Poznámky k epizodě
In 2023, GPT-3.5 invented 55% of the scholarly references it cited in a set of literature reviews. Stanford researchers Owen Queen and James Zou sent the same prompts to models released in 2026: Claude Sonnet 5.5 invented about 1%, GPT-6.1 Sol none, and Qwen3.7-Flash 69%. They did this for 50 widely cited findings about language models, across 20 models released between July 2024 and September 2026. 22 of the 50 were classified as expired, using a threshold of one-fifth of the original effect.
Two of those findings come from the same 2023 paper on sycophancy, published by Anthropic…
Web: youtube.com

