gpt-4o-2024-08-06· OpenAI
WildChat-4.8M (AI2)
IMPORTANT: You must ONLY return a valid JSON object in this EXACT format: {…
- User
System: IMPORTANT: You must ONLY return a valid JSON object in this EXACT format: { "needs_split": true/false, "split_positions": [] }
RULES:
- Return ONLY the JSON object, nothing else
- No explanations, no other text
- needs_split must be boolean
- split_positions must be array of integers
- split_positions should be empty if needs_split is false
- All positions must be within text length User: Analyze this text (line 9) and return ONLY a JSON with positions where paragraph breaks needed: Large language models (LLMs), such as ChatGPT, have received increasing attention in recent months due to their impressive performance on wide range of natural language processing tasks [ 16 ]. However, the robustness of these models against adversarial attacks and cross-domain data remains a challenge. While previous studies have evaluated the robustness of LLMs on other NLP tasks, such as text classification and sentiment analysis [ 19 ], their robustness on textto-SQL tasks has not been thoroughly investigated. Given the ever-increasing use of LLMs in real-world applications, it is essential to evaluate the robustness of these models on this task and understand their limitations to identify potential areas for improvement. Assistant:
- Assistant
{ "needs_split": true, "split_positions": [148, 370] }