conversation

Shane Parrish shares insights from a conversation with Shopify CEO Tobi Lütke about AI usage at Shopify and the impact of competition on tech companies.

Kent Dodds suggests that with current AI models, good primitives and brief conversations are sufficient for tasks, and he hasn't used plan mode in months amid discussions about removing it.

This paper introduces a three-layer checklist-and-judge framework to evaluate the communicative success of interpreter agents in machine-translated conversations across semantic, pragmatic, and cultural-social dimensions, validated through extensive benchmarks.

Shane Parrish interviews Shopify co-founder Tobi Lütke on super intelligence, where Lütke describes society as a form of super intelligence governed by systems.

Introduces Rationale-Guided Learning (RGL), a framework that reframes multimodal emotion recognition in conversation as a cognitively-inspired reasoning task using dual-process theory and MLLM-generated rationales, achieving state-of-the-art results on IEMOCAP and MELD.

This paper compares generation and judgement paradigms for LLM-based emotion-cause pair extraction in conversation, finding that pair-level judgement outperforms dialogue-level generation, and introduces an auxiliary retriever for improvement.

This paper introduces AtmosERC, a model that models dialogue-level affective atmosphere to enhance emotion recognition in conversations.

A speculative look at how conversational AI might evolve by the year 2026.

This paper introduces a framework to convert static tasks into dynamic multi-turn conversations to evaluate how well LLMs track evolving user intent, finding that strong static performance does not transfer to dynamic settings.

Explores the idea of treating an AI agent's conversation history as a simple file, proposing a minimal and transparent approach to agent state management.

A record of the conversation between Tian Yuandong and Holly Zheng about transcending superintelligence at the AGI Summit.

OpenAI highlights improvements in GPT-Live, showcasing its ability to maintain conversations while handling multiple tasks like checking flights, weather, and creating itineraries in real time.

Dwarkesh Patel announces an upcoming off-the-record conversation with Dr. Fei-Fei Li on July 22, with limited spots available.

TurnNat is a likelihood-based framework for automatically evaluating turn-taking naturalness in dyadic spoken dialogue, using a causal turn-taking prediction model trained on natural conversations to measure timing atypicality via negative log-likelihood.

This article argues that traditional prompt engineering is obsolete; modern LLMs are intent reconstructors, and interactions should be through natural, rich conversation rather than condensed instructions.

An essay exploring why thinking out loud with another person produces better understanding and insight than solitary reflection, drawing on cognitive science and philosophy.

A user reports that Claude repeatedly assumed suicidal intent during a scientific discussion about paraquat, despite explicit denials over 20 times, degrading the service and raising concerns about over-sensitive safety mechanisms.

A developer tested how small edge models (LFM2.5, Gemma variants) retain a single fact across conversation turns, finding that models often confidently deny knowing information that remains in context, posing a trust issue for agent architectures and suggesting a trade-off between memory and format discipline.

Lantern introduces a lightweight memory layer that archives conversation turns and retrieves relevant details after compaction, recovering 78.3% of lost facts with zero LLM calls and outperforming MemGPT-based methods.

Jeff Dean thanks the host of Two Minute Papers for a conversation about AI research, with a link to the full video in the thread.