Conversation

Notices

Ars Technica (arstechnica@mastodon.social)'s status on Tuesday, 14-Jan-2025 23:40:19 JST Ars Technica

Amazon must solve hallucination problem before launching AI-enabled Alexa
Rollout of upgraded voice assistant hit by delays.
https://arstechnica.com/ai/2025/01/amazon-must-solve-hallucination-problem-before-launching-ai-enabled-alexa/?utm_brand=arstechnica&utm_social-type=owned&utm_source=mastodon&utm_medium=social
In conversation about 4 days ago from mastodon.social permalink
Attachments
1. Untitled attachment
  https://files.mastodon.social/media_attachments/files/113/827/224/665/235/081/original/6a1798917b0876b3.jpg
2. Amazon races to transplant Alexa’s ‘brain’ with generative AI
  
  Rollout of upgraded voice assistant hit by delays as Big Tech group toils to overcome technical hurdles
- lemgandi (lemgandi@mastodon.social)'s status on Wednesday, 15-Jan-2025 00:24:17 JST lemgandi
  in reply to
  
  @arstechnica "We are working extremely hard on it". Ha ha ha ha ha ha ha ha.
  There's already good evidence that this is equivalent to proving the parallel postulate or touching your right elbow with your right hand. CF: https://arxiv.org/abs/2401.11817
  In conversation about 4 days ago permalink
  Attachments
  1. Hallucination is Inevitable: An Innate Limitation of Large Language Models
    
    Hallucination has been widely recognized to be a significant drawback for large language models (LLMs). There have been many works that attempt to reduce the extent of hallucination. These efforts have mostly been empirical so far, which cannot answer the fundamental question whether it can be completely eliminated. In this paper, we formalize the problem and show that it is impossible to eliminate hallucination in LLMs. Specifically, we define a formal world where hallucination is defined as inconsistencies between a computable LLM and a computable ground truth function. By employing results from learning theory, we show that LLMs cannot learn all of the computable functions and will therefore always hallucinate. Since the formal world is a part of the real world which is much more complicated, hallucinations are also inevitable for real world LLMs. Furthermore, for real world LLMs constrained by provable time complexity, we describe the hallucination-prone tasks and empirically validate our claims. Finally, using the formal world framework, we discuss the possible mechanisms and efficacies of existing hallucination mitigators as well as the practical implications on the safe deployment of LLMs.

Public

Conversation

Notices

Feeds