Magic AI Labs Unveils HashHop for Improved Long-Context AI Evaluations with New Funding

Aug 29, 2024, 06:13 PM

Magic AI Labs has introduced HashHop, a new algorithm designed to improve long-context evaluations in AI models, including their 100M context model. The approach involves synthetically generating data to enhance long-context reasoning abilities. HashHop provides an alternative to existing methods by using markers or hashes in a large codebase, which correspond to functions or lines, to improve model predictions. The innovation has been well-received, with notable recognition for addressing weaknesses in popular long-context evaluations. Magic AI Labs also announced new funding, supporting further development of this technology.