Keep pulling the thread on Brandon Foodi.
Mercore recently raised $100 million in a funding round.
Brandon Foodi predicts that LLM-based candidate assessment will soon become so effective that it will be 'foolish' for humans to believe they can perform better.
Brandon Foodi asserts that the data annotation market has shifted away from a crowdsourcing model, which worked for early models like ChatGPT, because current advanced models require high-quality, expert evaluators to identify complex errors.
Brandon Foodi predicts that data annotation companies reliant on high-volume, low-skill crowdsourcing pipelines will experience significant churn, while new players focused on high-quality talent will gain market share.
Brandon Foodi expects the market for human-generated data and evaluations to increase by orders of magnitude over time.
Mercore's strategy involves first dominating the human data market for AI labs, then expanding to high-volume contractor hiring for large tech companies to compete with firms like Accenture and Deloitte, and finally moving into all full-time hiring.
Brandon Foodi predicts that a significant portion of knowledge work will shift towards humans creating evaluations ('evals') for tasks that AI models cannot yet perform, as this is more efficient than humans repeatedly performing the tasks themselves.
Brandon Foodi claims that many types of Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF) data are no longer as useful as they once were, and corporate budgets for them are decreasing.
Brandon Foodi predicts that within the next two to three years, the primary public concern regarding AI will be that models are significantly better than humans at their jobs, creating a need to redefine people's roles in the economy.
Brandon Foodi believes that SFT and RLHF data are overhyped, stating that companies are spending billions on this data when they could be spending an order of magnitude less.
Brandon Foodi predicts that an AI software engineer capable of writing a pull request with a higher success rate than a human will be available later in 2024 or in the first half of 2025.
Mercore's platform is used for data labeling, talent screening, performance prediction, and the evaluation of both human and AI candidates.