How long can your skills be before your agent forgets what you told it? — Laurie Voss, Arize AI
Laurie Voss from Arize AI tested how many instructions AI models can follow when given a list of words that must appear in a response.
A year ago, models forgot instructions when the list grew to 200–300 words. Today, the latest models handle 5,000 instructions before failing — a tenfold improvement in twelve months. But models fail in different ways: some forget rules, others refuse due to safety checks, and some use all their computing power to verify instructions instead of answering. Voss says the problem is no longer getting instructions in, but verifying that the model actually follows them.
Open on YouTube →
Vibekollen prepared this summary with AI from the original publication. The content belongs to AI Engineer.
