Unverified50% confidenceOpinionExact time
Eric Bigelow believes that many capabilities already exist in pre-trained base models, and that subsequent SFT, RL, and agent fine-tuning are more about sharpening the distribution than endowing entirely new capabilities.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedLarge model training goes through three core stages: pre-training, supervised fine-tuning (SFT), and preference alignment.77% similarUnverifiedSpecialized super-resolution models carefully trained for specific domains often outperform general-purpose models.73% similarUnverifiedMainstream large model benchmarks such as MMLU, BigBench, and MATH have all been widely used in training data filtering and reinforcement learning reward signal design, making it increasingly difficult for high scores to map to real capability improvements.73% similarUnverifiedModels trained with reinforcement learning (RL) outperform models trained solely with SFT on Agent tasks.72% similarVerifiedTraining frontier large models requires enormous computational investment, and there are diminishing marginal returns as model scale increases.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1003515API
curl https://kongchang.com/api/v1/knowledge/claims/1003515MCP
get_claim(id=1003515)