Navigate Select ESC Close

The Hidden Reason LLMs Fail in Conversations: CCOPD

2026-05-31 Science & Technology
220
21
2
Discover AI
Discover AI
88.6k subscribers

Unlock all features

FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.

Description

Your AI Chatbot Is Gaslighting Itself. Why? And how can you fix it? All the answers in this video. See scientific pre-print below. All rights w/ authors: Same Evidence, Different Answers: Canonical-Context On-Policy Distillation for Multi-Turn Language Models Zizhuo Lin1,† Quanling Liu1 Jinsheng Quan1 Chao Zhang1 Yifan Zhu1 Xing Shi1 Jingtao Xu1 Zhihui Li2 Yawei Luo1 from 1 Zhejiang University 2 University of Science and Technology of China #airesearch #scienceexplained #artificialintelligence #aiagents #aiagent

Top Comments (4)

@hitmusicworldwide 2026-05-31

I never ran into this problem when dealing with the major language models

1 2 replies
@BobWidlefish 2026-05-31

Might be better to implement in harness instead. Hmm.

0
@LuisBustos-jq8sz 2026-05-31

Nice it reminds me of this paper about a teacher student method where you train the model solving math or code problems and the teacher have privileged information then i also read this paper about a method about using a method to by pass catastrophic forgetting using like small modules in the weights so i thought about doing a cybernetic loop to create an autonomous learning process where the model learn the inference when you talk with it sadly the cost of training a model is prohibitely large for me so i could never try it.

1 2 replies
@vascoduarte4519 2026-06-02

What conclusion did YOU draw for multi agent systems?

0

Unlock the Data Inside
Turn Videos into Knowledge

  • Get FREE 10/day: transcripts, summaries, chats
  • Chat with videos, export text & PDF
  • $1 free API credit for RAG, chatbots & research

Free forever plan • All features unlocked

App screenshot