frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: WorldBuild Bench repo: testing LLM world coherence with 3D games

https://github.com/sebnado/worldbuild-bench
2•sebnado•2h ago
I built WorldBuild Bench because, as we all know, llm bench scores often say something very different from what models actually feel like to use. It's really dependent on the type of tasks.

I personally want to test spatial, temporal, and causal coherence in an interactive 3D world. Does the model understand where things are, world stays consistent over time and do the consequences make sense? There is a million people generating random games here and there on yt, but I want something that I can reproduce every time a model comes out and gets scored.

For this first run, 9 models received the same three roughly 30-line game prompt. I just added Kimi k3 to the results.

They all ran in high-thinking mode through the same open source harness, with the same sub-agent setup and access to Three.js, Rapier, and Playwright. There is currently one run per model per brief, producing 27 browser-playable games.

Because the qualities I’m interested in are difficult to score automatically, the main evaluation happens through blind pairwise comparisons. You play two games built from the same brief without seeing the model names, then compare their game feel, world design, presentation, completeness, and overall quality.

I’m also publishing the prompts, generated artifacts, generation time, estimated cost, and code size.

Fable produced some of the strongest games from what I could see, but its three runs cost about $756. GPT-5.6 Sol cost about $108, while GLM-5.2 and Grok 4.5 each cost around $19. Opus also felt closer to Fable than I expected, considering the large cost difference. Kimi somehow ended up roughly at the cost of GPT, but performed somewhat similar to Opus (thats just my subjective opinion there)

This first run is small, and as stated above, human preference is subjective. But I plan on running more and hope to evolve the methodology. As long as I can afford all these tokens. Fable is ridiculously expensive.

If you look at it, I'd really appreciate criticism of the task methodology, design, blind evaluation, etc. What would make this rigorous enough to be truly useful.

N.J. MVC error registered 6,600 undocumented immigrants to vote

https://www.nj.com/politics/2026/07/thousands-of-undocumented-immigrants-were-registered-to-vote-...
1•speckx•1m ago•0 comments

Dementia symptoms that show up years before diagnosis

https://www.washingtonpost.com/wellness/2026/07/21/these-dementia-symptoms-can-show-up-years-befo...
1•brandonb•1m ago•0 comments

FBI Director Kash Patel Planning a Trip to Russia

https://www.reuters.com/world/us/fbi-director-kash-patel-planning-trip-russia-politico-reports-20...
2•Rooster61•1m ago•0 comments

Shardhash Protocol: Sharden Issuer

https://github.com/babakkarimib/shardhash
1•babakkarimib•1m ago•0 comments

OpenAI Shares Some Alignment Problems

https://thezvi.substack.com/p/openai-shares-some-alignment-problems
1•7777777phil•1m ago•0 comments

Jobflo is getting more useful

https://apps.apple.com/us/app/jobflo-private-job-tracker/id6759933192
1•jsrothwell•2m ago•0 comments

Pete Hegseth's Reckless Message to Helicopter Pilots

https://www.theatlantic.com/international/2026/07/hegseth-reckless-military-helicopter/688008/
2•Jtsummers•3m ago•0 comments

Why AI Needs a "Genie Coefficient"

https://spectrum.ieee.org/ai-agent-benchmark
1•mikelgan•3m ago•1 comments

SWARÓG: a highly powerful agent for network monitoring and content generation

https://medium.com/@a.galecki/swar%C3%B3g-highly-powerful-agent-for-network-monitoring-and-conten...
1•NodesHub•4m ago•1 comments

Fei-Fei Li says robotics simulation startup SceniX joins World Labs

https://runtimewire.com/article/fei-fei-li-scenix-joins-world-labs
1•ryanmerket•5m ago•0 comments

Show HN: Ounce, Stop measuring wealth in inflating fiat. Measure it in gold

https://useounce.com/
1•halfradaition•5m ago•0 comments

Show HN: Observability for Coding Agents and LLM Applications

https://telemetry.dev/
1•ephraimduncan•6m ago•0 comments

Impossible Colors

https://en.wikipedia.org/wiki/Impossible_color
1•akramachamarei•7m ago•1 comments

Ask HN: My Microsoft for Startups subscription is expiring

1•alexander2002•8m ago•0 comments

Substack adds an AI detector to help spot blogs written by no one

https://www.theverge.com/ai-artificial-intelligence/968855/substack-pangram-ai-detecting-tool
1•droidjj•8m ago•0 comments

Gritt: AI to Build Infrastructure with Robotics

https://www.gritt.ai/
1•wslh•8m ago•0 comments

Show HN: Collecte MTL – niche web app to know Montreal waste collection schedule

https://collectemtl.ca/
2•scastiel•9m ago•0 comments

Bruno Blocker – I made an free and lightweight focus app

https://www.brunoblocker.com/
2•aksel_salmi•9m ago•2 comments

France first European country to ban social media access for under-15s

https://www.theguardian.com/world/2026/jul/21/france-ban-social-media-access-under-15s
2•Erikun•10m ago•0 comments

Show HN: Posecode, a text language and renderer for human movement

https://www.posecode.org/play
4•brnbrn•10m ago•0 comments

Jimothy the raccoon has a rare spinal condition. Here's what that means

https://www.popsci.com/science/whats-jimothy-raccoon-condition/
2•speckx•10m ago•0 comments

Wordle

https://en.wikipedia.org/wiki/Wordle
3•chistev•11m ago•0 comments

Three-Body Scattering for Generative Modeling

https://arxiv.org/abs/2607.18198
2•E-Reverance•12m ago•0 comments

Judge approves $1.5B Anthropic settlement, reduces class counsel fees to 6.8% [pdf]

https://storage.courtlistener.com/recap/gov.uscourts.cand.434709/gov.uscourts.cand.434709.680.0_4...
4•ilamont•13m ago•0 comments

New HollowGraph malware uses Microsoft Graph for stealthy C2 comms

https://www.bleepingcomputer.com/news/security/new-hollowgraph-malware-uses-microsoft-graph-for-s...
4•devonnull•14m ago•0 comments

ACP v2 is available in Draft

https://agentclientprotocol.com/announcements/acp-v2-draft
2•artooro•16m ago•0 comments

June 23, 2026: The Night France Saved the British Grid

https://dispatches.kilowatts.io/p/a-midsummer-nights-near-miss
5•bookofjoe•17m ago•0 comments

Synthetic Consumer Panels Without Microdata via IPF

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7122180
1•7777777phil•18m ago•0 comments

Ask HN: Personal Goal Setting Method (Jim Rohn)

1•urnicus•18m ago•0 comments

Show HN: Stage is sunestting their product, what now?

1•othmanosx•19m ago•0 comments