Search

Tag: #inference-scaling6 results

自適應並行推理：高效推理擴展新範式

Berkeley AI Research 文章介紹自適應並行推理，讓 LLM 動態分解任務、產生並行執行緒，並依問題需求協調。分析並行推理進展，以克服順序推理的線性擴展問題，如上下文腐敗和高延遲。動機來自推理時擴展主導數學、程式設計和代理基準。

Berkeley AI ResearchResearch•May 8#parallel-reasoning#inference-scaling#context-rot

英偉達Q4營收因Agentic AI爆發飆升

英偉達公布Q4創紀錄681億美元營收，資料中心成長75%至623億美元。黃仁勳宣告Agentic AI拐點，Claude Code等工具驅動推理需求。Blackwell降低token成本10倍，Rubin接力。

虎嗅Media•Feb 26#agentic-ai#earnings-report#inference-scaling

Tokens Enable Emergent Resource Rationality

Inference-time scaling in language models leads to adaptive resource rationality without explicit cost rewards. Models shift from brute-force to analytic strategies as task complexity rises. LRMs show robustness on challenging functions like XOR/XNOR unlike IT models.

ArXiv AIResearch•Feb 12#research#language-models#v1

Inference Scaling vs Larger Tasks

Distinguishes inference scaling from natural compute increases for bigger tasks in LLMs. Uses Pareto frontiers of compute budget vs. task time-horizon to analyze efficiency. Argues true scaling concerns arise only when exceeding human-equivalent costs inefficiently.

AI Alignment ForumCommunity•Feb 11#research#llms#ai