The Agent Architect #36: Uncertainty Quantification

今週のパターン

Uncertainty Quantification

概要:

モデル由来の不確実性と、削減できないノイズを切り分け、点推定に加えて予測範囲と信頼水準を提示する。

使いどころ:

予測の信頼性をユーザーが知る必要がある重大な意思決定や、下流システムが信頼度に応じて挙動を変える必要がある場面。

注意点:

キャリブレーションが甘い区間はユーザーに誤った安心感を与える。ホールドアウトデータでの検証は不可欠だが、省略されがち。

今週のエージェントAI

• Experiment with Qwen3.8-Flash-Next on NVIDIA GB300 NVL72 for Agentic CodingNVIDIA Developer

  Alibaba released Qwen3.8-Flash-Next model weights, a multimodal mixture-of-experts model with 125B main parameters.

• Experiment with Qwen3.8-Flash-Next 176B Model on NVIDIA GB300 NVL72 for Agentic CodingNVIDIA Developer

  Alibaba released Qwen3.8-Flash-Next model weights, a 176B parameter multimodal mixture-of-experts model for developer evaluation.

• When Does Bigger Help? A Controlled Study of LLM Scale for Ontology LearningarXiv cs.AI

  Controlled evaluation of 13 LLMs across Qwen and GPT variants shows varying effects of model scale on ontology learning performance.

• A Model with No Head and Many ThoughtsarXiv cs.AI

  Method replaces vocabulary projection with lightweight projector to enable reasoning in embedding space.

• AI Agent Latency 101: How do I speed up my AI agent?LangChain

  Strategies to reduce AI agent latency include optimizing LLM calls, enabling parallelism, and improving user experience.