Unverified60% confidenceFactExact time
GPT-series models use unidirectional autoregressive pre-training and can only see historical information to the left of the current token.
2
Sources
60%
Confidence
Long-term
Relevance
7/5/2026
First Seen
Sources
Has AI Coding Assistant Fable Been Nerfed? Developer's Routing Anomaly Report Sparks Debate
twitterNickADobos7/2/2026
Related Claims
UnverifiedGPT系列采用单向（因果）注意力掩码，使模型在预训练阶段每个token只能关注其之前的上下文，与BERT等双向模型形成根本区别84% similarUnverifiedGPT 系列、Llama 等模型的核心能力绝大部分来自预训练而非后续微调阶段81% similarVerifiedThe GPT series adopts the autoregressive training objective of "predicting the next word".78% similarUnverifiedThe poster speculated that GPT-6 and GPT-6.1 Sol share the same pre-training base but adopt different post-training schemes and differ in the number of recurrent passes.75% similarUnverifiedSelf-supervised learning is the training foundation of the GPT series of models, which learn by predicting the next masked word in text.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/112419API
curl https://kongchang.com/api/v1/knowledge/claims/112419MCP
get_claim(id=112419)