karmakaze

Fun nostalgic read (also Korean). I grew up an Atari kid, having spent many hours in arcades and eventually getting an Atari 400 included BASIC, and Asteroids cartridge. No storage not even cassette.

I had friends that had Ataris of course, but also some that had C64s and even a TI-99/4A. The thing in my circle was that the C64 and TI folk were more users and the Atari kids were more hackers, not that there were very many of each at the time. I learned BASIC, machine/assembly, and got into coding and engineering. (One of our pastimes was cracking games just to see how it was protected and see if we could. We'd show off any success amongst ourselves but never distributed as that wasn't the fun part.) I get the feeling that I probably worked for some C64 kids as adults that always seemed to have more of a product or business sense about them than the Atari kids.

LLMs should use "TK" or stable diffusion (and the like) so as not to get hung up on sequential words/thoughts and fill them in later instead of hallucinating filler.

I found the low frequency surprising as it's so easy to pronounce--I suppose tc is used in most cases. Here's what I found for bigram freqs near TK:

Ratios (count / total) and percentages:

    PG: 0.00047%
    TK: 0.00046%
    KK: 0.00045%
    HQ: 0.00042%
    FN: 0.00042%

My bittersweet story is in the same circle from another side. I grew up making games on the Atari 8-bit, living and breathing the machine. The spiritual successor was the Amiga Lorraine which my friends and I were following intensely. It had a slow/rocky start, was funded by Atari for a while, and seemed could be the actual successor. The '83 videogames crash shuffled everything with the head of Commodore leading Atari and Amiga becoming part of Commodore. This mattered a lot because my family had a computer store in a mall selling Atari computers and games. A place where people hung out and chatted about all such things.

Well anyway when the 520ST came out I was to get one but had to choose between the color or monochrome monitor. For some reason this felt like a crucial choice and it was: med/low color or hires monochrome. Surprising to myself I picked monochrome. It was fun learning to program a new computer but it wasn't exciting the same way as on the 8-bit. There wasn't any low-level hardware tweaking, it was more about CP/M 68k, GEM, etc. One thing I did do was try some C compilers with Megamax being the most usable (for floppies). I missed the 8-bit's support chip programming which was lacking in the ST and living in the Amiga. Getting interested in MIDI (with Amiga dropped but the ST picked up) would have been great but I wasn't musically inclined. Well that was largely my cross-over from dreaming of becoming a great video game developer into more of a business professional programmer. Later at a co-op job I did use it again to prototype fast line joint computations for the 68k without an FPU to be used in PostScript firmware in EiconScript[0]. I also wrote the PCL 5 (most not all of 5e) compatible firmware (all except the macros) for their EiconJet (first with a hardware blitter then without). Turned out well in the end, with my interests shifting to OS/2 and multi-threading rather than only games.

Edit: Forgot to mention that I started out incessantly wanting the Timex Sinclair 1000 kit. My father flatly refused and eventually I got the Atari 8-bit. Which I thought was way cooler than the C64. I learned so much programming on that machine--with no way to save my programs for the first year.

Apparently only some aspects can be reduced and the mechanism is also in Windows 10.

I learned a solution is to turn the documents into vectors in say PostgreSQL (with pgvector) and do a cosine similarity search with a search vector. Doing a search for embed models on HuggingFace shows nomic-ai/nomic-embed-text-v1.5 and Qwen/Qwen3-Embedding-0.6B. I might have used a larger one like Qwen/Qwen3-Embedding-4B.

There's some info for AnythingLLM[0] which supports RAG. AnythingLLM has LanceDB out of the box but also supports others including pgvector.

[0] https://docs.anythingllm.com/features/embedding-models

I just discovered the name of this. I visualized that you can do a quicksort but only recurse one of the partitions each time--the one that contains the median index. It has the same worst-case O(n^2) and can be fixed the same way choosing the median pivot of 3 potential pivots. Apparently C++'s `std::nth_element` uses quick-select and since if you choose a different target index can find any percentile not just the median.

That's useless as a product, fun as a project.

I don't even like carrying the smallest of folding umbrellas around in case it might rain. Then if it does, this thing will hover so far above and away that you get wet anyway optimistically saving up to 50% of raindrops. 50% wet is still wet. If I have an umbrella I hold it close over my head unless I'm covering someone walking with me.

The problem isn't AI. The problem has been the mass fan-out of information and unchecked regulation, people, or algorithms that determine it. AI only makes it worse.

The first and mostly only time I use byobu was for production operations on a mysql database cluster. I thought it was some customized script or version of screen meaning Bring-Your-Own-BackUp. I later realized that's what those folding screens are called.

The general public will not care about these debates. If they have an individual humanoid AI that they've gone through many good times and bad and know each other throughout, they will be treated as a valued immediate member. Labels don't matter, just behaviour less of AI's but of humans. Shared experiences, values, and predictable behviour count for a lot (maybe everything).

An interesting shift is in how we routinely say "AI" and people hardly ever push back, vs saying "LLM" which how sounds pedantic in many contexts.

I don't recall him having any impact on the story. The movie didn't lose anything by omitting the detour. To me it's like hedonistic gods in Buddhism--all powerful but not practically consequential to humans.

I thought it was going say Samwise.

Woah, a return to old glory form? Simply good is great.

That's a ridiculous stat to report on if the other half say they're better off--it's a wash.

Actual headline "Nearly half of Americans say they're worse off financially than a year ago, NY Fed finds."

The best thing that I'd used for auto-updated software with weekly updates was to prefix user-facing visible commits with "uv:" Then each week we search for them and either use the text as-is or massage it slightly. We even got it into the product itself in the Help/Release-notes menu.

Funny to ask to stop doing something I don't do or never even heard of. I typically only mark database schema migrations or other major things with special prefixes.

First thing that comes to my mind is the procedural generation in Rescue on Fractalus! (Behind Jaggi Lines) 1984 by LucasFilm Games which blew my mind on Atari 6502.

I DDG searched for the posted title and could read the top link == TFA w/ Firefox + uBlockOrigin.

Searching a title is an easy cheat--it's rare for a story to only be published in one paywalled place.

1980. Neo-Amish.

I've not a new 'retirement' plan to voluntarily be stuck in the '80s.

Links are broken--proper ones below. They all have a similar tone/structure, like it had a few seeds of originality but then got 'produced'. More style than substance/depth for my liking.

[0] https://shvetsm.github.io/posts/who-looks-at-the-code/

[1] https://shvetsm.github.io/posts/keep-the-kitchen-clean/

Well that led me down an unexpected rabbit hole. I happened to be looking for ways to subset scatter-gather queries in a distributed database using composite hash keys and this seemed relevant. Grid partitioning (or multi-dimensional indexing) I knew of, but then also found Morton Z-Ordering (Bit-Interleaving), Coprime Matrix Multiplication (Lattice Hashing), Locality-Sensitive Hashing (LSH) / Random Projections and its relation to perceptual hashing.

Then I realized it's on Daniel Lemire's blog which is very fitting. Thanks HN.

It may be relocated in some reincarnation.

Since Dohmke’s departure, there’s been an ongoing talent drain at GitHub. Some GitHub employees have followed Dohmke to his Entire startup[0], a new developer platform that looks like it will compete directly with GitHub. Out of the 30 employees listed at Entire, at least 11 of them used to work at GitHub.

[0] https://entire.io/company

I wish the copy button for a response included the prompt (unless it's code). So many times I'm reading a response that starts running assuming the context of the prompt and I'm left wondering what it's about for the first half.

And can anything truly novel be created by a technology that feeds on what already exists?

When I was very young, I used to make up jokes that didn't really make sense, but they seemed novel and I cracked myself up. Late in life I wondered what truly original idea(s) I've had that wasn't an extension or mashup of existing ones or I'd learned from somewhere even subconsciously. It was very hard to come up with any. So this fault like many others that can be ascribed to machines can equally apply to ourselves.

I did finally think of a few, the clearest being the idea of 2*Pi being more fundamental than Pi back when I was in grade 7 math (I can visualize the room and lesson when the thought occurred to me) back in '77 long before I learned of Tau.

The idea of artificial scarcity and property in a capitalistic society runs deep.

It's the reason why copying => theft for them.

I'm hoping we'll have many of these pelican cyclist pictures collected. Then when all the models can do it well, we'll stop posting about them, and dhen the next generations of AIs train on the data we'll have these canonical archetypes.

I like to think of Gaussian Splatting as a stage full of improv actors who are iteratively/concurrently filling details in regions down to pixels until the full picture is painted.

I noticed that too. The "The story behind our list" link says:

But all human life is here.

I think that's why I don't resonate with many of the titles. I do enjoy reading a vivid account of a person in a situation, but I prefer one that's not primarily people drama, and interacts with something large in an objective sort of way.

Books are weird to non-regular readers. I've thoroughly enjoyed almost every book I've read (except those assigned in school). Even reading books much longer than I'd ever expect to finish. I grew up on early sci-fi books from the public library, later reading Microserfs, Cryptonomicon, Diamond Age, Godel Escher Bach, Lord of the Rings. I also enjoy accounts of personalities in music, automotive, and math/science (based on my Youtube history). I can enjoy but feel cheated by stories which are just human drama with a sci-fi backdrop.

I can't say I like Windows in any way, except perhaps nostalia for NT 3.51 or XP. I can however relate to 'modern Linux' experiences. I've tried using some new-fangled ones and instead of being delighted by the new hotness, just want my old Xubuntu the way I remember (but with good wifi, sound, sleep, hidpi, multimonitor support). My latest disappointment was KDE Plasma/Wayland--very modern in ways I don't appreciate. There was something simple I couldn't adjust (forget which) and tonnes of settings for things I don't care about at all.

The OS/Desktop is not the show, it's the stage/floor.

This has pretty pictures but its verbosity and frankly made-up complexity is making me not take it as seriously as I perhaps should.

The way I'd handle rebasing stacked PRs/branches is to rebase the very last one. Then simply `git branch -f a-branch <logical same point on rebased>` for each of the others, done.

I worked on a project that had weekly releases. We had git submodules, and submodules of that, and on rare occasions a 4th repo. I would manually keep those all up to date with rebases pinning each submodule to the logical new points. It all became muscle memory. The lesson I learned is don't use submodules unless you really need to. (All the repos were our own.)

JJ may be great but a stawman isn't going to sell it to me.

Now I can tell an AI to rebase a stack and as long as there weren't any conflicts easily check the results.

Everyone has things that bug them, git isn't one of mine (today :-). Instead I have a custom keyboard layout that no one else uses that makes me feel better, but I don't go around telling everyone they should switch--unless you're curious[0].

[0] https://github.com/qwickly-org/Qwickly

DeepSeek v4 pro is still rather large, DeepSeek-V4-Flash[0] becomes relatively more reasonable with smaller quantizations and eventually will be able to effectively offload 'facts' to system RAM. See DwarfStar 4[1] for current sweet spots.

[0] https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash

[1] https://news.ycombinator.com/item?id=48142108