been masking
an objection to the casting decision. I wanted to play God!
6 I might not have used the word “omniscient”, because I probably didn’t know the word
yet. But I knew the concept, and I certainly knew that my teacher was on spiritually-shaky ground to claim both that God knew everything and God needed an advisor.
On Wednesday, Vodafone
announced that they’d made the first ever satellite video call from a stock mobile phone in an area with no terrestrial signal. They used a mountain in Wales for their experiment.
It reminded me of an experiment of my own, way back in around 1999, which I probably should have made a bigger deal of. I believe that I was the first person to ever send an email from
the top of Yr Wyddfa/Snowdon.
Nowadays, that’s an easy thing to do. You pull your phone out and send it. But back then, I needed to use a Psion 5mx palmtop, communicating over an infared link using a custom driver
(if you ever wondered why I know my AT-commands by heart… well, this isn’t exactly why, but it’s a better story than the truth) to a Nokia 7110 (fortunately it was cloudy enough to not
interfere with the 9,600 baud IrDA connection while I positioned the devices atop the trig point), which engaged a GSM 2G connection, over which I was able to send an email to myself,
cc:’d to a few friends.
It’s not an exciting story. It’s not even much of a claim to fame. But there you have it: I was (probably) the first person to send an email from the summit of Yr Wyddfa. (If you beat
me to it, let me know!)
I’ve a notion that during 2025 I might put some effort into tidying up the tagging taxonomy on my blog. There’s a few tags that are duplicates (e.g.
ai and artificial intelligence) or that exhibit significant overlap (e.g. dog and dogs), or that were clearly created when I
speculated I’d write more on the topic than I eventually did (e.g. homa night, escalators1,
or nintendo) or that are just confusing and weird (e.g. not that bacon sandwich picture).
One part of such an effort might be to go back and retroactively add tags where they ought to be. For about the first decade of my blog, i.e. prior to around 2008, I rarely used tags to
categorise posts. And as more tags have been added it’s apparent that many old posts even after that point might be lacking tags that perhaps they ought to have2.
I remain sceptical about many uses of (what we’re today calling) “AI”, but one thing at
which LLMs seem to do moderately well is summarisation3. And isn’t tagging and categorisation only a stone’s throw away from
summarisation? So maybe, I figured, AI could help me to tidy up my tagging. Here’s what I was thinking:
Tell an LLM what tags I use, along with an explanation of some of the quirkier ones.
Train the LLM with examples of recent posts and lists of the tags that were (correctly, one assumes) applied.
Give it the content of blog posts and ask what tags should be applied to it from that list.
Script the extraction of the content from old posts with few tags and run it through the above, presenting to me a report of what tags are recommended (which could then be coupled
with a basic UI that showed me the post and suggested tags, and “approve”/”reject” buttons or similar.
Extracting training data
First, I needed to extract and curate my tag list, for which I used the following SQL4:
SELECTCOUNT(wp_term_relationships.object_id) num, wp_terms.slug FROM wp_term_taxonomy
LEFTJOIN wp_terms ON wp_term_taxonomy.term_id = wp_terms.term_id
LEFTJOIN wp_term_relationships ON wp_term_taxonomy.term_taxonomy_id = wp_term_relationships.term_taxonomy_id
WHERE wp_term_taxonomy.taxonomy ='post_tag'AND wp_terms.slug NOTIN (
-- filter out e.g. 'rss-club', 'published-on-gemini', 'dancast' etc.-- these are tags that have internal meaning only or are already accurately applied'long', 'list', 'of', 'tags', 'the', 'ai', 'should', 'never', 'apply'
)
GROUPBY wp_terms.slug
HAVING num >2-- filter down to tags I actually routinely useORDERBY wp_terms.slug
Many of my tags are used for internal purposes; e.g. I tag posts published on gemini if they’re to appear on gemini://danq.me/ and
dancast if they embed an episode of my podcast. I filtered these out because I never want the AI to suggest applying them.
I took my output and dumped it into a list, and skimmed through to add some clarity to some tags whose purpose might be considered ambiguous, writing my explanation of each in
parentheses afterwards. Here’s a part of the list, for example:
I used that list as the basis for the system message of my initial prompt:
Suggest topical tags from a predefined list that appropriately apply to the content of a given blog post.
# Steps
1. **Read the Blog Post**: Carefully read through the provided content of the blog post to identify its main themes and topics.
2. **Analyse Key Aspects**: Identify key topics, themes, or subjects discussed in the blog post.
3. **Match with Tags**: Compare these identified topics against the list of available tags.
4. **Select Appropriate Tags**: Choose tags that best represent the main topics and themes of the blog post.
# Output Format
Provide a list of suggested tags. Each tag should be presented as a single string. Multiple tags should be separated by commas.
# Allowed Tags
Tags that can be suggested are as follows. Text in parentheses are not part of the tag but are a description of the kinds of content to which the tag ought to be applied:
- aberdyfi
- aberystwyth
- ...
- youtube
- zoos
# Examples
**Input:**
The rapid advancement of AI technology has had a significant impact on my industry, even on the ways in which I write my blog posts. This post, for example, used AI to help with tagging.
**Output:**
ai, technology, blogging, meta, work
...(other examples)...
# Notes
- Ensure that all suggested tags are relevant to the key themes of the blog post.
- Tags should be selected based on their contextual relevance and not just keyword matching.
This system prompt is somewhat truncated, but you get the idea.
That post already has the following tags (but this wasn’t disclosed to the AI in its training set; it had to work from scratch): children, language, languages (a bit of a redundancy there!), spain, and unicode.
Testing it out
Let’s see what the AI suggests:
curl https://api.openai.com/v1/chat/completions \
-H "Content-Type: application/json"\
-H "Authorization: Bearer $OPENAI_TOKEN"\
-d '{ "model": "gpt-4o-mini", "messages": [ { "role": "system", "content": [ { "type": "text", "text": "[PROMPT AS DESCRIBED ABOVE]" } ] }, { "role": "user", "content": [ { "type": "text", "text": "My 8-year-old asked me \"In Spanish, I need to use an upside-down interrobang at the start of the sentence‽\" I assume the answer is yes A little while later, I thought to check whether Unicode defines a codepoint for an inverted interrobang. Yup: ‽ = U+203D, ⸘ = U+2E18. Nice. And yet we dont have codepoints to differentiate between single-bar and double-bar \"cifrão\" dollar signs..." } ] } ], "response_format": { "type": "text" }, "temperature": 1, "max_completion_tokens": 2048, "top_p": 1, "frequency_penalty": 0, "presence_penalty": 0}'
Running this via command-line curl meant I quickly ran up against some Bash escaping issues, but set +H and a little massaging of the blog post content
seemed to fix it.
GPT-4o-mini
When I ran this query against the gpt-4o-mini model, I got back: unicode, language, education, children, symbols.
That’s… not ideal. I agree with the tags unicode, language, and children, but this isn’t really abouteducation. If I tagged
everything vaguely educational on my blog with education, it’d be an even-more-predominant tag than geocaching is! I reserve that tag for things that relate
specifically to formal education: but that’s possibly something I could correct for with a parenthetical in my approved tags list.
symbols, though, is way out. Sure, the post could be argued to be something to do with symbols… but symbols isn’t on the approved tag list in
the first place! This is a clear hallucination, and that’s pretty suboptimal!
Maybe a beefier model will fare better…
GPT-4o
I switched gpt-4o-mini for gpt-4o in the command above and ran it again. It didn’t take noticeably longer to run, which was pleasing.
The model returned: children, language, unicode, typography. That’s a big improvement. It no longer suggests education,
which was off-base, nor symbols, which was a hallucination. But it did suggest typography, which is a… not-unreasonable suggestion.
Neither model suggested spain, and strictly-speaking they were probably right not to. My post isn’t about Spain so much as it’s about Spanish. I don’t
have a specific tag for the latter, but I’ve subbed in the former to “connect” the post to ones which are about Spain, but that might not be ideal. Either way: if this is how
I’m using the tag then I probably ought to clarify as such in my tag list, or else add a note to the system prompt to explain that I use place names as the tags for posts about
the language of those places. (Or else maybe I need to be more-consistent in my tagging).
I experimented with a handful of other well-tagged posts and was moderately-satisfied with the results. Time for a more-challenging trial.
This time, with feeling…
Next, I decided to run the code against a few blog posts that are in need of tags. At this point, I wasn’t quite ready to implement a UI, so I just adapted my little hacky Bash
script and copy-pasted HTML-stripped post contents directly into it.
If it worked, I decided, I could make a UI. Until then, the command line was plenty sufficient.
In this post, I shared that my grandmother and my coworker had (independently) been taken into hospital. It had no tags whatsoever.
The AI suggested the tags hospital, family, injury, work, weddings, pub, humour. Which at
a glance, is probably a superset of the tags that I’d have considered, but there’s a clear logic to them all.
It clearly picked out weddings based on a throwaway comment I made about a cousin’s wedding, so I disagree with that one: the post isn’t strictly about weddings
just because it mentions one.
pub could go either way. It turns out my coworker’s injury occurred at or after a trip to the pub the previous night, and so its relevance is somewhat unknowable from this
post in isolation. I think that’s a reasonable suggestion, and a great example of why I’d want any such auto-tagging system to be a human assistant (suggesting
candidate tags) and not a fully-automated system. Interesting!
Finally, you might think of humour as being a little bit sarcastic, or maybe overly-laden with schadenfreude. But the blog post explicitly states that my coworker
“carefully avoided saying how he’d managed to hurt himself, which implies that it’s something particularly stupid or embarrassing”, before encouraging my friends to speculate on it.
However, it turns out that humour isn’t one of my existing tags at all! Boo, hallucinating AI!
I ended up applying all of the AI’s suggestions except weddings and humour. I also applied smartdata, because that’s where I worked (the AI couldn’t have been expected to guess that without context, though!).
This post talked about Ash and I’s travels around the UK to see REM and Green Day in concert5 and to the National Science Museum in London where I discovered that Ash was prejudiced towards…
carrot cake.
The AI suggested: concerts, travel, music, preston, london, science museum, blogging.
Those all seemed pretty good at a first glance. Personally, I’d forgotten that we swung by Preston during that particular grand tour until the AI suggested the tag, and then I had to
look back at the post more-carefully to double-check! blogging initially seemed like a stretch given that I was only blogging about not having blogged much, but on
reflection I think I agree with the robot on this one, because I did explicitly link to a 2002 page that fell off the Internet only a few years ago aboutthe pointlessness of blogging. So I think it counts.
I was able to verify that I’d been in Preston with thanks to this contemporaneous photo. I have no further explanation for the content of the photo, though.
science museum is a big fail though. I don’t use that tag, but I do use the tag museum. So close, but not quite there, AI!
I applied all of its suggestions, after switching museum in place of science museum.
I wrote this blog post in celebration of having managed to hack together some stuff to help me remote-control my PC from my phone via Bluetooth, which back then used to be a challenge,
in the hope that this would streamline pausing, playing, etc. at pizza-distribution-time at Troma Night, a weekly film night I hosted back then.
If you were sat on that sofa, fighting your way past other people and a mango-chutney-barrel-cum-table to get to a keyboard was genuinely challenging!
It already had the tag technology, which it inherited from a pre-tagging evolution of my blog which used something akin to categories (of which only one
could be assigned to a post). In addition to suggesting this, the AI also picked out the following options: bluetooth, geeky, mobile, troma
night, dvd, technology, and software.
The big failure here was dvd, which isn’t remotely one of my tags (and probably wouldn’t apply here if it were: this post isn’t about DVDs; it barely even mentions
them). Possibly some prompt engineering is required to help ensure that the AI doesn’t make a habit of this “include one tag not from the approved list, every time” trend.
Apart from that it’s a pretty solid list. Annoyingly the AI suggested mobile, which isn’t an approved tag, instead of mobiles, which is. That’s probably a
tokenisation fault, but it’s still annoying and a reminder of why even a semi-automated “human-checked” system would need a safety-check to ensure that no absent tags are
allowed through to the final stage of approval.
This post!
As a bonus experiment, I tried running my code against a version of this post, but with the information about the AI’s own prompt and the examples removed (to reduce the risk
of confusion). It came up with: ai, wordpress, blogging, tags, technology, automation.
All reasonable-sounding choices, and among those I’d made myself… except for tags and automation which, yet again, aren’t among tags that I use. Unless this
tendency to hallucinate can be reined-in, I’m guessing that this tool’s going to continue to have some challenges when used on longer posts like this one.
Conclusion and next steps
The bottom line is: yes, this is a job that an AI can assist with, but no, it’s not one that it can do without supervision. The laser-focus with which gpt-4o was able to
pick out taggable concepts, faster than I’d have been able to do for the same quantity of text, shows that there’s potential here, but it’s not yet proven itself enough of a time-saver
to justify me writing a fluffy UI for it.
However, I might expand on the command-line tools I’ve been using in order to produce a non-interactive list of tagging suggestions, and use that to help inform my work as I tidy up the
tags throughout my blog.
You still won’t see any “AI-authored” content on this site (except where it’s for the purpose of talking about AI-generated content, and it’ll always be clearly labelled), and
I can’t see that changing any time soon. But I’ll admit that there might be some value in AI-assisted curation and administration, so long as there’s an informed human in the loop at
all times.
Footnotes
1 Based on my tagging, I’ve apparently only written about escalators once, while playing Pub Jenga at Robin‘s 21st birthday party. I can’t imagine why I thought it deserved a tag.
2 There are, of course, various other people trying similar approaches to this and similar
problems. I might have tried one of them, were it not for the fact that I’m not quite as interested in solving the problem as I am in understanding how one might use an AI to
solve the problem. It’s similar to how I don’t enjoy doing puzzles like e.g. sudoku as much as I enjoy writing software that optimises for solving such puzzles. See also, for
example, how I beat my children at Mastermind or what the hardest word in Hangman is
or my variousattempts to avoid doing online jigsaws.
If you’ve come across Tony Domenico’s work before it’s probably as a result of web horror video series Petscop.
3D Workers Island… isn’t quite like that (though quick content warning: it does vaugely allude to child domestic
abuse). It’s got that kind of alternative history/”found footage webpage” feel to it that I enjoyed so much about the Basilisk
collection. It’s beautifully and carefully made in a way that brings its world to life, and while I found the overall story slightly… incomplete?… I enjoyed the application of
its medium enough to make up for it.
Best on desktop, but tolerable on a large mobile in landscape mode. Click each “screenshot” to advance.
A lot of attention is paid, often in retrospect, to the experience of the first times in our lives. The first laugh; the first kiss; the first day at your job1.
But for every first, there must inevitably be a last.
I recall a moment when I was… perhaps the age our eldest child is now. As I listened to the bats in our garden, my mother told me about how she couldn’t hear them as clearly as she
could when she was my age. The human ear isn’t well-equipped to hear that frequency that bats use, and while children can often pick out the sounds, the ability tends to fade with age.
“Helloooo? Are you even listening to me?”
This recollection came as I stayed up late the other month to watch the Perseids. I lay in the hammock in our garden under a fabulously clear sky as the sun finished setting, and –
after being still and quiet for a time – realised that the local bat colony were out foraging for insects. They flew around and very close to me, and it occurred to me that I
couldn’t hear them at all.
There must necessarily have been a “last time” that I heard a bat’s echolocation. I remember a time about ten years ago, at the first house in Oxford of
Ruth, JTA and I (along with
Paul), standing in the back garden and listening to those high-pitched chirps. But I can’t tell you when the very
last time was. At the time it will have felt unremarkable rather than noteworthy.
First times can often be identified contemporaneously. For example: I was able to acknowledge my first time on a looping rollercoaster at the time.
The Tower of Terror at Camelot, circa 1994, was my first looping rollercoaster2.
The ride was disassembled in 2000 and, minus its “tower” theming3lived on for a while as Twist ‘N’ Shout at Loudoun Castle in Ayrshire, Scotland before that park
shut down. I looked at some recent satellite photography and I’m confident it’s now been demolished.
Last times are often invisible at the time. You don’t see the significance of the everyday and routine except in hindsight.
I wonder what it would be like if we had the same level of consciousness of last times as we did of firsts. How differently might we treat a final phone call to a loved one or the
ultimate visit to a special place if we knew, at the time, that there would be no more?
Would such a world be more-comforting, providing closure at every turn? Or would it lead to a paralytic anticipatory grief: “I can’t visit my friend; what if I find out that
it’s the last time?”
Footnotes
1 While watching a wooden train toy jiggle down a length of string, reportedly; Sarah
Titlow, behind the school outbuilding, circa 1988; and
five years ago this week, respectively.
2 Can’t see the loop? It’s inside the tower. A clever bit of design conceals the
inversion from outside the ride; also the track later re-enters the fort (on the left of the photo) to “thread the needle” through the centre of the loop. When they were running three
trains (two in motion at once) at the proper cadence, it was quite impressive as you’d loop around while a second train went through the middle, and then go through the middle while a
third train did the loop!
3 I’m told that the “tower” caught fire during disassembly and was destroyed.
Last month I implemented an alternative mode to view this website “like it’s 1999”, complete with with cursor trails, 88×31 buttons, tables for
layout1,
tiled backgrounds, and even a (fake) hit counter.
One thing I’d have liked to do for 1999 Mode but didn’t get around to would have been to make the images look like it was the 90s, too.
Back then, many Web users only had graphics hardware capable of displaying 256 distinct colours. Across different platforms and operating systems, they weren’t even necessarily
the same 256 colours2!
But the early Web agreed on a 216-colour palette that all those 8-bit systems could at least approximate pretty well.
I had an idea that I could make my images look “216-colour”-ish by using CSS to apply an SVG filter, but didn’t implement it.
But Spencer, a long-running source of excellent blog comments, stepped up and wrote an SVG
filter for me! I’ve tweaked 1999 Mode already to use it… and I’ve just got to say it’s excellent: huge thanks, Spencer!
The filter coerces colours to their nearest colour in the “Web safe” palette, resulting in things like this:
The flat surfaces are particularly impacted in this photo (as manipulated by the CSS SVG filter described above). Subtle hues and the gradients coalesce into slabs of colour, giving
them an unnatural and blocky appearance.
Plenty of pictures genuinely looked like that on the Web of the 1990s, especially if you happened to be using a computer only capable of 8-bit colour to view a page built by
somebody who hadn’t realised that not everybody would experience 24-bit colour like they did3.
Dithering
But not all images in the “Web safe” palette looked like this, because savvy web developers knew to dither their images when converting them to a limited palette.
Let’s have another go:
This image uses exactly the same 216-bit colour palette as the previous one, but looks a lot more “natural” thanks to the Floyd–Steinberg dithering algorithm.
Dithering introduces random noise to media4
in order to reduce the likelihood that a “block” will all be rounded to the same value. Instead; in our picture, a block of what would otherwise be the same colour ends up being rounded
to maybe half a dozen different colours, clustered together such that the ratio in a given part of the picture is, on average, a better approximation of the correct
colour.
The result is analogous to how halftone printing – the aesthetic of old comics and newspapers, with different-sized dots made from
few colours of ink – produces the illusion of a continuous gradient of colour so long as you look at it from far-enough away.
Zooming in makes it easy to see the noisy “speckling” effect in the dithered version, but from a distance it’s almost invisible.
The other year I read a spectacular article by Surma that explained in a very-approachable way
how and why different dithering algorithms produce the results they do. If you’ve any interest whatsoever in a deep dive or just want to know what blue noise is and why you
should care, I’d highly recommend it.
You used to see digital dithering everywhere, but nowadays it’s so rare that it leaps out as a revolutionary aesthetic when, for example, it gets used in
a video game.
Dithering can be so effective that it can even make an image “work” all the way down to 1-bit (i.e. true monochrome/black-and-white) colour. Here I’ve used Jarvis, Judice & Ninke’s dithering algorithm, which is highly-effective for picking out subtle colour differences in what would
otherwise be extreme dark and light patches, at the expense of being more computationally-expensive (to initially create) than other dithering strategies.
All of which is to say that: I really appreciate Spencer’s work to make my “1999 Mode” impose a 216-colour palette on images. But while it’s closer to the truth, it still doesn’t
quite reflect what my website would’ve looked like in the 1990s because I made extensive use of dithering when I saved my images in Web safe palettes5.
Why did I take the time to dither my images, back in the day? Because doing the hard work once, as a creator of graphical Web pages, saves time and computation (and can look
better!), compared to making every single Web visitor’s browser do it every single time.
Which, now I think about it, is a lesson that’s still true today (I’m talking to you, developers who send a tonne of JavaScript and ask my browser to generate the HTML for you
rather than just sending me the HTML in the first place!).
Footnotes
1 Actually, my “1999 mode” doesn’t use tables for layout; it pretty much only applies a
CSS overlay, but it’s deliberately designed to look a lot like my blog did in 1999, which did use tables for layout. For those too young to remember: back before CSS
gave us the ability to lay out content in diverse ways, it was commonplace to use a table – often with the borders and cell-padding reduced to zero – to achieve things that today
would be simple, like putting a menu down the edge of a page or an image alongside some text content. Using tables for non-tabular data causes problems, though: not only is
it hard to make a usable responsive website with them, it also reduces the control you have over the order of the content, which upsets some kinds of accessibility
technologies. Oh, and it’s semantically-invalid, of course, to describe something as a table if it’s not.
2Perhaps as few as 22 colours were defined the same across all
widespread colour-capable Web systems. At first that sounds bad. Then you remember that 4-bit (16 colour) palettes used to look look perfectly fine in 90s videogames. But then you
realise that the specific 22 “very safe” colours are pretty shit and useless for rendering anything that isn’t composed of black, white, bright red, and maybe one of a few
greeny-yellows. Ugh. For your amusement, here’s a copy of the image rendered using only the “very safe” 22 colours.
3 Spencer’s SVG filter does pretty-much the same thing as a computer might if asked to
render a 24-bit colour image using only 8-bit colour. Simply “rounding” each pixel’s colour to the nearest available colour is a fast operation, even on older hardware and with larger
images.
4 Note that I didn’t say “images”: dithering is also used to produce the same “more
natural” feel for audio, too, when reducing its bitrate (i.e. reducing the number of finite states into which the waveform can be quantised for digitisation), for example.
5 I’m aware that my footnotes are capable of nerdsniping Spencer, so by writing this
there’s a risk that he’ll, y’know, find a way to express a dithering algorithm as an SVG filter too. Which I suspect isn’t possible, but who knows! 😅
![Stylish (for circa 2000) webpage for HoTMetaL Pro 6.0, advertising its 'unrivaled [sic] editing, site management and publishing tools'.](/_q23u/2025/08/hotmetal-pro-6-640x396.jpg)
RSS
Spotify
Pocket Casts
Apple Podcasts
YouTube
