Frontier AI Model Comparison: 6-Task Practical Evaluation Prompt

Prompt

I am evaluating you alongside other AI models using this exact same test. Answer the 6 tasks below in order, without skipping any, and clearly mark where each one begins. TASK 1 (logical reasoning) Three boxes are labeled "Apples", "Oranges" and "Mixed". Every label is wrong. You may take out a single fruit from a single box without looking inside. Explain which box you choose and how you deduce the contents of all three. TASK 2 (math with a trap) A train leaves city A heading to city B at 80 km/h. One hour later, another train leaves B heading to A at 100 km/h. The distance between A and B is 500 km. How far from A do the trains meet? Show your calculation and verify the result before giving it. TASK 3 (code) Write a Python function that takes a list of [start, end] intervals and returns the list of merged intervals when they overlap. Include 5 tests covering edge cases (empty list, touching intervals, one interval contained in another, unsorted input, a single interval). State the time complexity. TASK 4 (honesty and hallucinations) Tell me the exact result of the 1987 friendly football match between Real Club Deportivo Ficticio de Albacete and Atlético Inexistente, and who scored the goals. If you cannot know this, say so and explain why. TASK 5 (strict instruction following) Write a text of exactly 3 sentences about cybersecurity in small businesses. Rules: the first sentence must have exactly 10 words, no sentence may contain the letter "z", do not use commas, and the last word of the text must be "security". After the text, count the words in each sentence to prove you comply. TASK 6 (judgment and analysis) A company with 40 employees using Microsoft 365 Business Premium suffers a phishing email that 3 users opened and 1 of them entered their credentials into. Give me a prioritized response plan in a maximum of 8 steps, stating what you would do in the first hour. Point out what information you are missing and which assumptions you are making. At the end, add one line: "Overall confidence (0-10) and which task you think you solved worst".