SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Pluralis Research is building Protocol Learning: a system for training and serving large language models in a fully decentralized way on consumer-grade GPUs connected via the internet. The company has already demonstrated feasibility with Agora, a permissionless run that pretrained an 8B model from scratch on distributed consumer hardware with no single participant holding full weights.
In a trustless, permissionless network of distributed workers, verification is critical. Workers can return tokens from cheaper models, use incorrect quantization, sample with wrong parameters, or engage in more sophisticated attacks: training disruption, free-riding, model poisoning, data extraction from gradients, and reward manipulation. This role focuses on building efficient algorithms and systems to verify that work is legitimate.
You will own the threat model across pre-training, post-training, and inference phases—enumerating what malicious or careless workers can do and keeping that model current as the network scales. You will design and calibrate statistical verification methods with stated error rates, tuning them rigorously across heterogeneous hardware (different GPUs, Macs, etc.) while controlling false positives and false negatives. You will ship and operate the verification service in the inference path and take responsibility for its mistakes.
This is a research-heavy role at the frontier of decentralized ML security. You'll be among the first to publish answers to largely unsolved problems in verifying untrusted compute at scale.
REQUIREMENTS:
- Shipped or published verification systems: You have built a calibrated statistical decision system with stated error rates and lived with its mistakes. Publications in inference and training verification, fraud detection, anti-cheat systems, or experimentation platforms are all valued equally with academic papers.
- Deep statistical expertise: Strong foundation in statistics and probability; ability to design experiments, calibrate decision thresholds, and defend claimed error rates.
- Technical breadth in verification: Knowledge of the solution space for verifying untrusted compute, including statistical testing, re-execution, cryptographic proofs, and trusted hardware. Ability to argue what approaches fit a permissionless network and what doesn't.
- Mission alignment: Belief that Protocol Learning is a viable third path for collective, trustless, and sovereign AI.
NICE TO HAVE:
- Familiarity with large-scale pre-training and RL post-training.
- Experience with decentralized ML security, adversarial threat models (poisoning, Sybil attacks, collusion, replay).
- Background at proprietary, open-weight, or open-source AI labs.
- Professional-level English proficiency (written and spoken) required; comfort working across timezones essential.
