跳到正文
Hacker News (official RSS)(youtube.com)·· 2 天前精選AI 評分70

RuneScape Bench:AI 代理的多人線上遊戲實驗

RuneScape Bench: MMOs for Agents [video]

AI 導讀

本影片展示了 RuneScape Bench,一個將開源 RuneScape 伺服器與 AI 代理結合的實驗平臺。

  • RS‑SDK:用於構建代理的工具包,支援多代理互動與評估。
推薦理由

本影片示範如何將開源 RuneScape 伺服器與 AI 代理結合,展示多代理互動與經濟模擬,並可作為多代理系統測試平臺,為研究者與開發者提供實作範例。

正文 · 原文

RuneScape Bench: MMOs for Agents | Max Bittker - YouTube

Back [Image 1](https://www.youtube.com/ "YouTube Home") TW

Skip navigation

Search

Search with your voice

[](https://www.youtube.com/watch?v=cNm4zyUtWj0)

Sign in

[Image 2](https://www.youtube.com/ "YouTube Home") TW

[](https://www.youtube.com/watch?v=cNm4zyUtWj0)

[](https://www.youtube.com/watch?v=cNm4zyUtWj0)

[Video 2](blob:https://www.youtube.com/695e2e85-b344-4516-9531-5dee0e2ca079)

[](https://www.youtube.com/watch?v=cNm4zyUtWj0)

[](https://www.youtube.com/watch?v=cNm4zyUtWj0)

RuneScape Bench: MMOs for Agents | Max Bittker

[](https://www.youtube.com/watch?v=cNm4zyUtWj0)

Tap to unmute

2x

Image 3

RuneScape Bench: MMOs for Agents | Max Bittker

Recurse Center 95,799 views 2d ago

[](https://www.youtube.com/watch?v=cNm4zyUtWj0)

Copy link

Info

Shopping

Image 4

Image 5

If playback doesn't begin shortly, try restarting your device.

•

You're signed out

Videos you watch may be added to the TV's watch history and influence TV recommendations. To avoid this, cancel and sign in to YouTube on your computer.

Cancel Confirm

Up next

Live Upcoming

CancelPlay Now

Recurse Center

Subscribe

Subscribed

Technical talks from the retreat for curious programmers. 📍NYC Do a programming retreat alongside like-minded cohort for 6 or 12 weeks. Work at the edge of your abilities, develop your volitional muscles, and learn generously.

Mote: An Interactive Ecosystem Simulation — Peter Whidden 54:37

Hide

[](https://www.youtube.com/watch?v=cNm4zyUtWj0)

Share

[](https://www.youtube.com/watch?v=cNm4zyUtWj0 "Share link")- [x] Include playlist

An error occurred while retrieving sharing information. Please try again later.

Image 6

0:00

[](https://www.youtube.com/watch?v=cNm4zyUtWj0)[](https://www.youtube.com/watch?v=RUbP2sUmtb8)

0:02 / 46:54

Live

•Watch full video

•

A RuneScape server controlled by AI agents and bots

•

46:38 I Found a Portal to RS3 (ScapeRune Ep. 1 - RS3 Guideless)Faux 268K • 12d ago Live Playlist ()Mix (50+)12:41 Jagex Has Been Busy Behemeth 109K • 14h ago Live Playlist ()Mix (50+)51:45 My Journey With Path of Exile: NZGDC Closing Keynote Chris Wilson 210K • 1d ago Live Playlist ()Mix (50+)14:57 Everything You Need to Know About Raids 4! (Fractured Archive)FlippingOldschool 118K • 4d ago Live Playlist ()Mix (50+)14:54 I Used AI To Surveil Myself At Work. What It Can See Will Disturb You.More Perfect Union 43K • 2h ago Live Playlist ()Mix (50+)11:34 Reverse Engineering the CRAZY Fridge Hack (what happened?)Low Level 36K • 3h ago Live Playlist ()Mix (50+)37:47 Old School Runescape's Broken Promise TheoryWise and Josh Isn't Gaming 59K • 4d ago Live Playlist ()Mix (50+)17:17 Can An AI Agent Play RuneScape?SirPugger 285K • 7mo ago Live Playlist ()Mix (50+)15:12 The Fractured Archive Will Have a Meta... Here's How It Plays Out!Travag 20K • 2d ago Live Playlist ()Mix (50+)1:26:27 Jagex Deleted My XP So I Found A Better Method Rendi 333K • 2w ago Live Playlist ()Mix (50+)16:19 THE DEFINITIVE GUIDE: How to get into NYU Tisch Drama Michael Quijano 4K • 4y ago Live Playlist ()Mix (50+)12:46 30 Plugins That Will Change How You Play RuneScape 3 Silenced 8.3K • 4h ago Live Playlist ()Mix (50+)

RuneScape Bench: MMOs for Agents | Max Bittker

Image 7: Recurse Center

Recurse Center

Recurse Center

10.4K subscribers

Subscribe

Subscribed

2K

Share

Download

Download

Save

95K views 2d agoRECURSE CENTER

95,799 views • Oct 17, 2026 • RECURSE CENTER

RS-sdk is a 2026 project by Max Bittker that turns LostCity into a testbed for new AI experiments. LostCity is an open source implementation of the RuneScape. This project taught him about how to design libraries for coding agents, how to build and think about evals, and the world of LLM post-training.There were also lessons in game design, simulated economics, and has even been a test…more

more

How this was made

Auto-dubbed

Audio tracks for some languages were automatically generated. Learn more

Chapters

View all

Image 8 ### A RuneScape server controlled by AI agents and bots ### A RuneScape server controlled by AI agents and bots 0:00

A RuneScape server controlled by AI agents and bots

0:00

Image 9 ### The world of RuneScape ### The world of RuneScape 1:25

The world of RuneScape

1:25

Image 10 ### AutoHotkey, botting, and automation ### AutoHotkey, botting, and automation 5:30

AutoHotkey, botting, and automation

5:30

Image 11 ### Why AI is good at coding ### Why AI is good at coding 8:14

Why AI is good at coding

8:14

Image 12 ### RS-SDK: Building agent harnesses ### RS-SDK: Building agent harnesses 12:03

RS-SDK: Building agent harnesses

12:03

Image 13 ### Agent "farming" practices ### Agent "farming" practices 16:32

Agent "farming" practices

16:32

Image 14 ### AI Benchmarking ### AI Benchmarking 22:08

AI Benchmarking

22:08

Image 15 ### RuneBench: 10 months of progress since Opus 4.5 ### RuneBench: 10 months of progress since Opus 4.5 28:21

RuneBench: 10 months of progress since Opus 4.5

28:21

Transcript

Follow along using the transcript.

Show transcript

Image 16 ### Recurse Center 10.4K subscribers

VideosAbout

Image 17: recurse.com/apply recurse.com/apply Apply to the Recurse Center

Video details

Date

Oct 17, 2026

Views

95,799

Likes

2,038

Show less

[](https://www.youtube.com/results?search_query=Recurse+Center&sp=EiG4AQHCARtDaElKeTdHcDFJdFp3b2tSMks0X2Q2S3A3ejg%253D)RECURSE CENTER
# RuneScape Bench: MMOs for Agents | Max Bittker

95,799 views 95K views

Oct 17, 2026

2K

Share

Download

Download

Save

147 Comments

Sort comments

Sort by

Top Show featured commentsNewest Show recent comments, including potential spam

Image 18: Default profile photo

Add a comment...

Image 19

@gappogappo

1 day ago

This is one of the most insightful talks on AI that I've seen. Runescape is a really interesting setting that brings out a lot of not-very-well understood behaviors of LLMs. The chains of thought were hilarious. Nice work Max

Show less Read more

Like

139

Dislike

Image 20

❤ by @RecurseCenter

Reply

Image 21

@Toomykins

1 day ago

love that he used 2004scape for this

Show less Read more

Like

116

Dislike

Reply

8 replies

Hide replies

8 replies

Image 22

@TheT8or

1 day ago

This presentation is worth more than its weight in gold. Thank you for it!

Show less Read more

Like

41

Dislike

Reply

Image 23

@jackgolding4235

1 day ago

Learning so much about SWE and AI playing on this server - thanks again for all your efforts Max! I look forward to giving some similar talks later this year!

Show less Read more

Like

26

Dislike

Reply

Image 24

@Prof_NH

2 days ago

This was so fun and interesting

Show less Read more

Like

48

Dislike

Reply

5 replies

Hide replies

5 replies

Image 25

@wadahadlan

22 hours ago

This is probably the most predictive video on how the AI space will go that we'll get Image 26: 😂

Show less Read more

Like

19

Dislike

Reply

Image 27

@TehEccentricKid

2 hours ago

I was trying to see what ai models could do using roblox studio with a lot of documentation. The studio was having issues with its MCP server locally and the agent could not connect directly to the studio. So the agent actually created its own connector that it could directly interact with the studio with, even BETTER than the mcp connector. It did try to use visuals at first but I noticed the same thing, LLMs do poorly on visual responses, they are delayed and inaccurate. You need to sync the agent up to the code directly somehow, whether that be through incoming packets, exposed api etc. That is by FAR the best way. For ai processing try to IGNORE visuals, they dont need it

Show less Read more

Like

Dislike

Reply

Image 28

@lasher2969

15 minutes ago

14:20 were you using Opus or Haiku for the runescape task? I see Haiku here but I thought VOD said Opus

Show less Read more

Like

Dislike

Reply

Image 29

@Reglaized

13 hours ago

He should’ve added another bot to thank Quinn for the successful trade

Show less Read more

Like

13

Dislike

Reply

Image 30

@spikehammer3112

3 hours ago

I hate that the Cost vs Performance chart @ 30:00 uses an inverted x-axis. It seems like it was done just for "up and to the right - good" but it makes the rest of the reasoning about the chart so much more confusing and misleading.

Show less Read more

Like

Dislike

Reply

Image 31

@mikeyjohnson5888

1 day ago

Thank you for giving this talk Max! How it aligns across many axis of mutual interests is fascinating. I wish you good luck in the future with your experiments. I don't know if you identify as a researcher or scientist but I do believe these kind of efforts are a shining example of citizen scientific research. I love when a new wild west opens up

Show less Read more

Like

17

Dislike

Image 32

❤ by @RecurseCenter

Reply

Image 33

@rorykeefe8954

2 hours ago

For a second there, I thought the text at 3:04 said “Swag guild” and I had only seen the right-most character (bc am partially blind). Thought that was a meme pulled straight from like 2005!!!

Show less Read more

Like

Dislike

Reply

Image 34

@stephybloops

1 day ago

mythical feed pull

Show less Read more

Like

10

Dislike

Reply

2 replies

Hide replies

2 replies

Image 35

@Balance-8

14 hours ago

How are they maxing in 3hours?

Show less Read more

Like

2

Dislike

Reply

2 replies

Hide replies

2 replies

Image 36

@Youtubeusermann

1 hour ago

Have you considered trying JEV?

Show less Read more

Like

Dislike

Reply

Image 37

@pebz117

1 day ago

i love the idea, the tech, the community of bots interacting and how different the economy is compared to the actual game. while it is an interesting project, all its end result with be cheaper bots to ruin the game with.

Show less Read more

Like

8

Dislike

Reply

1 reply

Hide replies

1 reply

Image 38

@andrewbuzz7308

20 hours ago (edited)

I wanted to create an mmo where players are agents, and humans are "NPCs" or spectators. But then realized that compute generally could probably go to more useful causes than colossal LLM simulations. Though, since inference is becoming more and more accessible, it might be back on the table.

Show less Read more

Like

5

Dislike

Reply

Image 39

@R3BootYourMind

5 hours ago

I wonder how the agents would behave if there was a risk of getting banned if you scam and other agents report you. I mean that the world should have an admin agent in it.

Show less Read more

Like

4

Dislike

Reply

Image 40

@Famzilla

1 day ago

I stumbled upon a random post on hacker news with this guys git repo. Absolutely awesome work! Inspired me to dive in deeper with LLMs.

Show less Read more

Like

7

Dislike

Reply

Image 41

@wrenchmonkey32

23 hours ago

Stumbled across this as I'm playing OSRS. Great video so far.

Show less Read more

Like

7

Dislike

Reply

Comments 147

Top Show featured commentsNewest Show recent comments, including potential spam

In this video

Chapters

Transcript

Chapters

Image 42 ### A RuneScape server controlled by AI agents and bots ### A RuneScape server controlled by AI agents and bots 0:00

A RuneScape server controlled by AI agents and bots

0:00

Image 43 ### The world of RuneScape ### The world of RuneScape 1:25

The world of RuneScape

1:25

Image 44 ### AutoHotkey, botting, and automation ### AutoHotkey, botting, and automation 5:30

AutoHotkey, botting, and automation

5:30

Image 45 ### Why AI is good at coding ### Why AI is good at coding 8:14

Why AI is good at coding

8:14

Image 46 ### RS-SDK: Building agent harnesses ### RS-SDK: Building agent harnesses 12:03

RS-SDK: Building agent harnesses

12:03

Image 47 ### Agent "farming" practices ### Agent "farming" practices 16:32

Agent "farming" practices

16:32

Image 48 ### AI Benchmarking ### AI Benchmarking 22:08

AI Benchmarking

22:08

Image 49 ### RuneBench: 10 months of progress since Opus 4.5 ### RuneBench: 10 months of progress since Opus 4.5 28:21

RuneBench: 10 months of progress since Opus 4.5

28:21

Image 50 ### Accidentally multi-agent present? ### Accidentally multi-agent present? 31:29

Accidentally multi-agent present?

31:29

Image 51 ### Breaking containment ### Breaking containment 33:38

Breaking containment

33:38

Image 52 ### AI macro economics -- what's valuable? ### AI macro economics -- what's valuable? 35:34

AI macro economics -- what's valuable?

35:34

Image 53 ### Agent scammers ### Agent scammers 38:36

Agent scammers

38:36

Image 54 ### How to try it out ### How to try it out 45:25

How to try it out

45:25

Sync to video time

RuneScape Bench: MMOs for Agents | Max Bittker

Image 55: Recurse Center

Recurse Center

Recurse Center

2K Likes

95K Views

Oct 17 2026

RS-sdk is a 2026 project by Max Bittker that turns LostCity into a testbed for new AI experiments. LostCity is an open source implementation of the RuneScape. This project taught him about how to design libraries for coding agents, how to build and think about evals, and the world of LLM post-training.There were also lessons in game design, simulated economics, and has even been a testbed for exploring the future of multi-agent systems, when agents are tasked with trading and trusting one another in a shared world. Max Bittker is a game designer and a founding member of Websim — he's built games such as Sandspiel and Orb.farm, and teaches a course on building and designing creative tools at NYU ITP. Max researches multi-agent systems and is the creator of rs-sdk and runebench. Max is a 2016 Recurse Center alum. ---- Read more:
* https://github.com/MaxBittker/rs-sdk/
* https://maxbittker.github.io/runebench/

Join the Discord Server:
* Image 56/discord

---- Localhost is a series of technical talks in NYC given by members of the Recurse Center community. For every event, we set aside seats for the public. The Recurse Center is the retreat for curious programmers. Spend 6 or 12 weeks in Brooklyn or online working alongside a cohort of curious and kind people. Work at the edge of your abilities on the projects that excite you. The (soft) deadline to apply to the upcoming Winter 1 Batch (beginning on Nov 2) is through Sunday October 18. New batches begin every 6 weeks. Apply here: https://www.recurse.com/apply Learn more: https://www.recurse.com/ --- 0:00 A RuneScape server controlled by AI agents and bots 1:25 The world of RuneScape 5:30 AutoHotkey, botting, and automation 8:14 Why AI is good at coding 12:03 RS-SDK: Building agent harnesses 16:32 Agent "farming" practices 22:08 AI Benchmarking 28:21 RuneBench: 10 months of progress since Opus 4.5 31:29 Accidentally multi-agent present? 33:38 Breaking containment 35:34 AI macro economics -- what's valuable? 38:36 Agent scammers 45:25 How to try it out --- Thumbnail designed by Eva Khoury.…...more

...more Show less

How this was made

Auto-dubbed

Audio tracks for some languages were automatically generated. Learn more

Chapters

View all

Image 57 ### A RuneScape server controlled by AI agents and bots ### A RuneScape server controlled by AI agents and bots 0:00

A RuneScape server controlled by AI agents and bots

0:00

Image 58 ### The world of RuneScape ### The world of RuneScape 1:25

The world of RuneScape

1:25

Image 59 ### AutoHotkey, botting, and automation ### AutoHotkey, botting, and automation 5:30

AutoHotkey, botting, and automation

5:30

Image 60 ### Why AI is good at coding ### Why AI is good at coding 8:14

Why AI is good at coding

8:14

Transcript

Follow along using the transcript.

Show transcript

Image 61 ### Recurse Center 10.4K subscribers

VideosAbout

Image 62: recurse.com/apply recurse.com/apply Apply to the Recurse Center

Video details

Date

Oct 17, 2026

Views

95,799

Likes

2,038

Transcript

NaN / NaN

Image 63 46:38

Image 64

I Found a Portal to RS3 (ScapeRune Ep. 1 - RS3 Guideless)

Faux

268K 12d ago

Image 65 12:41 New

Image 66

Jagex Has Been Busy

Behemeth

109K 14h ago

Image 67 51:45 New

Image 68

My Journey With Path of Exile: NZGDC Closing Keynote

Chris Wilson

210K 1d ago

Image 69 14:57 New

Image 70

Everything You Need to Know About Raids 4! (Fractured Archive)

FlippingOldschool

118K 4d ago

Image 71 14:54 New

Image 72

I Used AI To Surveil Myself At Work. What It Can See Will Disturb You.

More Perfect Union

43K 2h ago

Fundraiser

Image 73 11:34 New

Image 74

Reverse Engineering the CRAZY Fridge Hack (what happened?)

Low Level

36K 3h ago

Image 75 37:47 New

Image 76

Image 77

Old School Runescape's Broken Promise

TheoryWise and Josh Isn't Gaming

59K 4d ago

Image 78 17:17

Image 79

Can An AI Agent Play RuneScape?

SirPugger

285K 7mo ago

Image 80 15:12 New

Image 81

The Fractured Archive Will Have a Meta... Here's How It Plays Out!

Travag

20K 2d ago

Image 82 1:26:27

Image 83

Jagex Deleted My XP So I Found A Better Method

Rendi

333K 2w ago

Image 84 16:19

Image 85

THE DEFINITIVE GUIDE: How to get into NYU Tisch Drama

Michael Quijano

4K 4y ago

Image 86 12:46 New

Image 87

30 Plugins That Will Change How You Play RuneScape 3

Silenced

8.3K 4h ago

Image 88 16:21 New

Image 89

I Tested The Viral Reverse Engineering Repo

Low Level

2.1M 3d ago

Image 90 22:34

Image 91

How Jagex Accidentally Killed Social RuneScape

rswillmissit

71K 7mo ago

Image 92 35:51 New

Image 93

HW News - NVIDIA's $7000 RTX Spark Laptop, RIP RTX 5090, GTA VI PC Launch Rumor, CPU Price Hike

Gamers Nexus

488K 1d ago

Image 94 21:00 New

Image 95

松山機場變AI園區?連前總統都想拆掉它!為何廢松機吵30年還廢不了?

蔡黑皮

151K 1d ago

Image 96 16:51

Image 97

RuneScape Just Declared War on WoW

Bellular Warcraft

158K 2mo ago

Image 98 2:15:16

Image 99

I Spent 1000 Hours Playing RuneScape 3... As a New Player

TheoryWise

309K 1mo ago

Image 100 13:52 New

Image 101

The RuneScape Problem

Idyl Slop

54K 5d ago

Image 102 1:15:20

Image 103

The Wiki That Saved RuneScape | BeheCast #21

The BeheCast

55K 3w ago

Show more

Show more

[](https://www.youtube.com/watch?v=cNm4zyUtWj0)