And every time I ask it to do the same thing, it produces basically the same result. It will sometimes not produce a go.mod, but that’s probably because doing so isn’t as statically high as writing the code to sum numbers from stdin.

So going back to the understanding of how it generated this, is quite simply the most statistically relevant search space of it’s weights it has been trianed on and it has basically just produced a series of tokens, one after another that are relevant to the input, the next token and so on. It’s a trivial example I know, but it basically pattern matches it’s way through it’s vast search space just producing outputs based on context.

@movq@www.uninformativ.de I think your points are pretty clear to me, that’s fine. I’m just seeing if you can perhaps see things a different way maybe?🤔 I would challenge the assertion that you cannot understand how Claude Code generated an output; which I can demonstrate easily with a fairly trivial example by the input:

Write a program in Go that sums a list of numbers from stdin and prints the result.

You can basically think of this as pattern-matching. I’m very very good at very fast pattern matching and piecing pices of a puzzle together very quickly, sometimes with very little to go on, it’s often gotten me into a lot of trouble at work in my career because I can make a lot of assumptions very very quickly.

@bender@twtxt.net Fine, Let me answer properly and concretely 😅

Would you want your children not to learn anything, because “they have AI”?

No, children still need to learn. That will never change. What they learn however will over time.

Are you OK with your children using the AI for all of their homework?

Yes, frankly I am. Why? Because much of what we teach them in school is utterly pointless.
For example, learning to read Shakespear never taught me anything useful in my life. I regret much of my school years to be honest.
I leanred to read and write, sure. But I learned Math, Science, Computing and how things work on my own by being very curious.

What sense will it make?

That assumes I answered “no”, which I did not. So it all makes perfect sense :D

What kind of future would that bring for them?

This assumes I said “Yes”, which I did :D It will be an itneresting future that’s for sure. I don’t think we can just bury our heads in teh sand and pretend it’s all going to go away, It will not. It will make things very interesting for sure, as we’re already starting to see what’s possible and what’s changeing. For example; ordinary people are using these LLM(s) to write their legal suit and defense in courts with varying levels of success.

Even if AI were to become omniscient, what will it be of the human race then?

I’m not convinced it ever will. In fact, I am not convinced we know how to create true intellience at all.

What would we do?

What would be so different from say an Alien invasion from far superious beings?
What would we do that? Band together and defend humanity?

Serve the AI? Maintain the AI?

That assumes that “AI” will become intelligent and omniscient, which I don’t believe it ever will.

Would we have found the true meaning of life then?

If the meaning of life is to create our own sub-species liken to ourselves, sure, maybe. But is that even a reality? not sure, I doubt it. We barely understand ourselves at the best of times, let alone how our minds works.

To care for AI, Is that it?

How would this be different to caring for a friend, a family member If we could ever truly reate an actual sentient being with real feelings and intelligenace, is there any reason to worry? Could we not be freinds and have mutual goals and form relationships?

Aha, yesterday’s newly added support for LC_TIME to render localized timestamps also broke the feed parsing with my LANG=de_DE.UTF-8 and LC_CTYPE=de_DE.UTF-8 environment. :-)

Atom feeds make use of RFC 3339 timestamps. They are first converted into RFC 882 timestamp representation, which is the one that RSS feeds use. However, this conversion now results in localized RFC 882 timestamps, which cannot be parsed into Unix timestamp numbers via curl_getdate(…). I bet that it doesn’t know about the localization at all and expects English month and weekday names. Looking at its docs, I reckon that function was selected because of its myriad of supported timestamp formats: https://curl.se/libcurl/c/curl_getdate.html RFC 3339 is not included, though, hence the transformation up front.

The intermediate Item objects in the parser domain use std::string for the timestamp representation. This isn’t all that silly, because Newsboat supports all sorts of different feed formats with different timestamp formats. These RFC 883 timestamps are centrally parsed into time_t.

Speaking of time: It’s time to go to bed after this late bug hunting fun. :-)

@movq@www.uninformativ.de Thanks. I noticed the <updated> of the feed, too. But for some reason, some articles were suddenly marked as new.

On some YouTube feed <entry>s, I noticed updated <updated> fields showing today’s timestamps. But unless there is no <published>, the <updated> is not even considered. I verified that in the source code. Yet, all the affected articles in Newsboat show today’s timestamp, not the years old publication timestamp. I generate the YouTube feeds from the original feeds myself once a day, so I doubt that this is cause by some YouTube shenanigans.

Very weird, it doesn’t make any sense at all. What is going on here? O_o It doesn’t appear that I have duplicates in the database either.

Of course, @movq@www.uninformativ.de! Most of my points are also included in your list.

First of all, programming is what I really do enjoy the most. So, it doesn’t make any sense at all to not do this anymore. “But you could use your now free time to do something much cooler and more valuable!”, others might reply. Fuck no, I don’t want to waste my time with other shit that doesn’t fulfill me, why on earth would I want to do that?

All this hallucination reduces quality badly. In my experience, it’s also happening much more rapidly than I expected. Even though developers are still supposed to own and understand whatever has been generated under their name and even be responsible for that, the sad reality is that teammates often blindly trust the AI output. “But I asked the AI and it told me that $this was impossible”, “I’ve no idea either, but the AI just generated it” are responses I get more often. What really makes my angry is when I point out a flaw and suggest an alternative and this is the reaction. It happened several times that just trying it out and seeing it clearly work to proof my point only took me half a minute, but people still did something handwavy else instead.

The learning effect is drastically reduced. The more time I spend on a topic, the better the odds that whatever I learned actually makes it over into long-term memory. It’s like if a collegue just says “do it like that” or “this solves your problem”, but neither explains the why or how. Somehow, people are still convinced that it’s a completely different story when you replace the human counterpart with a computer program in this equation.

Skills are unlearned. It’s like with automation in general, just much worse. You end up in a state where you’ve no clue how anything works under the hood or how to actually find out important information that are needed to solve your problem. You’re screwed when a process breaks out of the blue. Even though it can become also rather terrible, with classical automation you’re typically still be able to decipher how exactly the thing was supposed to do something.

The energy consumption is sooo high, I absolutely do not want to be a part in burning down our planet. I’m sure I find (and probably have long found without knowing) other ways to contribute to worsen our climate crisis.

The scraper part is already covered in detail in your list. :-)

I’m convinced that license and copyright violations are only played down or even refused entirely because companies want to make big money quickly. With the work of others of course. Their double standards are obvious, they still try to actively keep their own stuff secret and out of any training sets. At most for internal use only. Virtually noone in charge is interested in good long-term solutions. Short-term for the win, when disaster eventually strikes, the causers are long gone, the responsibilities in other hands.

Vendor lock-in is something that lots of folks are only realizing very slowly. It’s completely crazy to me. This drug dealer routine should be well-known by now. It’s fucking everywhere. Yet, people are always surprised when they found themselves caught in it.

Adding new AI stuff only increases complexity. But complexity is the enemy that everybody should fear and reduce as much as possible. Of course, this is not limited to AI at all. And everywhere I look around, people in charge looooove to make things way more complicated than they ever need to be. Yet, simplicity is the real art and much harder to achieve.

I don’t understand why we have to go back full force to the ambiguity of natural languages. This alone should be more than enough to realize what a stupid idea all that is. Linked to that is that the “instruction set” is interpreted differently with newer model versions. I mean, is has to be. Why else would somebody want to upgrade in the first place than to get more Powerful™ Features™?

Some people argue that with AI the democratization is empowered. However, in my view, the exact opposite is the case. Models are getting so large that you can basically not run them locally or even train them. So, you have to rely on whatever the vendor offers you and runs for you. In the end, this only gives the owners more power, the multi billionaires. Not exactly what I understand by democratization.

Finally, technology assessments are missing completely. Or they are faked such that mostly only the (questionable) benefits are listed. But all the negative impact is just ignored.

Let’s keep some popcorn around for when this all explodes. :-)

@movq@www.uninformativ.de If you really like to, you can try to negotiate with your employer that you can leave earlier. At least some mates were successful in that. I mean, it’s also in the company’s interest to not have to pay someone who has already mentally resigned long ago.

And on the bright side, you don’t even have to hand over anything. Your boss doesn’t have to look for a successor, so they can just let you go even sooner. This AI shit will simply continue whatever you did, no problem!!

It’s so crazy. I should probably also look for something else. :-(

@movq@www.uninformativ.de ooooh, yes! I am going to subscribe to that. Well done.

@movq@www.uninformativ.de What are you going to do about this? 🧐

I went 1-for-2 again at Magic today, winning the first game with my (mostly standard) Fallout “Hail, Caesar” deck by creating a swarm of soldiers and slapping people across the face with them (LOL!), before quitting the 2nd game for lack of time after my board got wiped (I mean, I might have lucked into something eventually, but it was getting late, so I dropped out).

I hope to play more regularly going into the summer, but who knows.

@tftp@tilde.town Ah, I see. I have a feeling that a lot of stuff is going on under the hood all the time and it’s mostly the userland-visible things that stay the same? 🤔 But yeah, some stuff is really, really old, like the TCP code I’ve recently (tried to) read.

@tftp@tilde.town hey, my pleasure! Have a great weekend! I am going to off for a few days to do a whole lot of nothing. Can’t wait! 😂

@a9b5c: Just don’t go all Richard Lopez ok?

@prologic@twtxt.net Yep, that’s what I’m going for. 😃 I mean, I don’t want to make an exact clone, just catch some of the DOS vibe.

@lyse@lyse.isobeef.org LOL! That’s brilliant. (We’re going back to that. Some Amazon employees probably already are.)

Many people started to become distrustful of big tech in the wake of the COVID-19 pandemic. I began feeling pessimistic back in 2016, when AlphaGo beat master Go player Lee Sedol four games to one. Something about that event has soured me on the future of technology ever since.

I completely forgot, I saw my very first badger in the wild the day before yesterday. :-) That was absolutely cool! <3

I heard something comparatively large rustling in the bush right next to me and thought that it must be dear. Naturally, I stopped and tried to see what’s in there. The rustling went up the bank and it suddenly came down again towards the road I was on. That’s when I first layed eyes on it and identified it as a badger. For a split second I thought that it’s going to get after me and was ready to get running. But it just hadn’t noticed me yet. When it eventually spotted me, it froze for a few seconds and ran off uphill. My camera took too long to boot, so it was already gone by the time the photo machine was good to go.

Germans don’t have humour? Oh, yes we do! Get this:

I was pushing my bike slowly up a hill, coming across an elderly lady.

She goes: “Go slowly!!! 😡”

“???”

“Haha! Just kidding! 😊“

@bender@twtxt.net There ya go! 8-)

@lyse@lyse.isobeef.org I see him on those photos, and his singing starts ringing in my ears. I also thought, “I am going to miss him” because, you know, nothing is everlasting. But then I reproached myself; “enjoy the moment, dude, don’t taint it”. And so I shall.

@prologic@twtxt.net I am going to give it a more serious spin (meaning I am going to go read the help page). I’ve got to tell you though, most successful games do not need a help. But I am fully aware that there is a subset of gamers that would not mind—if not appreciate—a game with help, manual, and the likes.

@kiwu@twtxt.net Working on my game Frontier Crown – Going to push a new version today hopefully that includes much improved graphics, expanded ruleset and scope.

@lyse@lyse.isobeef.org maybe they go after the impact. A single? Meh. An album? “Woah! These guy(s)/gal(s) are busy!” Also more possibilities for people liking at least one song. Anyway, that’s my theory, and I am sticking to it! :-P

With all these new ways of digital publishing, I’m wondering for years why music artists still release entire albums. I would have imagined that most bands simply publish a new song whenever it’s good to go. But no, at least in my bubble, everybody still collects a bunch of new songs before throwing them as a collection into the crowd. I never used any of these streaming services, though, so maybe I’m just completely uninformed.

There’s a joke for win fans. Q: I have a problem with my win system. Can you help me? A: Yes. Go to the dos prompt and type format c:, then press OK and all your problems cease within a minute.

I don’t know what surprises me the most. Is it the dangerous in red, or the fact we need a 21 minutes video about it. LOL. Sorry, sorry, shouldn’t joke about such safety nightmare. My bad. I will go to my lane now. 🙈

As an enjoyer of delightfully bad graphic design, found on most Czech village center cork boards, I’m sad to see the stolen clipart and badly cropped watermarked stock images, gradually replaced with AI slop.

This is far from a serious rant, but generating images of my kind being telepathically hit with sharp rocks, surely gives me a right to complain.

So far these seem the most prominent slop categories, seem to be…

Architecture slop:

1. find a sketch of what an old building looked like
   
   
2. generate an AI version, without correcting any of the perspective errors - this one is diagonally levitating
   
   
3. generate a recreation of the buildings demise - after going through the AI, for the second time, it is now a completely different building
   
   

Moralizing slop:

History slop:

Just cancelled my sponsorship of two developers on Github, sorry 😞 – I’m not going to sponsor going forward if no-one else can be bothered to. It seems silly to be the sole sponsor of another’s work or project 🤦‍♂️

@lyse@lyse.isobeef.org Those are some very colorful shots. 👌 It was pretty warm here as well, health issues prevented me from going out, though.

(Have we established that Azabache is male? 😃)

how is everyone’s week going?

@lyse@lyse.isobeef.org Yes, and that’s why I’m 100% convinced that we’ll see a massive brain drain in a couple of years. This will affect young people even more, because they don’t have all the “old” knowledge to fall back on.

It’s concerning, I’ve warned about it many times, nobody listens.

I think the best thing one can do is explicitly not use any AI tools but keep your actual skills intact. Might be out of a (good) job for a while, but once this bubble bursts, this is who is going to get hired again. (I think.)

And considering how insanely expensive all this is, I’m still (mostly) convinced that the bubble will actually burst. This stuff just isn’t sustainable.

… or I might be wrong. And if so, I see an even darker future that I don’t want to put into words right now.

@xuu@txt.sour.is, what’s going on with y’all up in the mountains? The mouse has been mighty quiet for a while!

@lyse@lyse.isobeef.org AI result ahead, feel free to ignore.

I “asked” the AI at work the same question out of morbid curiousity. It “said” that SQLite converts that integer to floating point internally on overflows and then, when converting back, the x86 instruction cvttsd2si will turn it into 0x8000000000000000, even if the actual floating point value is outside of that range. So, yes, it allegedly actually saturates, as a side effect of the type conversion.

I couldn’t find anything about that automatic conversion in SQLite’s manual, yet, but an experiment looks like it might be true:

sqlite> select typeof(1 << 63);
╭─────────────────╮
│ typeof(1 << 63) │
╞═════════════════╡
│ integer         │
╰─────────────────╯

sqlite> select typeof((1 << 63) - 1);
╭──────────────────────╮
│ typeof((1 << 63) ... │
╞══════════════════════╡
│ real                 │
╰──────────────────────╯

As for cvttsd2si, this source confirms the handling of 0x8000000000000000 on range errors: https://www.felixcloutier.com/x86/cvttsd2si

The following C program also confirms it (run through gdb to see cvttsd2si in action):

<a href="https://yarn.girlonthemoon.xyz/search?q=%23include">#include</a> <stdint.h>
<a href="https://yarn.girlonthemoon.xyz/search?q=%23include">#include</a> <stdio.h>

int
main()
{
    int64_t i;
    double d;

    /* -3000 instead of -1, because `double` can’t represent a
     * difference of -1 at this scale. */
    d = -9223372036854775808.0 - 3000;

    i = d;
    printf("%lf, 0x%lx, %ld\n", d, i, i);

    return 0;
}

(Remark about AI usage: Fine, I got an answer and maybe it’s even correct. But doing this completely ruined it for me. It would have been much more satisfying to figure this out myself. I actually suspected some floating point stuff going on here, but instead of verifying this myself I reached for the unethical tool and denied myself a little bit of fun at the weekend. Won’t do that again.)

Eehhh, what the hell is going on here!?

SELECT
    printf("0x%x", (1 << 63) - 2),
    printf("0x%x", (1 << 63) - 1),
    printf("0x%x",  1 << 63     ),
    printf("0x%x", (1 << 63) + 1),
    printf("0x%x", (1 << 63) + 2)

SQLite yields:

0x8000000000000000 (instead of 0x7ffffffffffffffe)
0x8000000000000000 (instead of 0x7fffffffffffffff)
0x8000000000000000 (correct)
0x8000000000000001 (correct)
0x8000000000000002 (correct)

Huh!? O_o Am I stupid? What am I missing here? Or is this actually a bug? :-?

With 62 bits, everything is spot on:

0x3ffffffffffffffe
0x3fffffffffffffff
0x4000000000000000
0x4000000000000001
0x4000000000000002

And 64 bits rather unsurprisingly also yield:

0xfffffffffffffffe
0xffffffffffffffff
0x0
0x1
0x2

@lyse@lyse.isobeef.org easy come, easy go. They grow so fast! :-) Also, Azabache allows to be seeing when ready for it, you know, just like Gandalf “*a wizard is never late, nor is he early, he arrives precisely when he means to*”. :-D

@movq@www.uninformativ.de with the current regime, everything is, to put it politely, rather odd, and in disarray. They have yesterday’s window, otherwise the next one was on the 12 of April, or something like that. We knew it was going up for a few days, but we are used to that kind of thing, so it is not that super exciting any more. LOL.

Yeah, I saw it in person.

This whole thing was pretty weird, btw. I had no idea it was happening until basically yesterday. No news coverage, nobody mentioned it. 🤔 And suddenly, boom, we’re going to the moon. What? 😅

@movq@www.uninformativ.de pretty cool, thanks for sharing! We saw the rocket go up yesterday, but it didn’t occur to me to pull out the camera and take a shot until it was gone. I mean, the visible window is also quite short! 😅

Hey yall. Anyone got Gas in Upstate Arkansas? My 1999 Honda Civic LX is Low on fuel and I need to go to places soon.

2nd Van trip coming up this weekend, taking Friday off work. Gonna sleep in the Van tonight and see if I can fiddle with the town water supply (basically our outside tap near the Van haha 😆) and see if I can have a shower in the Van, brush my teeth and go to bed 🛌 – Basically I just want to figure out the rest of the plumbing 🪠

There you go, user-defined color schemes:

@bender@twtxt.net Thanks for the tip-off, fixed!

I hope to have some time this weekend to tease apart my current setup and build a couple of example sites with it (while also writing some docs along the way). But given the rate I’ve been going, it’ll probably be another month. 😢

I thought that YouTube finally destroyed all the feeds, because I didn’t get any new entries in my newsreader for days. Now I realized that Newsboat somehow just froze. No idea what happened. This is the very first time ever in all those years. Haven’t updated the version for literally years. I reckon I will compile the upcoming version then. This will require a new Rust toolchain, that’s going to be great fun, I’m sure. Already looking forward to that…

@lyse@lyse.isobeef.org A-ha! That means you haven’t spent enough time with these tools! Go on, try it! (If you don’t, we’ll fire you.) I’m sure you’ll like it!

When are the RAM/SSD prices going down?

I’m supporting incremental SQLite schema changes to just upgrade from an older database version to whatever the current software version supports. In the past, I already noticed that this is quite expensive in unit tests when each test case runs through the entire schema patches and applies them one by one.

To speed up test execution I now decided that I finally go through the troubles of maintaining both a set of incremental patches and a full schema setup in one go. A unit test verifies that both ways end up with the same structure. This gives me a set of SQLs to check the structures:

SELECT type, name, tbl_name, sql
FROM sqlite_schema
ORDER BY type, name, tbl_name

Unfortunately, the resulting CREATE TABLE SQL queries are formatted differently, depending on whether the full schema was set up in one big step or the structure had been modified with ALTER TABLE. Mainly, added columns are not on their own lines but appended in one physical line. That’s why I wanted an SQL formatting tool. Since I didn’t find one that works decently, I’m now doing some simple string manipulation. Joining consecutive whitespace into a single space character, removing spaces before commas and closing parentheses and spaces after opening parentheses. This works surpringly good enough. Of course, if it fails, the “diff” is absolutely horrendous.

Now for the cool part, my test execution dropped from around 5:05 minutes to just 1:32 minutes! I call that a win.

I just stumbled across PRAGMA table_info('tablename') https://sqlite.org/pragma.html#pragma_table_info, PRAGMA foreign_key_list('tablename') and friends. I guess, I have to play with that, now. It’s probably much better to use than the SQL text approach.

Turns out, I even go down to only 50% quality for my thumbnails: https://lyse.isobeef.org/tmp/galres.txt The difference between 50% and 80/90% is just barely noticeable.

$ convert -strip -quality 50 IMG_20251106_035048_448_size_400.jpg 50-stripped.jpg
$ convert -quality 50 IMG_20251106_035048_448_size_400.jpg 50.jpg
$ ls -lh 50*jpg | awk '{print $5 " " $9}'
26K 50.jpg
25K 50-stripped.jpg

@rdlmda@rdlmda.me most of our conversations used to be about twtxt, I am not going to lie. Lately? Not so much. It turns out (a) we don’t need a longer hash, (b) we don’t care so much about changing addressing, and © I am just Bender, what else can I say? :-D :-P