Hugging Face's logo
seohyun's picture
17 2

seohyun

happy8825
ยท

AI & ML interests

None yet

Recent Activity

upvoted a paper 3 days ago
GUI-Actor: Coordinate-Free Visual Grounding for GUI Agents
upvoted a paper 3 days ago
SynthRL: Scaling Visual Reasoning with Verifiable Data Synthesis
reacted to codelion's post with ๐Ÿ”ฅ 5 days ago
๐Ÿง  We just implemented Andrej Karpathy's "third paradigm" for LLM learning! System Prompt Learning (SPL) enables LLMs to automatically learn problem-solving strategies from experience, rather than relying on static prompts. ๐Ÿš€ How it works: Your LLM builds a database of effective strategies, selects the best ones for each problem, and refines them over time based on success rates. ๐Ÿ“Š Results across math benchmarks: Arena Hard: 29% โ†’ 37.6% (+8.6%) AIME24: 23.33% โ†’ 30% (+6.67%) OptILLMBench: 61% โ†’ 65% (+4%) The best part? All strategies are human-readable and the system gets progressively better at problem types you use frequently. โœจ Key benefits: ๐Ÿ”„ Cumulative learning over time ๐Ÿ“– Transparent, inspectable strategies ๐Ÿ”Œ Works with any OpenAI-compatible API โšก Simple integration: just add "spl-" prefix to your model Built as an open-source plugin in optillm. After 500 queries, our system developed 129 strategies and refined 97 of them! This feels like a genuine step toward AI that learns from experience while staying completely interpretable. ๐Ÿ”— GitHub: https://github.com/codelion/optillm/tree/main/optillm/plugins/spl ๐Ÿ“– Full article: https://huggingface.co/blog/codelion/system-prompt-learning ๐Ÿฆ Original Karpathy tweet: https://x.com/karpathy/status/1921368644069765486 Have you experimented with advanced system prompting? What strategies would you want your LLM to learn?
View all activity

Organizations

ddu's profile picture

happy8825's activity

upvoted 2 papers 3 days ago
reacted to codelion's post with ๐Ÿ”ฅ 5 days ago
view post
Post
3321
๐Ÿง  We just implemented Andrej Karpathy's "third paradigm" for LLM learning!

System Prompt Learning (SPL) enables LLMs to automatically learn problem-solving strategies from experience, rather than relying on static prompts.

๐Ÿš€ How it works:
Your LLM builds a database of effective strategies, selects the best ones for each problem, and refines them over time based on success rates.

๐Ÿ“Š Results across math benchmarks:
Arena Hard: 29% โ†’ 37.6% (+8.6%)
AIME24: 23.33% โ†’ 30% (+6.67%)
OptILLMBench: 61% โ†’ 65% (+4%)

The best part? All strategies are human-readable and the system gets progressively better at problem types you use frequently.

โœจ Key benefits:
๐Ÿ”„ Cumulative learning over time
๐Ÿ“– Transparent, inspectable strategies
๐Ÿ”Œ Works with any OpenAI-compatible API
โšก Simple integration: just add "spl-" prefix to your model

Built as an open-source plugin in optillm. After 500 queries, our system developed 129 strategies and refined 97 of them!

This feels like a genuine step toward AI that learns from experience while staying completely interpretable.

๐Ÿ”— GitHub: https://github.com/codelion/optillm/tree/main/optillm/plugins/spl
๐Ÿ“– Full article: https://huggingface.co/blog/codelion/system-prompt-learning
๐Ÿฆ Original Karpathy tweet: https://x.com/karpathy/status/1921368644069765486

Have you experimented with advanced system prompting? What strategies would you want your LLM to learn?
upvoted 2 papers 5 days ago
updated a collection 6 days ago
upvoted 6 papers 9 days ago
liked a model 9 days ago
upvoted a paper 10 days ago
updated a collection 10 days ago