dmitry kim (jsn) (@jsn13) profile picture
Moscow, Russia
Joined September 2007
Photo by dmitry kim (jsn) (@jsn13)Photo by dmitry kim (jsn) (@jsn13)Photo by dmitry kim (jsn) (@jsn13)Photo by dmitry kim (jsn) (@jsn13)Photo by dmitry kim (jsn) (@jsn13)Photo by dmitry kim (jsn) (@jsn13)Photo by dmitry kim (jsn) (@jsn13)
In our Heavenly Father's gambling house, many tables are rigged -- only to give us clueless schmucks better odds than we deserve. So, God, grant us courage to roll the dice, humility to accept how they fall, and clarity to know a good bet when it loses and a bad bet when it wins.
1
1
4
205
I don't get the hype around Ox Alpha. It struggled considerably in my simple coding benchmark (the one that e.g. Grok 4.5 and even GPT Luna one-shot pretty much instantly while looking bored). BTW, one thing I noticed, Ox Alpha on @opencode Go does *not* seem to be the same model, it's more stupid and/or somewhat glitchy (for example, it can't finish my benchmark on Go, while it can on OpenRouter). And btw, I've seen the same thing with Kimi K3: Kimi K3 on Opencode Go absolutely didn't feel like the same model as K3 on other providers (even on Opencode Zen, IIRC), it felt less smart.
3
196
So Sam Altman's story is basically that you either die a villain, or live long enough to become the hero.
1
50
Actually, even mid-tier coding models currently exhibit performance far beyond what humans can realistically do. I mean, a coding agent one-shotting IFCP2006 VM is like a human that, in a phone interview, hears the task description once, thinks for a while, and then is ultimately able to recite source code for a working VM in C, with proper formatting, without any errors, on the first try.
1
35
So, when you try to signup for tinkercad.com, and you're a child, they email your parent for consent before they let you in. But since you don't yet have an account, they generate some random name for you. And then they use that name in their email to the parent! What a mind job! So here's the email that I've got:
1
5
824
Talked to Fable in Claude Code about my health issues, got "safety measures flagged". I mean, seriously? :)
1
2
119
I remembered that I have a favorite programming benchmark, the one that I previously used to evaluate programming languages, that I can now use to evaluate coding models! I'm talking, of course, about ICFP Programming Contest 2006 virtual machine. So I asked models to read boundvariable.org/task.shtml and to implement the UM virtual machine capable of running `sandmark.umz`. I've made a few runs with a few models that are +- in the same price range on Opencode Zen / Go: DeepSeek V4 Pro, GLM-5.1, Kimi-2.6, MiniMax-M3, Claude Haiku 4.5 and GPT-5.4 mini (well, DeepSeek is way cheaper, actually). So, GLM, DeepSeek, Kimi: usually use C, usually get pretty close, but then invariably are confused about "self-decompressing" wording on the web page, which makes them run `sandmark` not directly on their VM (as they should), but using `um.um` (UM emulator written in UM) -- which, of course, is terribly slow. Then they get stuck trying random useless ideas to speed things up. I had to interrupt them at that point. Minimax-M3: usually uses C, is never confused about "self-decompressing", but invariably screws up op 13 every time. Sometimes it manages to dig itself out of that hole and delivers good results; other times it finds a way to dig itself deeper with some other mistakes. Haiku just did the whole thing in python (with terrible performance) and with some additional error causing `sandmark` to terminate early and without producing full output. Haiku then proudly declared that it has successfully completed its task. And GPT-5.4 mini just one-shots the whole thing every time, in Go or Rust, pretty fast, too -- definitely making it look too easy. It's a clear winner, and it's not close. I tried warning DeepSeek and Kimi about `um.um`; after that, DeepSeek one-shotted the task pretty fast, too, and Kimi decided that it has to ignore my advice, and ended up stuck trying useless performance tricks again. I tried warning MiniMax-M3 about op 13, but it found some new way to screw up the implementation.
1
86
So yes, we have our irreconcilable differences with China about rights/freedoms. It makes it easy to forget that it *is* one of the great civilizations, and that it's one of those that are very much based on a huge corpus of written texts -- which LLMs thrive on. Case in point: I have a medical condition for which I find the Traditional Chinese Medicine view quite helpful. I would argue that it might be better to discuss that part with something like Deepseek or Kimi than with whatever Western model is white-hot these days. Makes me wonder what other topics are like this. I suppose Chinese models would be more familiar with e.g. Chinese classical Chan texts ("zen" before it came to Japan). I'd also expect them to understand those texts better, since they are notoriously hard to translate, and Chinese models would just have a deeper understanding of the language of the original, see more original comments, etc. Of course, there must be many other topics like this.
1
41
Hot take: quite often, the most useful mode of using an AI coding agent is peer programming mode. That's strong evidence that the most useful mode of using a human coding agent is often also peer programming mode. The main difference is that humans are so damn expensive.
1
3
191
Do I enjoy the low-key ongoing struggle of keeping my everyday life and the space around me orderly and well-organized? No. But when it works, when I manage to get that predictable, efficient, frictionless flow -- do I enjoy *that* enough to make it all worth it? Also no. But ...
1
1
30
...still, when it works, when I'm not bogged down by friction of the mundane, not decision-fatigued from a thousand everyday nanodecisions, free to give it all to pursuing my big life goals -- do I then get the amazing results that *are*, indeed, the ultimate joy? Also no.
1
22
It's pretty much everything that "fuck you money" is, except for the money part!
25
Replying to @arkenoi
@arkenoi @adworse (OP мне там теперь ответить не даёт) нет, во всех приведённых вами примерах люди-умней-нас описываются исключительно как малопонятное явление природы, и совершенно никогда и принципиально ни в какой степени -- как герои, а речь была именно об этом.
1
1
116
Case in point: ГЛ. Мокрецы описываются только снаружи, любые попытки даже Банева, который a student of human nature, что-то в них понять *как в людях* натыкаются на стену. Даже дети, собственно, так же описываются всегда снаружи.
2
1
67
Более того, никакие их решения, выборы и стратегии тоже, кажется, никогда не описываются -- описываются только исходы и результаты, и даже реакции на эти исходы и результаты часто показаны амбивалентно, if at all.
1
66
dmitry kim (jsn) retweeted
ChatGPT refuses to say the name “David Mayer,” and no one knows why. If you try to get it to write the name, the chat immediately ends. People have attempted all sorts of things - ciphers, riddles, tricks - and nothing works.
3,011
3,491
1,534
51,393
10,467,571
Hard to believe someone would unironically touch the bottom row, yet, when I was very young, went through a period of seriously doing "Lisp style", but with 8-char indent. And that was way before I even heard of Lisp!
This Post is from an account that no longer exists. Learn more
3
5
151
dmitry kim (jsn) retweeted
I was looking up some history behind the x86-64 transition, particularly around the Pentium 4 time frame, and I found out that Bob Colwell (Pentium Pro chief architect) has been posting on Quora. Pentium 4 had a version of x86-64 that was fused off.
65
396
32
3,727
285,972