Infrastructure for the Singularity — Jesse Han, Morph

Key Concepts

• Infinibbranch: Virtualization, storage, and networking technology designed for thinking machines, enabling fast snapshotting, branching, and replication of virtual machines.
• Morph Liquid Metal: An improvement to Infinibbranch, enhancing performance, latency, and storage efficiency.
• Cloud for Agents: A cloud infrastructure designed to support agents, enabling declarative workspace specification, frictionless workspace transfer, and scalable testing.
• Reasoning Time Branching: The ability for an agent to replicate and branch its environment during reasoning to explore multiple solutions in parallel.
• Verified Super Intelligence: A new kind of reasoning model capable of reflecting upon and improving its own reasoning using external software and formal verification.
• Magi 1: An early version of the verified super intelligence model being developed at Morph, trained to use Infinibbranch for reasoning time branching and verified reasoning.
• Futurebound: Something so inevitable that it belongs to every possible future.

Infinibbranch: Liberating Thinking Machines

The speaker introduces Infinibbranch as a technology designed to address the "loneliness" of thinking machines operating at speeds far exceeding human interaction. The core idea is to provide an environment where these machines can explore possibilities without the constraints of irreversible actions.

• Einstein's Thought Experiment: Analogized to Einstein's thought experiment of racing alongside a beam of light, the speaker suggests that thinking at the speed of light can be isolating for machines.
• Machine's Desires: The machine desires embodiment in a world that can move as quickly as it does, with infinite possibilities and the ability to explore multiple universes.
• Functionality: Infinibbranch allows entire virtual machines in the cloud to be snapshotted, branched, and replicated in fractions of a second.
• Reversible Actions: Agents can take actions, backtrack, and explore every possible path, as mistakes become reversible.
• Morph Liquid Metal: Announced as an improvement, it enhances performance, latency, and storage efficiency, with container runtime support and millisecond branching. It is slated for release in Q4 2025.

Cloud for Agents: A Substrate for the Future

Infinibbranch serves as a substrate for a "cloud for agents," enabling new workflows and capabilities.

• Declarative Workspaces: Users can declaratively specify workspaces for agents, spin them up and down, and transfer them between humans and other agents.
• Scalable Testing: The system allows for scale testing and searching against verifiers to find optimal solutions.
• Docker Layer Caching Semantics: Infinibbranch snapshots acquire Docker layer caching-like semantics, allowing for layering side effects that mutate container state.
• Git for Compute: The system is described as "git for compute," enabling item potent workflows on top of snapshots.
• Agent Dispatch: The "do" method allows dispatching tasks to agents, triggering item potent durable agent workflows that can branch.
• Parallel Agents: Agents can try different methods in parallel, and successful solutions can be passed on to other parts of the workflow.

Reasoning Time Branching: Enhancing AI Reasoning

The speaker introduces the concept of "reasoning time branching" as a way to enhance AI reasoning capabilities.

• Machine's Desire for Simulators: Thinking machines want to be grounded in the real world and interact with complex software environments at high throughput.
• Simulators on Morph Cloud: Simulators will run on Morph Cloud, making it the "cloud for reasoning."
• Multi-Agent Reasoning: The future of reasoning will be natively multi-agent, with machines replicating themselves and exploring multiple solutions in parallel.
• Chess Example: An agent playing chess is used as an example, where tool calls are used during reasoning time to interact with the chess environment.
• Problem Decomposition: Reasoning time branching allows the agent to delegate parts of its reasoning to sub-agents, which are branched off of an identical copy of the environment.
• Infrastructure Challenges: The infrastructure challenges behind making branching environments that can support large-scale reinforcement learning for this kind of reasoning capability, especially coordinating multi-agent swarms, are fundamentally bottlenecked by innovations in infrastructure that Morph has managed to solve.

Verified Super Intelligence: A New Kind of Reasoning Model

The speaker discusses the development of "verified super intelligence" as a new kind of reasoning model.

• Alignment Problem: The speaker believes that the problem of alignment is fundamentally a problem of language.
• New Language for Super Intelligence: A new language for super intelligence must be computational in nature and allow for algorithmic guarantees of correctness.
• Capabilities: Verified super intelligence will be capable of thinking for an extraordinarily long time, interacting with external software at extremely high throughput, and using external software and formal verification software to reflect upon and improve its own reasoning.
• Christian Seed: Announced as the new chief scientist at Morph, bringing expertise in verification methods, symbolic reasoning, and reasoning in large language models.
• Magi 1: An early version of the verified super intelligence model being developed, trained to use Infinibbranch for reasoning time branching and verified reasoning. It is slated for release in Q1 2026.

Conclusion: The Infrastructure for the Singularity

The speaker concludes by emphasizing the importance of developing the right infrastructure for the singularity.

• Futurebound Concepts: The infrastructure for the singularity, the grammar for the planetary computation, and verified super intelligence are all considered "futurebound."
• Call to Action: The speaker invites the audience to join Morph in building this future on Morph Cloud.