Research Engineer – Decentralized Training and Inference Verification

This is a fully remote position, open to applicants in United States, +1 more country.

📋 Description

• Identify and sustain the threat model concerning malicious or negligent workers throughout pre-training, post-training, and inference phases.

• Address issues such as training disruption, denial-of-service, free-riding, model poisoning, backdoors, data extraction, reputation manipulation, and reward manipulation.

• Develop statistical verification techniques with clearly defined error rates.

• Optimize verification tests using stringent benchmarks.

• Manage false positives and false negatives across diverse hardware platforms, including various GPUs and Macs.

• Construct and manage the verification service within the inference workflow.

• Assume accountability for verifier performance and any errors that may occur.

• Create efficient algorithms and systems to confirm that tokens originate from the specified model and sampling parameters.

• Validate that the contributions from training participants align with their claims.

⛳️ Requirements

• Proven experience in developing a calibrated statistical decision system with specified error rates and managing its shortcomings.

• Published work in the fields of inference and training verification or relevant experience in fraud detection, anti-cheat measures, or experimentation platforms.

• Extensive knowledge of statistics and probability.

• Proficiency in designing experiments, calibrating decision thresholds, and justifying claimed error rates.

• Understanding of verifying untrusted computations, including statistical testing, re-execution, cryptographic proofs, and trusted hardware.

• Capability to evaluate which verification methods are suitable for a permissionless network.

• A strong belief in the feasibility of Protocol Learning for collective, trustless, and sovereign AI.

• Familiarity with large-scale pre-training and reinforcement learning post-training.

• Knowledge of decentralized machine learning security and adversarial threat models, such as poisoning, Sybil attacks, collusion, and replay attacks.

• Experience working in proprietary, open-weight, or open-source AI laboratories.

• Professional-level proficiency in English, both written and spoken.

🏝️ Benefits

• Substantial equity ownership for key technical contributors, complemented by a competitive base salary.

• Flexible working environment with team members located globally.

• Optional full visa sponsorship and relocation assistance to either Australia or the United States.