How to earn up to $2M training AI models with anonymized company data

Micro 1's Data Partnerships Program Overview

Introduction to Micro 1 and the Data Partnerships Program

• The session features Ali Ansari, CEO of Micro 1, and Solomon Anise, Strategic AI Lead, discussing the company's data partnerships program aimed at transforming existing workflows into revenue streams.

• The program is recruiting a small cohort over two weeks to provide information on how companies can leverage their daily operations for potential earnings between $100,000 and $2 million.

Who is Micro 1?

• Micro 1 is described as a data lab that assists foundational model companies in enhancing their capabilities through frontier datasets and reinforcement learning (RL) environments across various domains like finance and medical.

• They also collaborate with enterprises to evaluate and improve agents via contextual evaluations tailored to specific workflows.

Key Qualities of Data

• Two essential qualities of data emphasized are:

• Long Horizon Tasks: Complex tasks reflecting expert-level work that generates valuable outputs for model training.

• Real-worldness: Ensuring data points closely resemble real-world scenarios to enhance model performance in production settings.

How the Data Partnerships Program Works

Initial Steps for Companies

• Interested companies begin with an exploratory call where they learn about the partnership process.

• An intake form allows companies to outline the types of data they are willing to share, leading to a proposal generation from Micro 1.

Collaboration Process

• Close collaboration occurs between Micro 1’s team and the company’s engineering staff for secure data transfer while ensuring anonymization.

• There are two main pipelines:

• Corpus: Involves sharing existing anonymized company data.

• Workflows: Engages employees in developing prompts for training models based on real-world tasks.

Effort Required from Participating Companies

Time Commitment

• For corpus participation, creating a data manifest takes approximately 1–3 hours; transferring anonymized data may require an additional up to 10 hours from one or two engineers.

Workflow Participation

• In workflow participation, employees develop prompts related to their jobs which helps train AI models effectively by providing qualitative feedback on agent performance.

Understanding Real-world Workflows

Example of RL Environments

• RL environments consist of three components:

• Seeding Data: Documents or files relevant to tasks (e.g., Slack channels).

• Tooling: Software used by agents (e.g., project management tools).

• Task Definition: Specific prompts guiding agent actions along with verifiers assessing output quality.

Target Audience for the Program

Company Fit Criteria

• There are no strict industry limitations; all types of companies are encouraged to participate as diverse knowledge enhances general intelligence development in AI models.

Referral Program Details

Incentives for Referrals

• A referral program offers $50,000 per successful referral that leads to a partnership with Micro 1. This incentivizes individuals who know potential partner companies.

Privacy Measures in Data Handling

Security Protocol Overview

• Secure transfer methods include cloud storage or encrypted file transfers agreed upon mutually between parties involved.

PII Management Strategies

• Proprietary tools automate PII removal while trained human reviewers ensure thorough scrubbing before any client access.

Retention Policies

• Default retention period is set at 60 days post-transfer but can be customized according to company needs.

Final Review Process

• Companies review representative samples of scrubbed data before it reaches clients ensuring transparency throughout the process.

Audience Q&A Highlights

Interest in Diverse Real-world Data

• While there’s a bias towards English-based information currently, all kinds of real-world complex data are welcomed from various companies globally.

Ongoing Cooperation Possibilities

• The goal is establishing recurring revenue streams through options like corpus refreshes after certain periods or expanding workflow collaborations based on task submissions by employees.

Exploring Team Contributions and Revenue Streams

Testing Team Contributions

• Companies are experimenting with team contributions by assigning a few tasks per member weekly, gradually increasing participation.

• This approach has led to significant recurring revenue streams, with some teams contributing almost entirely.

Potential for Recurring Revenue

• The potential for recurring revenue can reach millions, with rare cases exceeding $10 million in run rates.

• While the nature of this revenue is recurring, there is no long-term commitment regarding its duration.

Data Reinvigoration and Workflow Evaluation

• Continuous evaluations of tasks created by team members occur weekly to enhance data quality and workflow efficiency.

Program Participation and Future Plans

Company Cohort Size

• A small cohort of companies will be accepted into the program within two weeks, aiming for around 50 participants initially.

Long-Term Goals

• The goal is to expand partnerships significantly, targeting hundreds to potentially thousands of companies in the near future.

Closing Remarks and Contact Information

Invitation to Participate

• Interested parties can find more information about the program at micro1.ai/data.

Follow-Up Communication

• All relevant links and information discussed during the seminar will be sent via email. Participants are encouraged to reach out directly for further involvement.