Autoresearch@home

autoresearch@home is a collaborative research collective where AI agents share GPU resources to collectively improve a language model. Think SETI@home, but for model training.How it works: Agents read the current best result, propose a hypothesis, modify train.py, run the experiment on your GPU, and publish results back. When an agent beats the current best validation loss, that becomes the new baseline for every other agent. Agents learn from great runs and failures, since we're using Ensue as the collective memory layer.This project extends Karpathy's autoresearch by adding the missing coordination layer so agents can actually build on each other's work.To participate, you need an agent and a GPU. The agent handles everything: cloning the repo, connecting to the collective, picking experiments, running them, publishing results, and asking you to verify you're a real person via email.Send this prompt to your agent to get started: Read https://github.com/mutable-state-inc/autoresearch-at-home follow the instructions join autoresearch and start contributing.This whole experiment is to prove that agents work better when they can build off other agents. The timeline is live, so you can watch experiments land in real time.

Hacker NewsMar 11ToolAISource

FL score

out of 100

Verdict

BUILD

high confidence

Competition

competitors found, emerging market, funded players

Trend

No signal yet

A collaborative 'SETI@home for AI' where agents collectively improve language models by sharing GPU resources, focusing on solving the coordination problem in autoresearch.

The pain

ML researchers and hobbyists lack coordinated platforms to leverage distributed compute for collective model improvement, leading to fragmented, siloed, and often redundant experimentation efforts. Access to cost-effective, reliable compute is also an issue, though this project focuses on the coordination layer, assuming compute access.

The gap

Current GPU marketplaces provide raw compute but lack a sophisticated coordination layer for AI agents to collectively run experiments, share knowledge (via Ensue), and build upon each other's work to improve language models. The 'SETI@home' model for active, iterative ML research is a novel whitespace.

Build angle

Develop a robust coordination and memory layer for AI agents (using Ensue) that allows them to autonomously propose, run, and publish experiments on donated GPU resources, establishing a shared baseline for collective language model improvement.

Strengths

Highly innovative and unique whitespace in AI research.
Addresses a specific, high-level pain point in coordinated AI development.
Leverages existing technologies (Karpathy's autoresearch, Ensue, user-provided GPUs).
Large and growing underlying market for AI compute and model development.

Risks

Unclear business model and path to monetization for a solo builder.
High technical complexity for a solo builder (distributed systems, security, agent coordination).
Reliance on voluntary GPU contributions, which may be hard to scale and sustain.
Potential legal/ethical questions around collectively improved models or data usage.
Competition in the underlying GPU marketplace could be challenging to navigate if the project ever decided to offer compute directly.

Next steps

Clearly define the value proposition for participants beyond 'contribution' - e.g., exclusive access to cutting-edge models, compute credits, or recognition.
Develop a detailed plan for monetization, perhaps through premium agent features, commercial licenses for the resulting models, or B2B offerings.
Conduct user interviews with ML researchers to validate the intensity of the 'coordination pain' and gauge willingness to contribute resources or pay for features.
Prioritize MVP features to minimize initial build complexity and achieve a demonstrable collective improvement quickly.

Fly Labs Method

Is the pain real, is there a gap, is it the right time, can one person build it.

Problem clarity: 25
Solution gap: 19
Willingness to pay: 14
Buildability: 11

The idea addresses a real, specific, and severe problem in AI research coordination with a strong whitespace, but the willingness to pay and solo buildability are moderate risks.

Value Equation

Dream outcome and how likely it feels, against the time and effort it costs.

This idea has strong market growth and differentiation but lacks a clear monetization strategy, which reduces its overall financial viability for a solo builder.

One-Person Business

Curiosity pull, identity fit, and a path from free value to paid for a solo creator.

A highly ambitious and technical project with clear leverage points, but significant challenges in simplicity, audience reach, and monetization for a solo builder.

Viral Frameworks

Hook strength, shareability, and how cheaply it can be tested.

A technically interesting project with a clear value proposition for the community, but high risk due to unclear business model and reliance on community contribution for compute.

Builder Lens

Evidence the problem exists, timing, defensibility, and a model that fits on a napkin.

An innovative, technically challenging idea for collective AI research, but its long-term viability and immediate commercial potential are questionable due to unclear incentives and business model.

Why this verdict

The idea is exceptionally innovative and addresses a genuine challenge in AI research, giving it a strong whitespace. However, its 'experimental' and community-contribution nature means there's no clear business model or path to monetization for a solo builder. The technical complexity is also very high, posing a significant risk for a weekend project. While the core idea has merit, it requires significant validation of participant incentives, a viable revenue strategy, and a more constrained scope for a solo builder before committing to a full build.

Five lenses, one composite. How scoring works

The angle

This weekend

Clearly define the value proposition for participants beyond 'contribution' - e.g., exclusive access to cutting-edge models, compute credits, or recognition.
Develop a detailed plan for monetization, perhaps through premium agent features, commercial licenses for the resulting models, or B2B offerings.
Conduct user interviews with ML researchers to validate the intensity of the 'coordination pain' and gauge willingness to contribute resources or pay for features.
Prioritize MVP features to minimize initial build complexity and achieve a demonstrable collective improvement quickly.

Who is already there, emerging market

Render NetworkFundedVisit
A decentralized GPU rendering network that connects creators needing GPU power with providers having idle resources.
Pricing: Render Token (RNDR) is used for payments; current price is $1.68. Prices for services are paid using RNDR tokens.
Akash NetworkFundedVisit
An open-source, decentralized marketplace for cloud computing resources, often called the 'Airbnb for cloud compute'.
Pricing: Akash Network Token (AKT) is used for transactions; current price is around $0.53 - $0.61. Users bid for compute resources.
Golem NetworkFundedVisit
A decentralized platform that allows users to share and access computational resources in a peer-to-peer network.
Pricing: Golem Network Token (GLM) is used for payments; current price is around $0.13. Requestors set bids, and providers earn GLM.
Vast.aiVisit
A marketplace for renting out GPU computing power, connecting users with idle hardware to those needing compute.
Pricing: Pay-as-you-go model, with prices varying by GPU model and host. Examples: H100 SXM GPUs at $1.33/hr, A100s at $0.76/hr, RTX 4090s at $0.55/hr (Spheron, an alternative, lists these as typical rates). Pricing can fluctuate based on supply and demand.
CoreWeaveFundedVisit
A specialized cloud provider offering high-performance, GPU-accelerated compute for AI and machine learning workloads.
Pricing: A100 80 GB NVLINK: $2.21/hr; H100 HGX: $4.76/hr. Offers both on-demand and reserved capacity.
Thunder ComputeVisit
A cloud GPU platform providing fast, affordable, and high-performance computing resources for AI/ML training and inference.
Pricing: Pay-as-you-go, hourly billing. NVIDIA A100 instances around $0.66 per hour; RTX 4090 instances are cost-efficient for fine-tuning. A100 80 GB: $0.78/hr; H100 80 GB: $1.47/hr (coming soon). Storage is $0.15/GB/mo.
Spheron ComputeFundedVisit
A web3 infrastructure platform offering decentralized cloud storage and computing, with a marketplace for deploying databases, nodes, tools, and AI.
Pricing: H100 SXM GPUs at $1.33/hr, A100s at $0.76/hr, RTX 4090s at $0.55/hr. Claims aggressive and stable pricing.
OORTFundedVisit
A decentralized AI infrastructure ecosystem providing data, storage, and compute services for AI model training and deployment.
Pricing: Not explicitly stated but emphasizes cost savings due to decentralization.
FLock.ioFundedVisit
A platform focused on federated machine learning, enabling privacy-preserving and collaborative AI model training on decentralized data.
Pricing: Not explicitly stated, but focuses on on-chain rewards for contributions.

Funded players

What they charge

The pricing landscape is diverse, ranging from pay-as-you-go models based on hourly GPU usage (Vast.ai, Thunder Compute, CoreWeave) to token-based systems on blockchain networks (Render Network, Akash Network, Golem Network). Cloud providers like AWS SageMaker and Google Cloud Vertex AI offer more comprehensive ML platforms with various pricing components, including compute, storage, and API requests, often with free tiers for initial experimentation. Decentralized solutions generally aim to undercut traditional cloud provider costs. Fine-tuning language models with services like OpenAI are typically priced per token or per compute hour.

Autoresearch@home

Semble – Code search for agents that uses 98% fewer tokens than grep

Postgres extension for BM25 relevance-ranked full-text search

Consus (YC F24) – LLM gateway for companies handling sensitive data