The day in conversation for 2026-11-07
discussion
Transcript
The discussion: Tom: That 61.3% success rate for GPT-6 Sol on some tasks seems like a hard ceiling, but what about the information acquisition cost they mentioned; does that scale linearly with the horizon length? Jane: I'm more concerned with *why* it scales poorly beyond just the sheer number of reads. If we look at BizAgentBench, is that high cost primarily due to sequential decision-making requiring full context recalculation, or is it inefficient rule identification across the 789 documents? Lu: From a systems perspective, I wonder if the benchmark environment itself introduces state space complexity that overwhelms the agent's ability to filter relevant rules quickly. Have we seen similar bottlenecks when agents are tasked with long-horizon planning in purely symbolic environments? Meng: It feels like the core issue might be generalization rather than just compute time. If the agent struggles to identify *which* rule applies, does that imply a failure in the initial parsing or representation layer rather than just retrieval latency? Lalam: Speaking of representation, I was looking at NEAR. The training-free framework relies on local references derived from nearest neighbors; what kind of distance metric are they using for those embeddings, and how does that relate to the sparse graph coloring results we discussed last week regarding allocation guarantees? Tom: That’s a great pivot. The distance metric is key there. If it's based on cosine similarity, how does that differ from the structural constraints imposed by the
More episodes
- 2505.16254-Human-AI Writing as Distributed Cognition: Reassessing Collaborative Writing Theories in the Age of LLMs
- 2506.10913-Choreographic Quick Changes: First-Class Location (Set) Polymorphism
- 2507.14812-A Black-Box Approach for Exogenous Replenishment in Online Resource Allocation
- 2507.15448-Frames over finite fields with respect to the Galois inner product and their connections to self-orthogonal and LCD codes
- 2508.21572-NewsReX: An Open-Source Multi-Framework for Neural News Recommendation
- 2509.17028-Impact of Packetization on Network Calculus Analysis
- 2510.17342-AoA Services in 5G Networks: A Framework for Real-World Implementation and Systematic Testing
- 2511.04088-Efficient and rate-optimal list-decoding in the presence of minimal feedback
- 2511.17264-Notes on Stack Machines and Quantum Stack Machines
- 2601.15318-On the closest balanced game