Drafts

Previously this post was called "Feature request to self: draft posts". Its original contents will be below. But I've decided to make it into a public scratch pad for drafts after all. It's important to allow others to see my sloppy incomplete thinking, or things I was interested in at some point. I should try to date things too.

Original post content:

I should really add some draft post functionality to this blog. Over the years I've had several and some of them I even finish but the way I keep the drafts around is just keeping the new post tab around as one of my hundreds of tabs. This used to be a lot more flaky and I'd lose some drafts that way, as the new post page is mostly just a simple textarea tag.

I also think I lost a few drafts when I did a tab migration from my old computer to my new computer and didn't think about some of the implicitly saved form state before I closed all the tabs on the old computer.

Anyway, it's just dumb to keep drafts around forever like this... And some I realize I should probably not post ever, or at least not for a long time, but I don't want to delete them, and I don't want to just stuff them in a corner of my file system....

As a compromise, maybe I'll use this post to anchor some drafts but hidden in HTML comments so only people who are really bored (or AIs) will care about reading them.

(End original content.)

Now I'm freeing things from their prior HTML comment state!

---

Draft Posts

Deciding where to move
Summarizing some thoughts on recent AI Doom
Rant about modern soft language and words no longer meaning things
Nippon 雪月花 lyrics
On resisting anime until 2011
Every life is wasted, or no life is.
More on LLMs

---

Deciding where to move

(Probably started writing in late 2021)

Recently (ed: summer 2021) I took a 25 day road trip across the country with my friend and housemate. The purpose was to see if there was a place we liked (individually or together) to consider moving to (ditto), and also have a bit of fun now that the pandemic is basically over.

No decisions have been made yet... but since I need to write down my thoughts, might as well do it here. Maybe pictures will wind up here eventually. Starting off, here's a loose drawing of our route:

In a bit more detail, with some highlights:

Down through WA towards Boise (the route I usually go) though stopped at Richland WA which was surprisingly nice, one of the 3 WA candidates for me. Boise was a surprise too, never thought I'd want to live there but it's kind of nice... at least the northern side of it. Then over past Pocatello and Driggs to reach the border of Wyoming (where I think I want to move to). Those west tetons are great... but out of my price range

Anyway went through those mountains down to Jackson Hole, then down the western side of WY to Evanston, detouring to SLC (showed friend the temple square area of the capital, got a park pass at an REI, then went out to the great salt lake and he almost killed his drone flying low and fast), then towards PG the long way through Tooele and Eureka and up the west side of Utah lake, drove past all my old homes before heading to my dad's. Next day went up AF canyon just to the reservoir, then met brother at Sundance (he got a job there, yay discounted meals). Back to Evanston WY from there, up to Kemmerer this time, and through the central valley part of western WY. Detoured a bit but made it back to Jackson and to Yellowstone, stayed on the western side for a couple nights

Yellowstone was surprisingly not super packed. About as busy as I remember it being last time I was there.

Anyway east from yellowstone through another route, Dubois WY (the area around there I really liked). Down to Lander, also a good candidate. Continued south and then east to Cheyenne, back north to Buffalo (next to another forest region). From there east to Mt Rushmore, then more east, dipped into Iowa for a night and as I predicted it was nothing but corn from the moment we crossed the border. Back up and through Minnesota, then some Wisconsin, then the upper peninsula of Michigan and entering the main state over the north bridge. Best fudge ever there, though apparently there's even better fudge on the island next to the bridge. =P

Friend's parents happened to be in Michigan (he's from there) so we had a lunch with them, and a dinner with his brother on the other east side of the state, then down back to the center area around Lansing where he grew up. Spent some time there going back and forth for him to check places out. Then down to Indiana, went through one of the Amish towns, just west of it is Das Dutchmann Essenhaus where we stayed and had an amazing fried chicken buffet.

More south after that through Kentucky, Tennessee, Arkansas, hit Texas. Checked out Dallas a bit (mostly the northern side, there's a sub-town where a lot of tech companies are). It was a potential candidate for me but I've now decided it's not worth it unless there's a job attached... Continued west through Texas and eventually New Mexico, checked out the UFO museum at Roswell. Google tried to route me around an unconstitutional border patrol stop in the interior, which was basically the same as the Canadian border crossing, but the guard ended up waving us and several other cars in front of us through without stopping so I didn't get to ask if he's aware of a dirt road that bypasses the whole thing.

Friends parents have their actual house in Arizona, just south of Tuscon, so we made it there for a night (they were still in Michigan though). Continued up through Arizona, through Prescott, then hit the grand canyon for a bit, back down towards Vegas... had to burn an extra day to save $500 because Vegas hotels are absurdly expensive Saturday night but not other nights. Stayed in Vegas, original plan was 2 nights but ended up staying a third since friend got minorly rear-ended the first day and wanted to file a police report (yay dashcam footage). Saw Penn & Teller, probably the highlight of the trip =P Like Dallas though, I now no longer want to consider Vegas area unless a job was attached. Way too hot..

After that headed back home, went through St George then back over to Nevada to look at the eastern side, back up to Twin Falls and Boise and then home

So, been home for a bit now, and returned to my usual habits. But that can't do! I need to work on this a bit while memories are fresh. Even if no decisions are made until next year, I want this post to be something I can look back on. I also am almost given up on finding an ideal location this year because of the crazy housing cost rises, but I can still maybe find something better than I have, and something that puts me a bit closer to the ideal.

One useful exercise is to come up with lists of things I care about (wants and not-wants about candidate locations and property attributes, mostly), and then try to rank them into an order of care-most and care-least. Since I don't think I'll find an ideal right now, I won't have any true must-haves. You can get fancier with a full scoring system, but I think instead I'll just try and tag sections/items if there's something more to them; like for some X I might not care much about them, but if I have them, they'd be nice. For other X, I might not care at all, really, whether I have them or not.

1. Being able to visit in person casually with some friends I've made around the Bellevue area, and not have to make a trip out of it
2. Being closer than the current 12 hour drive to what's left of my family
3. No additional state capital gains taxes
4. Sales tax rate (almost everywhere is lower, though Montana and Oregon have 0...)
5. Useful trees on the property -- I like pines/spruces/firs and quaking aspens the most aesthetically, pines and others should be big enough to use for a theoretical log cabin or some structure
6. 100 Mbit internet -- however Starlink would be very doable, and maybe cellular service until Starlink if there's nothing else
7. Rocky mountains (living on a mountain is best)
8. Four seasons
9. Access to a real grocery store
10. Gas furnace
11. Gas stovetop
12. AC
13. Garage
14. Option of off-grid electricity/gas/water/sewage
15. On-grid electricity/gas/water/sewage
16. 2-bed 2-bath minimum, preferred even if by myself, though in that case 1 bath would be ok
17. Access (less than an hour) to a dentist that practices biomimetic dentistry
18. Can go outside home and not see any neighbor
19. Neighbors at least a good distance away
20. Great restaurants
21. Access to airport

If there's something not here that you personally care about, and are wondering if I do too, perhaps I forgot, but also perhaps I just don't care at all and maybe if asked about it would be happier if such things were not present (e.g. bars).

Summarizing some thoughts on recent AI Doom

(Wrote around May 2023.)

Just making a more permanent record of some off the cuff thoughts from the bird site, which could go up in flames at a moment's notice and while I do daily archives of images/videos that I hit the "like" button on, I don't archive my own tweets except occasional yearly/not-even-yearly twitter data backups. And while I'm at it, might as well expand or put some context on some things.

Unless you've been living under a rock, or just don't care to follow tech news at all (fair enough), AI news has been gaining more and more attention. Especially with the release of GPT-4, it has caused AI safety people to pull the trigger and launch a wide-scale media campaign to draw attention to the issue that capabilities progress is increasing very fast while progress on "alignment" / Friendliness / AI-not-kill-everyone Theory continues at the usual glacial pace it's had for the last 20 or so years.

I've long held the idea that the only real problem humanity faces is ensuring a positive Singularity, mostly in the sense of I.G. Good:



That is, if nothing big gets in our way -- like a planet-ending asteroid for instance, but by big I mean existential threats, not horrible but survivalable things like a plague that wipes out 90% of the planet but where everyone else is immune -- such a technology seems both very feasible and very desirable. Beyond the problem of building it is the problem of making it actually care, for sufficient meanings of care, about humanity's desires, and stay caring even as it makes modifications to itself (presumably to become smarter and more efficient and capable at solving problems).

I was convinced of this by readings from the SL4 mailing list, Yudkowsky's work (from mailing lists, off-blog pages, Overcoming Bias, and finally the Less Wrong sequences; Harry Potter fanfic was entertaining but not enlightening to anything new). So you could say I'm in the MIRI-style camp of the issue. By default, it seems that if we do create an intelligence equal to / very quickly superior to our own, the likely outcome is that we all end up dead soon after. The way to avoid this is to have very solid reasons to believe such an intelligence would not wipe us out, directly or indirectly. These reasons, should we ever find them, will probably be in the form of first order logic, math, and code, not abstract English arguments.

A lot of very bad abstract English arguments have been put forth over the years for why this "alignment" of interests isn't a concern. One of the early ones was someone who thought we could just tell the AI to optimize for making people happy. As a way to measure happiness, people's frequency of smiling could be used to gauge it. The end result is that a super-intelligent AI will want to figure out a way to convert every molecule of matter into something resembling a small always-smiling face. I think this is where the later paperclip example came from originally (Bostrom did not originate it), but I could be wrong and haven't verified this sequence. Perhaps ChatGPT knows?

Anyway, that class of arguments seems defeated by pointing at the orthogonality thesis, which in a nutshell is that intelligence is a separate thing from goals, and it's consistent for something to possess a high degree of intelligence while only wanting to pursue some really dumb idea like maximizing the number of paperclips in the universe, but it could just as well be to provide as much gold to its creators as possible, or many other problems related to obtaining material goods.. Some researchers don't believe in the orthogonality idea, fine, at least they're explicit about it, even if I haven't bought their arguments.

Worth rereading:
https://web.archive.org/web/20070927211335/http://yudkowsky.net/singularity.html
https://web.archive.org/web/20000819020750/http://dmoz.org/Society/Philosophy/Current_Movements/Transhumanism/Singularity/
https://web.archive.org/web/20000815053120/http://dmoz.org/Games/
https://web.archive.org/web/20000815052645/http://dmoz.org/about.html

Rant about modern soft language and words no longer meaning things

(Unsure when I wrote this, but probably in 2023 or 2024. I occasionally think back to it.)

Being lied to, or misled, or not getting the whole context, is not gaslighting.
Being unsure of yourself is not impostor syndrome.
Being sad, melancholic, or cynical is not depression.
Being obsessive about someone, especially if there's at all a bidirectional communication channel, is not parasocial.
Wanting to spend time on social media instead of paying attention to a dull math lecture is not having ADHD.
Not identifying very strongly with a particular Hollywood trope representation of your gender is not gender dysphoria.
Being annoyed by a crooked picture frame is not having OCD.
Being nervous about giving a talk is not social anxiety disorder.
You are not more cool from having more self-diagnosed problems/excuses you can apply to yourself.

Nippon 雪月花 lyrics

(Tried to write this in late 2022 after growing to like this idol quite a bit, but didn't finish my translation and so left it incomplete. It's still an incomplete mess. I did post a French song though...)



The song is also available on practically every other platform, please support!

(Ed: actually there are official lyrics posted that I have access to now.. and anyway, I should just redo this whole post, but whatever... indefinite draft state.)

I liked this song so I made my own translation of it. Please note my Japanese level is still very low, this was made with the assistance of machine translation and dictionary lookups, with my own bias for more literal translations (as opposed to smartly localized as maybe should be for songs), so there may be (probably are) severe errors... I also want to thank じょなさん-san for providing a transcription of the JP lyrics (along with a wotagei draft), I've assumed his version is correct (maybe it's the official version?) where my ear/AI speech-to-text differ from it (especially kanji).



The message is really nice. It reminded me of another song about connections from my favorite rapper, Eyedea:



----

On resisting anime until 2011

(Started writing this I think in either late 2025 or early (Jan) 2026.)

dbz and pokemon were my first 'animes' in the 90s but I just saw them as cartoons, like extreme dinosaurs or spiderman. It wasn't until 2011 when a friend finally convinced me to watch/I finally relented against watching the ghost in the shell movie (and then the two seasons of SAC), was surprised by how much I thought it was actually good even if not 'new' to me in its sci-fi ideas or philosophy. It was still a fantastic presentation and while not my favorite anime possibly the best anime of all time

My main issue was: it's unlikely to show me something new.

But I mellowed and realized that's ok. Sometimes it's enough, at least as far as making me happy and clap like a seal, to see some themes and topics being broached at all in media.

Another example is Dark Souls. It's maybe my favorite game, though it's hard to have a singular favorite since I like so many different things about different games. Nevertheless, as MrBtongue put it, back in April of 2012, noting that Dark Souls came out in the fall of 2011, it's one of those few games that so brilliantly demonstrate Gesamtkunstwerk -- a game of total art. It's not just the gameplay that makes Dark Souls amazing (and some people really bounce off of it and prefer the faster iterations from later installments like Elden Ring, while I still think DS1 fits DS1 best and would like to see them make another attempt at this pacing). It's not just the level design, though it's very good too. It's not just the themes, the visuals, the overarching silence broken by the character themselves and occasional orchestral pieces for boss fights... It's not just the architecture or world building (even if some people bounce off world-building-through-item-descriptions)... It's all of that put together. There's a gazillion VaatiVidya videos about a lot of these things, and still to this day new people make videos exploring themes of Dark Souls and its connections to real-world real-history icons and religions.

And yet... it's just a swords and sorcery dark fantasy game, with kind of bad sorcery. The themes of life and death, of struggle against adversity, of human agency and free will, of order and chaos, of cycles and rebirth, of gods shirking their duty or choosing a corrupted path, they've all been explored in other media. I don't think there's perhaps a single really unique thing in Dark Souls thematically. Even the minor themes explored through NPCs aren't unique, but many players won't pick up on them at all. But that's ok, it works so well, and it's so fun, and all of these things and more are present in a singular medium, rather than focused on individually and scattered across several pieces.

And yet again... while I've really enjoyed every game after, and even Demon's Souls when I went back to play it, I still put Dark Souls at the top. The rest of the games feel like riffs of the same underlying things in DS1, and minor polishing of other elements, with regression in other areas. They're all excellent, but they don't really do much new, either in taking the themes somewhere new or introducing more themes at all.

It's also somewhat unfair of me, but again part of this earlier framing of my opinions about things, that I can't take the mere presence of something as a sign of quality or deepness or another good attribute. Let's take the crestfallen warrior for example: from a gameplay perspective, he's there to offer the player advice on what to do next, since there's no quest guide markers. He's also a warning, since he's likely to mess you up if you attack him on sight. Thematically, he shows the spiral of depression, and feelings of inadequacy, and eventually he goes hollow and will attack you on sight.

On the one hand, cool, this minor NPC has got some characterization and explores an emotional theme to some extent. You aren't getting that from a Goomba in Super Mario Bros. I think it's nice to see and like it when games do it, especially in a way that's optional and not forced. You can "save" him by not speaking to him at all!

It is sad that most players won't engage with this at all, and of those who do engage, a lot won't pick up on the obvious themes, or will stop engaging after the first encounter, or some of them who continue talking to him just don't parse the object level of what he's saying. I recently saw another player come up from Blighttown, speak to him, and totally ignore the bit about "a new problem". They were too distracted by not having the bonfire and just went straight towards Andre's bonfire instead. So, they missed seeing Frampt entirely.

On the other hand, it's not that deep. We don't really know anything about his life, his friends, his relationships, his family, how he grew up, his opinions about other things. He doesn't join us for any adventures, he just sits in his spot, rotting, until he loses it.

There's a type of person who looks at any media that shows some metaphor or exploration of depression, say, and thinks "Cinema! So well written! Such character depth! People who don't pick up on it suck! People who don't like this media must not have picked up on it!"

I really dislike that way of looking and thinking about things.

(Added Jan 25)

This was also a long-winded way to say that while I enjoy watching JJK, I think it's overrated by its fans, and it has severe flaws when it comes to writing, characterization, and direction. I'm an anime-only but couldn't help being spoilered on some future manga stuff -- the spoilers don't make me hold any different opinion. My opinion on JJK after s1 and knowing nothing of the manga has been: it's a fun action anime that does nothing new but is competent in what it does. After the movie and the second season, and then hearing some manga spoilers, and now as season 3 has started, my opinion hasn't really changed. Some minor nuances would be: it's much weaker on characters, narrative, and pacing after season 1, and what could have been truly incredible animation in season 2 was hamstrung by (reportedly) horrible working conditions. They still put out something amazing (even if there was one or two jarring cuts I hate) despite the odds -- unlike OPM S3 which put out trash no matter the excuses -- and they fixed some things up in the blurays -- but yeah, it was compromised.

Compare JJK's S3E4 with Frieren (any episode), just in the way new and side characters are introduced or reintroduced and even killed. Frieren does it so much better. There's also no omniscient narrator as became common in JJK since S2, which is such a reliable sign of hack writing.

Every life is wasted, or no life is.

(Started writing Jan 25, 2026.)

All humans have been, thus far, mortal. Only a handful have had any sort of lasting impact through time, most of us only accomplish a little bit in our own time and then we're gone, with the most impact over the long term being genetic continuation with offspring.

When people deride others for "wasting" their life, how are they any better? They will die soon, too, at least until biological immortality is achieved (at which point I will be willing to reconsider this, but I don't think my conclusions will change all that much).

So many people have died in their 20s or younger -- were those lives any more or less wasted? Some left an impact through music, some died before they got going, some left negative impacts with their hurt to other people, some left children behind, some didn't. Who can say which lives were more or less wasted? They're all dead well before the 'normal' life of 80ish years.

What makes the young retired engineer's dull life of reading and playing games more or less wasteful than an old retired manager's dull life of watching CNN and RV travel?

For careers, what makes the rocket scientist's life more or less wasteful than a mountain biker's life?

We can say one provides more utility to humanity, but even that measure is hard to do accurately, and besides, it's a mistake to conflate utility with morality. (Sorry utilitarians.) It's all stupid pontificating and some people trying to feel better than others.

Make money and die, that's the American way, it don't matter what name you gave the bucket that you play -- How Much Do You Pay? https://www.youtube.com/watch?v=DwnQj4o6uig

LeGuin also has a line about "deserving" that I think expresses a similar sentiment.

And yet this is strange thinking -- when you're tempted to apply a label to either all or none, or almost all or almost none, it's either a very poor label, or you're approaching it from a strange point of view. I think some people do this with the word / label "bad", applied to people: either we're all bad (sinners) (because we all have done bad things sometimes?) or none of us are. But I think that's wrong, bad can be properly applied to an individual empirically based on the degree and frequency of bad actions they do. When it's successfully applied, it has predictive power in that bad people will continue to do bad things, and different kinds of bad things too. However there must be room for repentance and no longer becoming a bad person -- when the label stops fitting because bad actions are no longer done at any higher frequency or intensity than normal humans.

I suspect a lot of people just don't want to apply the label to people at all, and especially not to consider it applying to themselves. There's a persistent and false cliche that "no bad person sees themselves as bad" or "everyone is the hero of their own story". No, some bad people know they're bad, and some people enjoy doing bad things and know they are bad things.

To me, the line that divides a bad person from someone who is not a bad person (not being bad doesn't necessarily mean good) must be somewhat fuzzy and permit crossing from either side to the other over time, i.e. it's not a fixed permanent label and it's entirely appropriate in some circumstances to change "is bad" to "was bad (but is no longer)". But it would be a mistake to think there's no boundary at all, or that the label should apply to either everyone or to no one.

More on LLMs

(May 18, 2026)
I wrote some stuff 3 years ago above, my core opinions haven't changed. I don't think our situation has changed much either. Trump 2 got elected, and the VP is strongly anti-regulation for AI, so the surprising "wait they're actually listening?" bits of the previous administration when it comes to real AI risk vaporized. Now we've got Bernie (lol).

Capabilities have increased. But not tremendously. The systems can do more useful things, faster, with less supervision; they are tying together even more bits of context and finding security issues, interesting biological research threads, and taking on much bigger coding challenges.

Alignment research has gone (as far as I can tell) basically nowhere. It's been more commonly called "interpretability" these days. Still not very encouraging research. They are still basically black boxes, though black boxes that can be vaguely pushed in certain directions, and distilled.

This year was the first year I've thought since my quasi-retirement that it actually might be kind of hard to get a job again. At least a lot harder than every year prior. I have played with Claude Code some, but only with a local model, I just find the UX terrible and hate to let these machines do work for review instead of just doing it myself. So I do most things myself, still, when I bother to do anything. At most I'll "consult", but then I'll consult with a frontier model in the chat interface. And those have gotten better. (And when I need a lot of context, Gemini has been great for dumping stuff into.)

Even when the tools get better, I'll still do a lot of things myself. I've described programming in Common Lisp as if having a heroin needle sticking out of a vein; it's just so pleasurable compared to other kinds of programming, I don't even care so much what I'm programming or how much (or little) progress I'm making. In my little game port project, it's so fun to just make some changes, recompile the system, and see everything take effect at once. Or sometimes there's a bug, but it doesn't crash, the game is just paused, so I inspect things and then fix it and resume (skip that frame) and all is well. I implemented a single frame step / pause feature incrementally, it was great, and pretty easy. So having AI do that misses the point. I'm not opposed to having the AI do some things for me in the future, especially once it gets better, but I want to still be doing something. I still want the needle hanging from my arm, recently pumped heroin circulating in me. If I left it all to the AI, that'd be like going cold turkey. I'd die.

I've been planning this ever since accepting the inevitability of AGI (either killing us all or leading us to Friendly AI utopia). It fundamentally doesn't matter to me that an AI is going to be faster, better, smarter, etc. at everything compared to me. I still have my life, and I want to fill it with what I please. If my form of heroin is programming in Lisp, so be it.

In the book Excession, there's a character by the name of Gestra Ishmethit who is my favorite and is my most memorable thing from that otherwise meh book. He is a hermit who has spent years growing his own forest and making tools and scale model naval ships from the wood. Completely pointless in the world of the Culture. But he enjoys it. I look at that and see a kindred spirit. I probably won't ever take up such a grand hobby (but you never know, in the vastness of the future where death is defeated), but I can be certain to take up similar pointless pursuits. Indeed, all of my programming hobbying post-retirement has been mucking around, pointless self-indulgence. There have been a few instances of programming something for someone, a few bug reports or patches here and there to other projects, but it's mostly self-service to myself.

In this transitionary period, though, where our AIs are kinda smart but also still really dumb, and even more our corporate culture is embracing them in dumb ways, I do feel a bit like a horse put out to pasture. i.e. the feelings of difficulty in getting a new job. It's not that I've become a worse programmer, or that I can't use AI tools, but I don't have the same flexibility as someone who's been using them all the time and happily (or unhappily) undergone the pain of the treadmill as things come and go. I skipped copilot, the vs code integrations, etc., and have barely tried claude code. It's all stupid. But I see why businesses like it, and it has made me more productive in some things, it's a valuable tool, with so much gifted to us for free.

Another way of looking at it, in my imagination at least, is comparing our time to that of the printing press and mechanical typesetting losing out to digital printers. Here is also where some of my irritation with anti-AI people (especially if they're programmers) comes from. We are mechanical typesetters. We have already embraced a significant level of automation, causing it and being affected by it. The world of handwriting is dead or dying. Recently the typewriter was invented, which has spurned our productivity even more because now we don't need to do as much manual typesetting, though it still is very useful for mass production of books, newspapers, magazines, and so on. We will never have a "burning of the library of alexandria" moment again because our important books are replicated orders of magnitude more often, instead of being limited to handwritten copies.

Now along comes the 1970s and digital printing and layout, and bam, mechanical typesetting is completely dead. Or at least relegated to the same niche uses as handwriting. I see a similar parallel with how LLMs are progressing. They aren't quite there yet, they're more like dot matrix printers if anything, good but not fully replacing us, but that time doesn't seem far because they're so close even if they never get to full AGI.

And it's kind of sad? Like, this skill I made from hobby to career will no longer have much economic value. Taste of our own medicine, perhaps, as we happily inflicted the same on many others. But in all, I find a lot to be cheerful about. There are many dumb facts I've learned over my career, some I've forgotten, some are no longer relevant, but so many dumb things. Syntax obscura, semantics obscura, dead platforms, dead frameworks.. it's inevitable in any career, you can't escape it, you can't perfectly optimize learning only things that will have relevance in the future. But the nice thing with LLMs is that they cut down on having to learn *more* of that sort of crap, by so much. I've let myself get out of date on web standards and features, but that's fine, because I'm letting the AI design any web pages I make in the future, I don't need to refresh and learn about all the new stupid stuff that obsoletes some of the old stupid stuff I had to learn. LLMs present an even higher level programming language that frees us from having to care about even more tedious minutia more of the time.

The anti-AI people have seemingly gotten more hysterical and I really hate it. They pollute the environment everywhere they go. I can only roll my eyes so hard. I feel bad for them in some ways, because they've been perpetually wrong these past few years and will continue being wrong going forward. I also find it sad how many go into a rage about AI art but have no idea how much AI is already in software they're using, hidden. And of course there are many programmers who go into rage at AI use in code. I understand why some programmers will hide their usage of it, rather than disclose it, to avoid these rabid freaks.

On the other hand, I also understand somewhat some of the freaks' feelings. There was a really terrible final lightning talk at ELS this year. The dude just threw some shit together and giggled the whole time about how he didn't write any code shown, it was all slop generated, giggling like a child bragging at show-and-tell that he stole some cookies from the cookie jar and got away with it. Not only is that not what show-and-tell is about, it's just a shitty act in isolation. Whoever encouraged him to whip something up should be ashamed as well. I'm generally very positive about AI, and AI-assisted works too, but this kind of pointless shit.. no one at a programming conference for programmers wants to see your public masturbation dude.

I do get annoyed by so much of the slop, really. There's even a first slop comment here on my blog that made it past my dumb captcha system and was posted the other day. I'm leaving it for now because I don't know if deleting it will trigger "retaliation" and force me to put in some sort of effort. I sympathize so much with open source project maintainers having to deal with so much shit slop pull requests. I hate having to read AI-speak with their sycophantic mannerisms and reddit-esque ways of phrasing things. There is so much to hate about our current transitionary time, these systems are retarded and people hyping them up like to pretend they're not or like to dismiss all retardation with "this is the worst they'll ever be". That's true, but it still sucks now, man!

I'm mostly positive about the security bugs direction things are going. Yeah it sucks that a lot of us are finding out our pants have been down for years, but that's just the way things go with security and people's continual refusal to stop writing C and C++. Whether it's bots finding vulns or people finding them, disclosing them (and ideally fixing them) is a good public service. Even if it sucks that we can maybe predict that for the next year+ we'll have to be a bit more vigilant in keeping things up to date. And again, AIs that file false/hallucinated issues suck, it's just unwanted noise.

Back to my core opinion about AI... still is that the default outcome for AGI (or ASI as it's more commonly called now) is human extinction. I'm not as pessimistic as EY, but I also don't have any real argument against EY, and I haven't seen anyone else make a decent argument either. Lots of other idiots and retards have arisen, the latest retard even gave him $10k for the privilege of having his idiocy broadcast to the world. My own best "arguments" against the idea that AGI's default outcome is human extinction are from a mix of Hanson, Goertzel, and a smattering of others. It's two parts, though they are somewhat related. They're also not so much arguments as hopium.

The first is roughly "progress (and disaster) are incremental". The hope is that progress continues being somewhat steady, as it has been for history, and that the magnitude of progress and disasters will also be somewhat steady, with disasters especially not likely to be sudden foom and extinction as that's not steady. For now, "disasters" have ranged from miscounting letters in words, hallucinating data or citations, wiping out software production environments, and driving unstable people further into forms of psychosis. Before escalating to extinction level disasters, the hope is that first they might "merely" cause anywhere from a kilodeaths to a couple gigadeaths instead. At that point, when the risk is so much more clearly not just theoretical at all, surely people would take alignment more seriously, including proposals about global regulation for hardware to try and mitigate one path to further capabilities.

The second is roughly "neural net architectures can't ever scale to world-ending capabilities or meaningfully accelerate research on alternate AI architectures that could, at least not without also meaningfully accelerating alignment research on those alternate architectures". The hope here is that while sure, deep learning architectures basically obsoleted every other AI technique over the past 100+ years all within the past 10-15, and capabilities have gotten quite good since even the last 5 years, and they keep getting better every 6 months or so, nevertheless the increments seem to be stalling out and in the bigger picture there still isn't a lot to show for it all, therefore we don't need to be any more worried about AI extinction concerns than we were 20 years ago.

There is some overlap here with the first part, e.g. every automation advance in the past led to lost jobs but further economic efficiency and this is just another example of incremental progress. Both parts are still based on hope that things will just work out, though.

So what's my plan? Continue trying to enjoy life. Be thankful that I was able to have a little career and instead of spending all my money invested most of it instead so that I have a buffer large enough to perhaps not have to work again. Mostly keep my head down in social environments I hang out in where the anti-AI sentiment is irrationally high. (That's another thing: I've been against intellectual property for a long time, arguments around "theft" of the AI companies will not reach me, especially when they all offer free tiers and others have made pretty cool free local models by distilling them.) If I see something that can reasonably benefit from my marginal input as a way of trying to make the singularity a positive one, I'll consider it, but not much is looking great so far. I have many qualms with the global regulation approach, not least of which being "it doesn't work" (long term), though it can delay things a little while and it's at least a serious proposal with hundreds of policy pages written on it. It's a thing that could be done, and doing it, would set the precedent for being able to do things, so in the future, something better might be done too. We'll see.

I'll also keep occasionally using LLMs as tools, much like I use vim, but the heroin needle stays in.

---

Posted on 2025-02-13 by Jach

Tags: fodder

Permalink: https://www.thejach.com/view/id/441

Trackback URL: https://www.thejach.com/view/2025/2/drafts

Back to the top