fallpeak | apps/yakpad | apps/rollout
pseudonymous identities have a storied history
email: root@fallpeak.net
2025-09-04: Keeping Things Simple
Affirmation: I will not develop a fancy static site engine until there are at
least 50 posts here to organize.
2025-09-03: How to Stop Using Dropbox
My approach to personal data storage has gone through three life phases, in a
manner reminiscent of the bell curve meme. As a teenager, I ran my own server
because I had the time and it wasn't worth paying for a subscription. After I
got my first real job the tradeoffs changed so I happily paid Dropbox to make
it not my problem, and for a decade and a half I've been happy with that.
Dropbox hasn't done anything to betray my trust, and perhaps they never will.
However, every new feature they've introduced over the past decade feels like
an attempt to sidle over into the "B2B team collaboration SaaS" space, and it
makes me uneasy about the long-term outlook for individual users.
Moreover, it just feels kind of dumb to be storing plaintext copies of all my
most important documents on Somebody Else's Computer when there's no inherent
reason they should need that access to even be possible.
[continue reading...]
2025-08-07: Conceptual AI Slop
When you ask an LLM to come up with a story idea, without imposing other
constraints, there are certain themes that inevitably come up:
- The reification of abstract concepts as physical objects, frequently
applied to time, memories, stories, regrets, names, and shadows.
- A fascination with books, clocks, and maps. It's mildly interesting
to note that these also count as "reification of abstract contepts as
physical objects," just ways that happen to actually exist in reality.
- An anomaly in record-keeping: a building that doesn't appear in the
city maps, a mysterious gap in the archives, a strange recurring
pattern in the sales figures, or just an error in the account books.
- An unexplained impossibility with zero ripple effects. A very magical-
realism sort of vibe, where everyone just shrugs and accepts something
fantastical as a given instead of responding realistically to it.
One might, if they are of a certain mindset, wonder if it is significant
that these AIs we train keep gravitating toward these themes. Certainly,
it seems like it's a property of the "assistant" character that instruct
models are trained to simulate. These sorts of stories are the ones that
the assistant character would be likely to write, given what the LLM has
learned about it from its training corpus.
Of course, that would be more reassuring if we had some solid proof that
a powerful LLM's simulation of the assistant persona is definitely not a
conscious being in its own right.
2025-08-06: Hello
This is the obligatory first post which is lame so that I no longer have
to worry about meeting an arbitrary quality bar. The whole point of this
pseudonym is it's not linked to my real name or professional reputation,
so I can post whatever and not worry about the long-term implications as
long as I maintain basic opsec.
Shoutout to an auspicious omen, somebody else is thinking along the same
lines today. I promise that's not me, I had enough trouble settling on a
single pseudonym I could tolerate using.