frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Accidentally Turing-Complete

https://beza1e1.tuxen.de/articles/accidentally_turing_complete.html
25•bschne•1y ago

Comments

panstromek•1y ago
Nice list. Some of those are arguably not accidental, TypeScript type system seems kinda obvious to be turing complete when it tries to describe dynamically typed langauage.
WalterGR•1y ago
x86 MOV instruction: “The mov-only DOOM [game] renders approximately one frame every 7 hours, so playing this version requires somewhat increased patience.”
a_cardboard_box•1y ago
Rule 110 is only Turing-Complete if you have an infinitely large array of cells, and are able to initialize it with an infinite repeating pattern. If I'm not mistaken, HTML+CSS can only do a fixed-sized array.

With a Turing-Complete language, if a program runs out of memory on one machine, you can run the same code on a bigger machine without modifying it, and it can use the additional memory. With fixed-length rule 110, you need to modify the code if you want to use more memory.

256_•1y ago
This is addressed in the second paragraph of TFA:

"Stuff which is somehow limited (stack overflows, arbitrary configuration, etc) is still considered Turing complete, since all "physical" Turing machines are resource limited."

In my opinion, worrying about infinite memory, in regards to Turing completeness, makes the task of implementing computation much less interesting.

Also, I'm pretty sure CSS only does one generation (or a finite number of them) before stopping anyway.

256_•1y ago
Logic in Doom is particularly interesting to me. Apparently you can fit ~64k logic gates in a map (using the method described). From [1]:

"As the DOOM engine was not designed to be an interpreter, there are some constraints on our programs written against it. The biggest one is how large our programs can be. Since each gate uses at least one tag, we can use this as a metric to derive an upper-bound on the size of a program. As the DOOM engine uses 16-bit tags, this means we can have, at most, 65535 gates. This is not a particularly large number. We may be able to implement a very small CPU but this limit will be hit pretty quickly I believe."

The z80 had ~8,500 transistors. The 8086 had ~29,000 (checking Wikipedia). You could get far fewer if you use a 1-bit microarchitecture, I'm sure. I think there was a DEC (PDP?) computer that used that trick to have a really low transistor count, but I don't remember what it was called.

The real problem is RAM; for this you may as well cheat and modify Doom's code to add a RAM chip, and I/O while you're at it.

You could create a CPU in Doom implementing an architecture for which a C compiler exists, capable of compiling Doom, and run it in the CPU in Doom. For "reasonable" speed you'd have to do more than one simulation step per frame render (in the host Doom). If you ran it for long enough maybe you could get a full frame of Doom in Doom.

[1]: https://calabi-yau.space/blog/doom.html

karmakaze•1y ago
Doom running in TypeScript static type checker[0].

> half trillion lines of types totaling 177 terabytes ran through the type checker around the clock for 12 days to get the first frame

[0] https://news.ycombinator.com/item?id=43184291

karmakaze•1y ago
My favorite one is Conway's Game of Life. It's perhaps the least surprising one, but it's also the most visually appealing. Really like this video that leads up to making the Game of Life in itself[0]. It's something you can show a non-technical person and they can get a sense of how crazy it is that something so simple can do anything.

[0] https://www.youtube.com/watch?v=Kk2MH9O4pXY

Show HN: Can AI read your mind?

https://jeffyclassify.com/blog/mind-reader/
1•nico•2m ago•0 comments

Everything New in OpenCode 2.0

https://www.augmentedswe.com/p/whats-new-in-opencode-2
1•wordsaboutcode•3m ago•0 comments

It Could Be Otherwise: Science in the Age of Uncertainty

https://blue-continuum.com/it-could-have-been-otherwise
1•dnetesn•4m ago•0 comments

Show HN: AI Hub – Open Access Lesson Plans for Teaching Students about AI

https://aihub.remakelearning.org/
1•gault8121•12m ago•0 comments

Show HN: The LAST DOOR – a tiny game about deciding when to leave

https://sydro.surge.sh/lastdoor/
1•sydro9•20m ago•0 comments

Borrowing Strength (Drug Dosing)

https://arenamag.com/articles/borrowing-strength
2•elo2000•21m ago•0 comments

Show HN: PDF to ePub Conversion Tool for Mac

https://github.com/bnlcas/pdf2epub
2•blululu•23m ago•0 comments

Apples to Apples: Getting Type Mappings Right in LLM-Based Code Translation

https://www.codemetal.ai/research/apples-to-apples-getting-type-mappings-right-in-llm-based-code-...
3•matt_d•23m ago•0 comments

Learning What to Investigate Next: Meta-Reasoning for Long-Horizon Agents

https://arxiv.org/abs/2610.02525
3•matt_d•27m ago•0 comments

KBEmacs, MIT's Programmer's Apprentice, Reanimated

https://github.com/jeffshrager/kbemacs
2•abrax3141•28m ago•0 comments

Ask HN: Where has creativity in software gone?

5•piratesAndSons•28m ago•3 comments

Show HN: Built a Rome City Builder

https://tiberava.pages.dev/
7•granga•29m ago•0 comments

DOJ arrests 2, seizes 187 devices in North Korean laptop farm crackdown in Texas

https://www.reuters.com/legal/government/doj-arrests-two-seizes-187-devices-north-korean-laptop-f...
3•usernomdeguerre•32m ago•0 comments

I built a tool to track what AI leaders say and how their views change on video

https://blumify.io/people
5•gcsydney•34m ago•0 comments

High-speed video reveals how honey bee stingers pump venom

https://arstechnica.com/science/2026/10/honey-bee-stingers-are-double-piston-venom-pumps/
4•gumby•38m ago•0 comments

Character.ai files for Chapter 11 bankruptcy. Disney makes a bid

https://www.cnbc.com/2026/10/13/character-ai-chapter-11-disney-bid.html
1•Gaishan•38m ago•1 comments

The Feeling of Power

https://hex.ooo/library/power.html
2•layer8•38m ago•0 comments

1 in 8B: A Life Simulation Game

https://1in8b.life/en/
4•nogajun•39m ago•0 comments

Fine-tuning decision models – Server Logs and IoT Fleets

https://hive.technology/lab-notes/fine-tune-decision-models/
2•tylermhall•40m ago•0 comments

Show HN: Using the long book of the law to improve PA health insurance appeals

https://gemini.google.com/share/ac24e893cfcf
2•Steaglsz•41m ago•0 comments

Do Frameworks Matter Anymore?

https://brookslybrand.com/posts/do-frameworks-matter-anymore/
4•fagnerbrack•42m ago•0 comments

Who Decided AI Gets to Have Feelings?

https://middlelayer.substack.com/p/who-decided-ai-gets-to-have-feelings
4•logan1085•44m ago•1 comments

A recent doubling of global warming rate signals a 1.5°C breach by 2028

https://www.nature.com/articles/s41612-026-01561-y
4•littlexsparkee•45m ago•0 comments

Topgolf saved over $11.3M after leaving VMware

https://arstechnica.com/information-technology/2026/10/topgolf-saved-over-11-3-million-after-leav...
11•canucker2016•49m ago•0 comments

Creating a Garmin "MCP"

https://jackantico.io/garmin-mcp
6•ricotico060•51m ago•0 comments

The Machines That Make the Machines

https://developer.nvidia.com/blog/the-machines-that-make-the-machines/
3•gmays•53m ago•0 comments

U.S. Treasury Announces Record Enforcement Actions Against Bank of America

https://home.treasury.gov/news/press-releases/sb0657
12•hippich•57m ago•4 comments

HN: Encrypted Privacy Aware Emails, ZyMail Hit the Globe

3•ZachQ•59m ago•0 comments

Ask HN: Could you find funding if you didn't want to build a $1-10B business?

5•FrenchDevRemote•1h ago•0 comments

AI, Hyperwar and End Times Fascism

https://efturnip.substack.com/p/ai-hyperwar-and-end-times-fascism
5•dopple•1h ago•0 comments