Superbloom Labs

We built the first AI to teach itself to beat Minecraft. Follow for research and updates.

[Superbloom Labs @superbloom_labs]

Meet Steve: the first self-improving AI to master Minecraft without human help.

It's a step to whole civilizations growing in virtual worlds, and a leap towards game characters that come to life.

Steve out-progresses 97% of players; 26x more than attempts by NVIDIA, DeepMind and OpenAI.

Characters can now develop new abilities and evolve through experience. Soon, in-game towns, civilizations, and whole ecosystems will increasingly look and feel like our own, or like a great story, and cheaply:

Once Steve’s capabilities exist, a System One model like Jev can take over, and it’s 92% cheaper. Soon, hundreds will be able to run locally alongside any game.

Steve completes ~96% of the game’s advancements, 100/105. It also defeats the Dragon and returns with the Egg on any seed. (1/4)

NOTE: Minecraft has been a challenging, unsolved AI benchmark for a decade, and in that time OpenAI (VPT, 2022), DeepMind (DreamerV3, 2023) and NVIDIA (Voyager, 2023) got their agents about as far as a diamond pickaxe, which is where we derive the 26x figure.

[Superbloom Labs @superbloom_labs]

Minecraft AGI takeoff incoming 👀

Steve seems to have figured the basics out now. Slowly, momentum builds.

Watch live: https://t.co/j7u90Cqsbb

[Superbloom Labs @superbloom_labs]

Meet Steve: the first self-improving AI to master Minecraft without human help.

It's a step to whole civilizations growing in virtual worlds, and a leap towards game characters that come to life.

Steve out-progresses 97% of players; 26x more than attempts by NVIDIA, DeepMind and OpenAI.

Characters can now develop new abilities and evolve through experience. Soon, in-game towns, civilizations, and whole ecosystems will increasingly look and feel like our own, or like a great story, and cheaply:

Once Steve’s capabilities exist, a System One model like Jev can take over, and it’s 92% cheaper. Soon, hundreds will be able to run locally alongside any game.

Steve completes ~96% of the game’s advancements, 100/105. It also defeats the Dragon and returns with the Egg on any seed. (1/4)

NOTE: Minecraft has been a challenging, unsolved AI benchmark for a decade, and in that time OpenAI (VPT, 2022), DeepMind (DreamerV3, 2023) and NVIDIA (Voyager, 2023) got their agents about as far as a diamond pickaxe, which is where we derive the 26x figure.

[Superbloom Labs @superbloom_labs]

For games, this opens a possibility we care deeply about: characters that can develop new abilities through experience.

Soon, in-game towns, civilizations, and whole ecosystems will increasingly look and feel like our own, and cheaply.

https://t.co/j7u90Cqsbb

[Superbloom Labs @superbloom_labs]

Meet Steve: the first self-improving AI to master Minecraft without human help.

It's a step to whole civilizations growing in virtual worlds, and a leap towards game characters that come to life.

Steve out-progresses 97% of players; 26x more than attempts by NVIDIA, DeepMind and OpenAI.

Characters can now develop new abilities and evolve through experience. Soon, in-game towns, civilizations, and whole ecosystems will increasingly look and feel like our own, or like a great story, and cheaply:

Once Steve’s capabilities exist, a System One model like Jev can take over, and it’s 92% cheaper. Soon, hundreds will be able to run locally alongside any game.

Steve completes ~96% of the game’s advancements, 100/105. It also defeats the Dragon and returns with the Egg on any seed. (1/4)

NOTE: Minecraft has been a challenging, unsolved AI benchmark for a decade, and in that time OpenAI (VPT, 2022), DeepMind (DreamerV3, 2023) and NVIDIA (Voyager, 2023) got their agents about as far as a diamond pickaxe, which is where we derive the 26x figure.

[Superbloom Labs @superbloom_labs]

We recognize that previous demonstrations have beaten Minecraft with substantial gameplay capability already built into the starting setup. We credit & appreciate earlier work for advancing autonomous play.

The milestone we’re claiming is autonomous development of the capabilities behind the victory:

For example, the most recent viral Dragon defeat, @rronak_ Astra + Jev demonstration, used a preselected seed on Peaceful difficulty, a route surveyed & mapped by a human in advance, supplied coordinates for resources, and an already-active End portal. Its configuration includes village and chest locations, portal destinations, and travel waypoints. These conditions are openly documented in the project’s repository.

Mine AI MCP similarly supplies human-authored skills, gameplay walkthroughs, and advice.

Steve’s experiment requires the player to identify and develop that gameplay competence itself. There are no human-authored gameplay skill libraries, prebuilt combat or survival controllers, progression scripts, Dragon walkthroughs, or Minecraft-specific guides.

We credit earlier work for advancing autonomous play, and applaud the work of Ronak and the long lineage before for advancing the field.

We also recognize the lineage of the name “Steve-1” from prior art, and distinguish that this is not a continuation or part of that research.

Yours,
Superbloom Labs

[Superbloom Labs @superbloom_labs]

Meet Steve: the first self-improving AI to master Minecraft without human help.

It's a step to whole civilizations growing in virtual worlds, and a leap towards game characters that come to life.

Steve out-progresses 97% of players; 26x more than attempts by NVIDIA, DeepMind and OpenAI.

Characters can now develop new abilities and evolve through experience. Soon, in-game towns, civilizations, and whole ecosystems will increasingly look and feel like our own, or like a great story, and cheaply:

Once Steve’s capabilities exist, a System One model like Jev can take over, and it’s 92% cheaper. Soon, hundreds will be able to run locally alongside any game.

Steve completes ~96% of the game’s advancements, 100/105. It also defeats the Dragon and returns with the Egg on any seed. (1/4)

NOTE: Minecraft has been a challenging, unsolved AI benchmark for a decade, and in that time OpenAI (VPT, 2022), DeepMind (DreamerV3, 2023) and NVIDIA (Voyager, 2023) got their agents about as far as a diamond pickaxe, which is where we derive the 26x figure.

[Superbloom Labs @superbloom_labs]

Steve is based on a simple insight: externalized cognition is a severely underutilized avenue for continual learning.

Imagine your favorite skill to practice. Maybe an instrument.

Learning your first musical scale requires intense focus from your ‘intelligence;’ your fingers feel alien to use all the sudden, but as you practice, it becomes intuitive and automatic. The behavior that used to take all your mental energy is embedded into the nervous system where it becomes a sort of automatic, recallable pattern; this finger here, that one there, and so on. You no longer need to use the same underlying intelligence each time, and can focus on greater ambitions like testing improvisation or learning Moonlight Sonata.

LLMs can do the same without updating the weights. Steve builds it outside the model as working software, and our proprietary system preserves useful patterns-of-thought & something akin to 'muscle memory' for the model to continually improve, till it reaches mastery of an environment.

In other words, we supply the general-purpose, ‘intuitive’ mechanisms for Steve to create, transfer, and propagate learned capabilities across future tasks as if it had a nervous system.

We are releasing information on how it works in a series of research papers over the coming weeks. We are also working with a few YouTubers, famous for gaming and AI content, who will act as independent reviewers and make long-form video essays about Steve, and large groups of intelligences like him living together.

Until now, externalized cognition for LLMs has mostly meant Markdown notes. As Dwarkesh Patel (@dwarkesh_official) puts it, an AI that learns by writing itself notes is like a saxophone student who inherits another student's notes: she gets an account of the practice, importantly not the muscle memory and patterns of thought, and has to learn the instrument all over again. So, he argues, real learning has to go into the weights.*

We flip that by treating the model as the underlying intelligence (what the student uses to figure things out in the first place) to task it with building the muscle memory and patterns of thought that represent true, progressive learning in the environment.

One run works out a better way to ‘play the instrument’ and captures it in something like a trained controller, so the next can use it to focus on greater ambitions like Giant Steps.

The weights never change (no re-training!) and the new student picks up the instrument already able to play it.

Once the abilities exist, a model 92% cheaper can take over and run them.

It is now time to make the most of the intelligence we already have. How could we not already have?

Models often already possess much of the expertise a difficult problem requires. The challenge is bringing that expertise to bear: asking the right questions, investigating uncertainty and translating understanding into something that works.

Steve connects that investigation to the agent’s actual software. Useful understanding becomes things like calculations, controllers, diagnostic tools, tests, procedures, and more, that operate on fresh observations and keep working while the model concentrates elsewhere.

[Superbloom Labs @superbloom_labs]

Steve's live run starts soon. Its world seed comes from drand round 32818211, public randomness, published at 8:39:57pm PT.

Run ID: steve-live-001
Seed = SHA-256 of the run ID, round and randomness

Watch & verify: https://t.co/fH3iUEnILK