Adaptation Odyssey in LLMs: Why Does Additional Pretraining Sometimes Fail to Improve? | Cohere | Research | TechiSeek

AddAdd your Listings

Research

Adaptation Odyssey in LLMs: Why Does Additional Pretraining Sometimes Fail to Improve?

In the last decade, the generalization and adaptation abilities of deep learning models were typically evaluated on fixed training and test distributions.

Research Cohere

ResearchNo Need for Explanations: LLMs can implicitly learn from mistakes in-contextCohere ResearchFrom Tools to Teammates: Evaluating LLMs in Multi-Session Coding InteractionsCohere ResearchAgents Explore but Agents Ignore: LLMs Lack Environmental CuriosityCohere ResearchProcedural Knowledge in Pretraining Drives Reasoning in Large Language ModelsCohere ResearchHow to Improve the Robustness of Closed-Source Models on NLICohere ResearchRewardBench 2: Advancing Reward Model EvaluationCohere BlogsVision Model Platform Updates: Enhanced Capabilities and New FeaturesFireworks BlogsWhy do all LLMs need structured output modes?Fireworks BlogsFrontier-lab training infrastructure, now as a serviceFireworks BlogsGLM 5.2 Fast is live on FireworksFireworks BlogsIntroducing OpenAI gpt-oss (20b & 120b)Fireworks

BlogsInference Providers vs. API Routers: Where Do Your Tokens Actually Come From?Fireworks BlogsIntroducing the Flags Explorer, first-party integrations, and updates to the Flags SDKVercel BlogsThe value of llms.txt: Hype or real?Mintlify BlogsWhat is llms.txt? Breaking down the skepticismMintlify BlogsWhat is MCP and how to get startedMintlify BlogsThe Developer’s Intro to Core Web VitalsNetlify Products & ServicesAI docs readership increased over 500% in 2025. What does it mean for you?Gitbook Products & ServicesSupporting AI standards at GitBook: What OKF, MCP and llms.txt tell us about the future of docsGitbook BlogsOpenAPI Operation IDs make good API designRedocly