HN user

s-macke

2,097 karma

OpenRISC Emulator running Linux: http://s-macke.github.com/jor1k/

My Github site: https://github.com/s-macke

My personal site: www.simulationcorner.net

Posts55
Comments402
View on HN
twitter.com 1mo ago

When Washington switched off Fable/Mython 5: What happened, hour by hour

s-macke
3pts0
github.com 1mo ago

366-byte C program emulator can run Linux and Doom – IOCCC29 winner

s-macke
6pts1
www.youtube.com 1mo ago

IOCCC29 Awards Presentation and Source Code Reveal [video]

s-macke
1pts0
github.com 1mo ago

Show HN: I reverse-engineered the world maps of Test Drive III (1990 DOS game)

s-macke
215pts56
github.com 1mo ago

Abstruse Goose Comic Archive

s-macke
3pts0
resources.anthropic.com 5mo ago

2026 Agentic Coding Trends Report

s-macke
1pts0
www.understandingai.org 6mo ago

AI is just starting to change the legal profession

s-macke
4pts2
www.amd.com 6mo ago

Taking Local AI Development to the Max with AMD Ryzen AI Halo

s-macke
3pts0
github.com 6mo ago

Abstruse Goose Comic Archive

s-macke
3pts0
github.com 6mo ago

Show HN: Reverse engineering and porting a C64 game with AI

s-macke
2pts0
phys-sim-book.github.io 12mo ago

Physics-Based Simulation Free Online Book

s-macke
3pts0
github.com 1y ago

Can diffusion models solve visual Sudoku?

s-macke
1pts0
twitter.com 1y ago

OpenAI Researcher Jason Wei: It's obvious that it will not be a "fast takeoff"

s-macke
36pts16
arxiv.org 1y ago

Reinforcement Learning Finetunes Small Subnetworks in Large Language Models

s-macke
3pts0
github.com 1y ago

The simplest, fastest repository for training/finetuning small-sized VLMs

s-macke
2pts0
www.twitch.tv 1y ago

Gemini 2.5 Pro won Pokémon Blue in 106k moves

s-macke
2pts3
arxiv.org 1y ago

SWE-Smith: Scaling Data for Software Engineering Agents

s-macke
1pts0
twitter.com 1y ago

Sam Altman: we added one million users in the last hour

s-macke
4pts0
arxiv.org 1y ago

Measuring AI Ability to Complete Long Tasks

s-macke
1pts0
www.lesswrong.com 1y ago

FrontierMath Was Funded by OpenAI

s-macke
22pts4
arxiv.org 1y ago

Grokking at the Edge of Numerical Stability

s-macke
1pts0
anokas.substack.com 1y ago

LLMs struggle with perception, not reasoning, in ARC-AGI

s-macke
5pts0
huggingface.co 1y ago

A Llama 70B finetune that has reflection baked into it's weights

s-macke
6pts0
openai.com 2y ago

OpenAI releases ChatGPT on your desktop for macOS

s-macke
27pts20
arxiv.org 2y ago

Examination of Large Language Model Performance on Grade School Arithmetic

s-macke
2pts0
github.com 2y ago

Llama3 is the first open model that can also win text adventures

s-macke
30pts16
blog.qaware.de 2y ago

Java Microservices and the Battle for Stability Under High Load

s-macke
2pts0
arxiv.org 2y ago

MambaByte: Token-Free Selective State Space Model

s-macke
4pts0
blog.qaware.de 2y ago

Java Microservices and the Battle for Stability Under High Load

s-macke
2pts0
github.com 2y ago

Show HN: Benchmarking language models by playing text adventures

s-macke
2pts1

Yes, you haven’t tried it. LLMs are actually awesome at deobfuscation, but terrible at obfuscation. They just can’t do it yet.

They also lack the creativity needed for those entries. Obfuscation is only one part of it. Coming up with the idea is another. Many entries also have special qualities that make them true works of art.

Author here. The Ross’s Game Dungeon video was, to some degree, a motivation for me to start the reverse-engineering process, just for fun. The analysis took place over many years, and I figured out 95% of the format myself. AI helped me then with the coloring, which is pretty complex. I have read the Accursed Farms forum, of course. But it didn't help much.

The map mentioned there can be found as “Unknown Map” in the dropdown. However, I haven’t yet connected that map to the menu screen. It might be called “Tom’s Test Track,” because that string appears in the disassembly.

Integration into noclip might be easy, because I also have the geometry as a standalone .obj file.

Voxel Space (2017) 2 months ago

Author here. Yes, it is integral. I chose this approach to first show how to draw it from back to front, because the code is easier to understand this way.

Nice to see an MCP integration here as well. In my experience, coding agents are great at analyzing MOS6502 code. Because the code is limited to only 64 kB, it does not overwhelm the agent. And in parallel it can write specs and even extract assets via normal coding tools.

Using my similar tool [0], I feel I get roughly a 100x speedup. I will definitely try regenerator2000.

[0] https://github.com/s-macke/OpcodeOracle

Most such emulators have Internet access on the IP level. Therefore, this is a very cheap way to test anything on the Internet.

    apk add nmap
    nmap your.domain.com
However, the speed is heavily throttled. You can even use ssh and login to your own server.

It can also be used as a very cheap way to provide a complete build environment on a single website, for example to teach C/C++. Or to learn the shell. You don't have to install anything.

AI is very effective for reverse engineering. Unless you’re doing it purely for fun, it makes sense to use AI where it helps.

I’ve tried to visualize the “navigate and modify” process you mentioned in [0]. It’s mesmerizing.

Because reverse engineering outcomes are comparatively easy to verify, it’s a good fit for training for AI. I expect major progress in the next few years, potentially to the point where reverse engineering many binaries becomes highly automated.

[0] https://github.com/s-macke/OpcodeOracle

About halfway through my reverse-engineering process, I came across those documents and realized that roughly 50% were missing. I searched pretty much the entire web to see if anyone had stored these files elsewhere, but so far I haven’t had any luck.

Author here. I’m happy to see one of my projects on Hacker News. This has been a fun one. One evening you just try to disassemble it and wonder where the code is. The following months were a truly satisfying experience, reverse-engineering this diamond.

There is still a functioning Forth interpreter implemented in the game. If they hadn’t removed all the word names, it would have been possible to debug at any time and analyze the program state and call functions live with real Forth code. Some crazy feature at that time.

There’s a small misunderstanding here. Reverse engineering by annotating a disassembly file is fine. However, the next obvious steps would be to write a sensible, high-level documentation of the internals of the game and then port it to a high-level language. Indeed, whether this would run on the original hardware might be questionable for such an Atari Game. But have you seen modern C compilers like Oscar64 [0] for such old hardware? Never say never …

[0] https://github.com/drmortalwombat/oscar64

I’ve performed many experiments using AI to reverse-engineer old games like this one, and it looks like the newest generation of models has no trouble with it. They’re actually awesome.

Even with the current models, we might be able to automatically reverse-engineer all those old games, decode all assets, and even rewrite them in a more sensible language than assembler.

I’ve also been playing around with reverse engineering, and I’m very impressed. It turns out that Codex with GPT-5.2 is better at reverse engineering than Claude.

For example, Codex can completely reverse-engineer this 1,300-line example [0] of a so-called C64-SID file within 30 minutes, without any human interaction.

I am working on a multi-agent system that can completely reverse-engineer C64 games. Old MS-DOS games are still too massive to analyze for my budget limit.

[0] https://gist.github.com/s-macke/595982d46d6699b69e1f0e051e7b...

... emitting a NYC worth of CO2 in a year is dizzying

Simplified comparisons like these rarely show the full picture [0]. They focus on electricity use only, not on heating, transport, or meat production, and certainly not on the CO2 emissions associated with New York’s airports. As a rough, back-of-the-envelope estimate, a flight from Los Angeles to New York with one seat is on the order of 1,000,000 small chat queries CO2e.

Of course we should care about AI’s electricity consumption, especially when we run 100 agents in parallel simply because we can. But it’s important to keep it in perspective.

[0] https://andymasley.substack.com/p/a-cheat-sheet-for-conversa...