First screen, captured 1 day ago. Hover to scroll the page, or open the video.

GPT-5.6 Sol can beat Factorio on peaceful mode

What HN thinks

An AI finally plays Factorio like a tired human: chests everywhere, belts optional, and the factory must grow by hand.

Impressive but heavily pre-trained
Many argue the model has absorbed every wiki, guide and blueprint, and peaceful mode removes real-time pressure, so this is far from learning a game from scratch.
It skipped the core mechanic
Commenters note Sol hand-fed chests instead of belt routing, treating the player as a logistics unit; humans would find that tedious and it doesn't scale.
LLMs in games as features
A large tangent wants small LLMs enriching NPC dialogue, diplomacy and strategy opponents, with debate over whether AI-written dialogue ruins the art.

Consensus: Most agree the run is progress but that peaceful mode and the manual ferrying make it a weaker result than the headline implies; biters would likely expose real-time weaknesses.

Recommended in the comments

FactorioLab ↗ (tool): Cited as one of the planners that makes the game highly calculable.
Factorio Prints ↗ (service): Mentioned as a source of ready-made factory blueprints.
autorio ↗ (tool): A commenter was more interested in the harness itself for an LLM base gardener.
Infinite Ores ↗ (other): Suggested instead of an LLM to avoid babysitting depleted patches.

Summarised by Claude from the comment thread 1 day ago.

Points: 148
Comments: 179
Best rank: #1
Posted: 2026-10-31 by ryanmadden
Category: Ai ml

factorio gpt-5-6 openai-codex ai-agents llm-benchmarks game-playing-ai

All captures:

Full page

Captured 1 day ago. Click for the whole screenshot.

Looks similar

Screenshot of datavorous.com: I post trained a model on Demoscenes

I post trained a model on Demoscenes (datavorous.com)

39 points by datavorous_ 2 days ago | 26 comments

A blog post describing post-training a Qwen-based 7B model on 25k Dwitter JavaScript demos with curriculum learning and human-preference RL to generate demoscene continuations.

demoscene dwitter llm-fine-tuning rlhf

Screenshot of vosen.github.io: Zluda 6 release (run unmodified CUDA applications on non-Nvidia GPUs)

Zluda 6 release (run unmodified CUDA applications on non-Nvidia GPUs)(vosen.github.io)

139 points by Tiberium 4 months ago | 13 comments

Screenshot of olivergrabner.com: LLM agents play Pokemon showdown

LLM agents play Pokemon showdown (olivergrabner.com)

25 points by grabnero 11 days ago | 14 comments

A research write-up on a multi-agent environment where LLM teammates coordinate via chat in Pokémon Showdown 2v2 battles, comparing win rates across talk modes.

multi-agent-llm pokemon-showdown qwen agent-coordination

Screenshot of pydong.org: Python’s Preprocessor

Python’s Preprocessor (pydong.org)

375 points by rbanffy 2 years ago | 81 comments

Blog post explaining how Python source code encoding declarations (PEP 263) can be abused to implement a preprocessor that transforms code before execution.

python source-encodings pep-263 preprocessor