Giacomo Tesio argues that Large Language Models (LLMs) function as lossy compressions of their training data, rather than demonstrating true intelligence. He posits that LLMs mimic understanding by assembling statistically linked data fragments, which users are unlikely to recognize as direct quotations due to the vastness of the training material and the user's limited exposure to it. Tesio suggests this process is a misunderstanding of both the Shannon theorem and how LLMs operate. AI
IMPACT Challenges the notion of LLM intelligence, suggesting they are sophisticated data compression tools rather than cognitive entities.
RANK_REASON Opinion piece by a named individual discussing the nature of LLMs.
Read on Mastodon — fosstodon.org →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
