Unverified50% confidenceBenchmarkExact time
The researchers conducted downstream evaluation on a GPT-2-scale model with 25 million parameters; the Functionalizer brought significant improvements in code syntactic validity and code character perplexity, while prose text coherence remained largely unchanged.
1
Sources
50%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
Unverified开发者将GPT-2（355M参数版本）微调后使其学会函数调用，正确且可解析的函数调用成功率达到88%73% similarUnverified同样使用GPT-4，经验丰富的开发者通过结构化提示获得的代码质量可以比随意提问高出40%以上72% similarUnverifiedCarlini等人2023年的研究显示，GPT-4等模型可以在特定提示下逐字复现训练集中的代码片段70% similarUnverifiedClaude series models excel in long-context understanding and code reasoning, while GPT series models may be faster at code completion for certain specific languages70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924771API
curl https://kongchang.com/api/v1/knowledge/claims/924771MCP
get_claim(id=924771)