Yes, the models got dumber — Better Than Good.
GPT-4 lost 95 points on prime number detection in three months. Model degradation via quantisation, routing changes, and silent updates is real and measurable.
GPT-4 lost 95 points on prime number detection in three months. Model degradation via quantisation, routing changes, and silent updates is real and measurable.
What remains if/when coding is ‘solved’