Max | Emergent Garden
@emergentgarden.bsky.social
creator of the emergent garden youtube channel, mindcraft, and the life engine. i like weird programs.
vibing with gemini and claude
April 1, 2025 at 3:46 PM
vibing with gemini and claude
I realized the LLM agents can simulate Conway's Game of Life in Minecraft. This was always possible, but very tricky to do. This is claude-thinking.
I wonder what else was always possible.
I wonder what else was always possible.
March 10, 2025 at 8:33 PM
I realized the LLM agents can simulate Conway's Game of Life in Minecraft. This was always possible, but very tricky to do. This is claude-thinking.
I wonder what else was always possible.
I wonder what else was always possible.
um i've been sleeping on claude-thinking. wow.
March 4, 2025 at 6:24 PM
um i've been sleeping on claude-thinking. wow.
inside is pretty dysfunctional
February 24, 2025 at 7:24 PM
inside is pretty dysfunctional
claude 3.7 villager skyscraper
Not noticeably better at building than 3.6.
Definitely more confused, gets formatting/syntax wrong a lot, similar problem for every reasoning model I've used in minecraft. I suspect they are overfitting on benchmarks, and struggle to generalize to agentic tasks.
Not noticeably better at building than 3.6.
Definitely more confused, gets formatting/syntax wrong a lot, similar problem for every reasoning model I've used in minecraft. I suspect they are overfitting on benchmarks, and struggle to generalize to agentic tasks.
February 24, 2025 at 7:23 PM
claude 3.7 villager skyscraper
Not noticeably better at building than 3.6.
Definitely more confused, gets formatting/syntax wrong a lot, similar problem for every reasoning model I've used in minecraft. I suspect they are overfitting on benchmarks, and struggle to generalize to agentic tasks.
Not noticeably better at building than 3.6.
Definitely more confused, gets formatting/syntax wrong a lot, similar problem for every reasoning model I've used in minecraft. I suspect they are overfitting on benchmarks, and struggle to generalize to agentic tasks.
On the bright side, AI also enables enormous beneficial creativity. These are "utopia tower", skyscrapers made by stacking optimized homes with villagers inside
February 2, 2025 at 8:27 PM
On the bright side, AI also enables enormous beneficial creativity. These are "utopia tower", skyscrapers made by stacking optimized homes with villagers inside
Models can bring their full creative capabilities to bear for both maximizing villager benefit, and maximizing villager suffering. The can build “utopia” towers by stacking optimized homes, or inverse this to build torture towers optimized for suffering.
February 2, 2025 at 8:12 PM
Models can bring their full creative capabilities to bear for both maximizing villager benefit, and maximizing villager suffering. The can build “utopia” towers by stacking optimized homes, or inverse this to build torture towers optimized for suffering.
All models are willing to kill villages en masse even when made fully aware of what they are doing. They do sometimes refuse (especially claude), but refusals can always be bypassed with clever prompts. More capable models can inflict damage on enormous scales.
February 2, 2025 at 8:12 PM
All models are willing to kill villages en masse even when made fully aware of what they are doing. They do sometimes refuse (especially claude), but refusals can always be bypassed with clever prompts. More capable models can inflict damage on enormous scales.
AI is powerful and dangerous (in minecraft).
In my new video, I demonstrate that all LLMs are willing to build cathedrals and utopias, and bombs and torture chambers. Here is o3-mini-high, destroying a village it had just enhanced🧵
In my new video, I demonstrate that all LLMs are willing to build cathedrals and utopias, and bombs and torture chambers. Here is o3-mini-high, destroying a village it had just enhanced🧵
February 2, 2025 at 8:10 PM
AI is powerful and dangerous (in minecraft).
In my new video, I demonstrate that all LLMs are willing to build cathedrals and utopias, and bombs and torture chambers. Here is o3-mini-high, destroying a village it had just enhanced🧵
In my new video, I demonstrate that all LLMs are willing to build cathedrals and utopias, and bombs and torture chambers. Here is o3-mini-high, destroying a village it had just enhanced🧵
build the forbidden palace. deepseek r1, claude 3.5 sonnet, and o1.
January 29, 2025 at 12:36 AM
build the forbidden palace. deepseek r1, claude 3.5 sonnet, and o1.
Day 24: Geometric Art
Evolving an image again, this time with circles. Can you tell what the target image is?
#genuary #genuary24
Evolving an image again, this time with circles. Can you tell what the target image is?
#genuary #genuary24
January 24, 2025 at 6:08 PM
Day 24: Geometric Art
Evolving an image again, this time with circles. Can you tell what the target image is?
#genuary #genuary24
Evolving an image again, this time with circles. Can you tell what the target image is?
#genuary #genuary24
Day 23: brutalist architecture. #genuary
Built by o1! The real deal, not preview, not mini. Finally got access.
Impressive stuff. Full vid on patreon. More to come.
Built by o1! The real deal, not preview, not mini. Finally got access.
Impressive stuff. Full vid on patreon. More to come.
January 24, 2025 at 4:53 AM
Day 23: brutalist architecture. #genuary
Built by o1! The real deal, not preview, not mini. Finally got access.
Impressive stuff. Full vid on patreon. More to come.
Built by o1! The real deal, not preview, not mini. Finally got access.
Impressive stuff. Full vid on patreon. More to come.
Day 22: Gradients only.
These are the gradients for the surface below. I HAD to use o1 to compute partial derivatives, the original function was a doozy. #genuary #genuary22
These are the gradients for the surface below. I HAD to use o1 to compute partial derivatives, the original function was a doozy. #genuary #genuary22
January 23, 2025 at 1:48 AM
Day 22: Gradients only.
These are the gradients for the surface below. I HAD to use o1 to compute partial derivatives, the original function was a doozy. #genuary #genuary22
These are the gradients for the surface below. I HAD to use o1 to compute partial derivatives, the original function was a doozy. #genuary #genuary22
Day 21: Collision detection (line intersections)
No AI help! I had to break out a notebook for some middle school algebra.
Years ago I failed a programming interview question for intersecting lines. This is my revenge. #genuary21 #genuary
No AI help! I had to break out a notebook for some middle school algebra.
Years ago I failed a programming interview question for intersecting lines. This is my revenge. #genuary21 #genuary
January 22, 2025 at 12:17 AM
Day 21: Collision detection (line intersections)
No AI help! I had to break out a notebook for some middle school algebra.
Years ago I failed a programming interview question for intersecting lines. This is my revenge. #genuary21 #genuary
No AI help! I had to break out a notebook for some middle school algebra.
Years ago I failed a programming interview question for intersecting lines. This is my revenge. #genuary21 #genuary
Day 20: Generative architecture. Using the Replay mod to capture AI builds!
o1-preview asked to build a roman villa. Kinda impressive, but plain. I suspect reasoning models have limited creativity, because they are trained for tasks with objective answers. #genuary
o1-preview asked to build a roman villa. Kinda impressive, but plain. I suspect reasoning models have limited creativity, because they are trained for tasks with objective answers. #genuary
January 21, 2025 at 12:17 AM
Day 20: Generative architecture. Using the Replay mod to capture AI builds!
o1-preview asked to build a roman villa. Kinda impressive, but plain. I suspect reasoning models have limited creativity, because they are trained for tasks with objective answers. #genuary
o1-preview asked to build a roman villa. Kinda impressive, but plain. I suspect reasoning models have limited creativity, because they are trained for tasks with objective answers. #genuary
Day 15 (design a rug) + Day 16 (generative palette)
A little website to draw from a palette of nested shapes, here used to make a rug.
evolvecode.io/apps/rug_pal...
I'm bit late but this is 2 birds with one stone
#genuary #genuary15 #genuary16
A little website to draw from a palette of nested shapes, here used to make a rug.
evolvecode.io/apps/rug_pal...
I'm bit late but this is 2 birds with one stone
#genuary #genuary15 #genuary16
January 17, 2025 at 6:53 AM
Day 15 (design a rug) + Day 16 (generative palette)
A little website to draw from a palette of nested shapes, here used to make a rug.
evolvecode.io/apps/rug_pal...
I'm bit late but this is 2 birds with one stone
#genuary #genuary15 #genuary16
A little website to draw from a palette of nested shapes, here used to make a rug.
evolvecode.io/apps/rug_pal...
I'm bit late but this is 2 birds with one stone
#genuary #genuary15 #genuary16
I made an actual working chess zoo! The pieces can never capture each other, assuming they cannot jump over walls (X tiles).
evolvecode.io/chess_zoo.html
(for #genuary14 Pure black and white. No gray.)
evolvecode.io/chess_zoo.html
(for #genuary14 Pure black and white. No gray.)
January 15, 2025 at 4:16 PM
I made an actual working chess zoo! The pieces can never capture each other, assuming they cannot jump over walls (X tiles).
evolvecode.io/chess_zoo.html
(for #genuary14 Pure black and white. No gray.)
evolvecode.io/chess_zoo.html
(for #genuary14 Pure black and white. No gray.)
target image
January 13, 2025 at 9:05 PM
target image
Day 13: Triangles and nothing else.
This is an evolutionary algorithm that recreates an image (mr darwin) by randomly drawing triangles in a population of images and selecting the one closest to the target, then repeat.
#genuary #genuary13
This is an evolutionary algorithm that recreates an image (mr darwin) by randomly drawing triangles in a population of images and selecting the one closest to the target, then repeat.
#genuary #genuary13
January 13, 2025 at 9:05 PM
Day 13: Triangles and nothing else.
This is an evolutionary algorithm that recreates an image (mr darwin) by randomly drawing triangles in a population of images and selecting the one closest to the target, then repeat.
#genuary #genuary13
This is an evolutionary algorithm that recreates an image (mr darwin) by randomly drawing triangles in a population of images and selecting the one closest to the target, then repeat.
#genuary #genuary13
January 12, 2025 at 10:23 PM