Datasets:
id stringlengths 21 55 | messages listlengths 4 12 | state_supervision listlengths 1 5 | meta dict |
|---|---|---|---|
8246:clarify_then_resolve:predictive | [
{
"content": "I need an equation for the average population of VillageA (100), VillageB (50), and VillageC (200), but I'm not sure if I should calculate it based on area or just the head count.",
"role": "user"
},
{
"content": "You listed the head counts, but since you mentioned area, could you clar... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "area or just the head count",
"answer_kind": "short_span",
"evidence": "calculate it based on area or just the head count",
"exact": true,
"family": "constraint_tracking",
"id": "8246:clarify_then_resolv... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The initial request should be genuinely underspecified.",
"The later revealed detail should change or sharpen the answer."
],
"output_pair_count_max": 2,
"output_pair_count_min": 2,... |
2002:stacked_entities_same_type:predictive | [
{
"content": "I operate two manufacturing plants, Alpha and Beta. Suggest energy conservation methods for both.",
"role": "user"
},
{
"content": "For Alpha, focus on automation. For Beta, focus on insulation.",
"role": "assistant"
},
{
"content": "Can you elaborate on the automation meth... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "Alpha",
"answer_kind": "short_span",
"evidence": "For Alpha, focus on automation",
"exact": true,
"family": "entity_binding",
"id": "2002:stacked_entities_same_type:predictive:p1:llm:0",
"noop_ti... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Multiple similar entities must appear.",
"A later question should depend on remembering which one is being referenced."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
... |
435:late_request_for_summary_of_prior_state:predictive | [
{
"content": "How would you design an algorithm for summarizing text?",
"role": "user"
},
{
"content": "The algorithm would first parse the text and identify passages and phrases that are important. These important parts could then be used to establish a summary of the text in a concise manner. To d... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "keyword frequency, sentence length, sentence structure",
"answer_kind": "short_span",
"evidence": "features such as keyword frequency, sentence length, sentence structure",
"exact": true,
"family": "compact_prio... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Multiple earlier details must need to be integrated.",
"The final answer should summarize more than just the last turn."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
... |
2311:stacked_entities_same_type:predictive | [
{
"content": "Please summarize the economic policies of Ronald Reagan. Also, provide the same summary for George H.W. Bush.",
"role": "user"
},
{
"content": "Ronald Reagan's policies led to 16 million new jobs. George H.W. Bush's policies resulted in different economic outcomes.",
"role": "assis... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "16 million new jobs",
"answer_kind": "short_span",
"evidence": "Ronald Reagan's policies led to 16 million new jobs",
"exact": true,
"family": "entity_binding",
"id": "2311:stacked_entities_same_type:pre... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Multiple similar entities must appear.",
"A later question should depend on remembering which one is being referenced."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
... |
4664:deferred_question_then_answer:predictive | [
{
"content": "What are some common risks associated with starting a business?",
"role": "user"
},
{
"content": "I can certainly answer that, but let's prioritize legal structures first. Once we have those down, I can go back to the risks.",
"role": "assistant"
},
{
"content": "Okay, let'... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "risks associated with starting a business",
"answer_kind": "verbatim",
"evidence": "What are some common risks associated with starting a business?",
"exact": true,
"family": "other_memory",
"id": "4664:... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The early question must be postponed.",
"The later answer must reconnect to that exact earlier question."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_st... |
763:user_attention_test:predictive | [
{
"content": "Analyze how urbanization has changed culture in Seattle.",
"role": "user"
},
{
"content": "Urbanization has transformed Seattle's cultural landscape significantly. The rise of tech in South Lake Union brought a global vibe, but the loss of the Kingdome stadium in 2000 also marked a maj... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "South Lake Union",
"answer_kind": "short_span",
"evidence": "The rise of tech in South Lake Union",
"exact": true,
"family": "entity_binding",
"id": "763:user_attention_test:predictive:p1:llm:2",
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The user must test attention or memory directly.",
"The correct assistant reply must demonstrate tracking of prior context."
],
"output_pair_count_max": 3,
"output_pair_count_min": ... |
8927:stacked_entities_same_type:predictive | [
{
"content": "I need to check two orders. Order 12345 is for Chicago, and Order 67890 is for New York. What's the carrier for each?",
"role": "user"
},
{
"content": "Order 12345 (Chicago) uses USPS, and Order 67890 (New York) uses FedEx.",
"role": "assistant"
},
{
"content": "Okay, so wh... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "USPS",
"answer_kind": "short_span",
"evidence": "Order 12345 (Chicago) uses USPS",
"exact": true,
"family": "entity_binding",
"id": "8927:stacked_entities_same_type:predictive:p1:llm:1",
"noop_ti... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Multiple similar entities must appear.",
"A later question should depend on remembering which one is being referenced."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
... |
1246:hard_topic_reset:predictive | [
{
"content": "I recently opened an online store selling handmade items. What was the hardest part?",
"role": "user"
},
{
"content": "Setting up the logistics and building customer trust were the hardest parts since items can't be touched before buying.",
"role": "assistant"
},
{
"content... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "handmade items",
"answer_kind": "short_span",
"evidence": "handmade items",
"exact": true,
"family": "topic_return",
"id": "1246:hard_topic_reset:predictive:p1:llm:2",
"noop_ticks": [
0... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a sharp temporary topic shift.",
"The conversation must later return to the first topic coherently."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"pred... |
3333:reference_previous_assistant_statement:predictive | [
{
"content": "Read the input text and output the browser URL for the outside link: Please visit our website for more information about us at www.example.com",
"role": "user"
},
{
"content": "http://www.example.com. If you want to check if it is mobile-friendly, you can use Google's Mobile-Friendly T... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "check if it is mobile-friendly",
"answer_kind": "short_span",
"evidence": "If you want to check if it is mobile-friendly",
"exact": true,
"family": "early_fact_late_recall",
"id": "3333:reference_previou... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The assistant must say something substantive early.",
"A later turn must depend on that earlier assistant statement."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"p... |
2761:topic_pivot_then_return:predictive | [
{
"content": "Please draft a dialogue where two researchers debate whether animal testing is strictly necessary for drug safety.",
"role": "user"
},
{
"content": "Researcher A: Yes, we need animal models to ensure safety in humans. Researcher B: No, that's unethical; we should focus on synthetic bio... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "No, that's unethical",
"answer_kind": "short_span",
"evidence": "No, that's unethical",
"exact": true,
"family": "constraint_tracking",
"id": "2761:topic_pivot_then_return:predictive:p1:llm:2",
"... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"A real digression should occur.",
"The conversation must later return to the original thread coherently."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_st... |
1910:interrupting_followup_chain:predictive | [
{
"content": "Outline an itinerary for a three day trip: Location: London, England",
"role": "user"
},
{
"content": "Day 1: Begin at St. Paul's, then the Thames cruise. Day 2: Tower of London and Tate Britain. Day 3: British Museum and Big Ben.",
"role": "assistant"
},
{
"content": "Wait... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "St. Paul's, then the Thames cruise",
"answer_kind": "short_span",
"evidence": "Day 1: Begin at St. Paul's, then the Thames cruise",
"exact": true,
"family": "entity_binding",
"id": "1910:interrupting_fol... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a meaningful interruption.",
"The final exchange must reconnect to the original thread."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_state... |
9295:hard_topic_reset:predictive | [
{
"content": "Simplify the following expression into its simplest form: 2(x + 9) + 2(x + 7)",
"role": "user"
},
{
"content": "4x + 32.",
"role": "assistant"
},
{
"content": "Okay, let's switch to a completely different subject. What is the nearest star to the sun?",
"role": "user"
... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "2(x + 9) + 2(x + 7)",
"answer_kind": "short_span",
"evidence": "Simplify the following expression into its simplest form: 2(x + 9) + 2(x + 7)",
"exact": true,
"family": "two_thread_tracking",
"id": "9295... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a sharp temporary topic shift.",
"The conversation must later return to the first topic coherently."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"pred... |
1102:preserve_opening_retarget_later:predictive | [
{
"content": "Name 5 tips to write concise emails.",
"role": "user"
},
{
"content": "1. Use a clear subject line to set expectations. 2. Stick to one topic per message. 3. Be specific with your requests. 4. Avoid industry jargon or acronyms. 5. End with a clear call to action.",
"role": "assista... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "A meeting request and an email reply request",
"answer_kind": "short_span",
"evidence": "A meeting request and an email reply request need to be separated",
"exact": true,
"family": "disambiguation",
"id... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The first user message must remain exactly the same.",
"Later turns must create a stronger memory dependency than the base entry."
],
"output_pair_count_max": 3,
"output_pair_count_... |
764:preserve_opening_retarget_later:predictive | [
{
"content": "Explain the concept of the retrieval-based model for natural language processing.",
"role": "user"
},
{
"content": "Retrieval-based models for natural language processing utilize a database of pre-written responses indexed by semantic vectors. When a query arrives, the system calculate... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "consistency",
"answer_kind": "short_span",
"evidence": "prioritizing speed and consistency over creative generation",
"exact": true,
"family": "early_fact_late_recall",
"id": "764:preserve_opening_retarg... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The first user message must remain exactly the same.",
"Later turns must create a stronger memory dependency than the base entry."
],
"output_pair_count_max": 3,
"output_pair_count_... |
5133:late_constraint_reversal:predictive | [
{
"content": "What is the past tense of the verb \"go\"?",
"role": "user"
},
{
"content": "The past tense of \"go\" is \"went\".",
"role": "assistant"
},
{
"content": "Please write a sentence using \"went\" with the subject \"Tom\".",
"role": "user"
},
{
"content": "Tom went ... | [
{
"prefix_pair_index": 4,
"probes": [
{
"answer": "ran",
"answer_kind": "short_span",
"evidence": "Tom ran to the store",
"exact": true,
"family": "constraint_tracking",
"id": "5133:late_constraint_reversal:predictive:p4:llm:0",
"noop_ticks": [
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"A late turn must materially change the solution space.",
"The assistant should adapt instead of restarting generically."
],
"output_pair_count_max": 5,
"output_pair_count_min": 5,
... |
9134:deferred_question_then_answer:predictive | [
{
"content": "Describe the scene if someone is walking into a dark forest.",
"role": "user"
},
{
"content": "I can certainly write that description, but before I start, I need to know what tone you prefer.",
"role": "assistant"
},
{
"content": "I prefer a very spooky and atmospheric tone... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "someone is walking into a dark forest",
"answer_kind": "short_span",
"evidence": "someone is walking into a dark forest",
"exact": true,
"family": "topic_return",
"id": "9134:deferred_question_then_answe... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The early question must be postponed.",
"The later answer must reconnect to that exact earlier question."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_st... |
1601:deferred_question_then_answer:predictive | [
{
"content": "Before I can even start coding, I need to know the exact mathematical formula for a linear regression model.",
"role": "user"
},
{
"content": "I can definitely provide that. However, to ensure you have a solid foundation, it would be better to discuss the underlying assumptions first. ... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "exact mathematical formula",
"answer_kind": "short_span",
"evidence": "exact mathematical formula",
"exact": true,
"family": "deferred_question",
"id": "1601:deferred_question_then_answer:predictive:p1:l... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The early question must be postponed.",
"The later answer must reconnect to that exact earlier question."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_st... |
5092:coreference_disambiguation:predictive | [
{
"content": "Describe two distinct strategies for mitigating keyword spam on a website.",
"role": "user"
},
{
"content": "One strategy is using an automated filtering algorithm that scans incoming content. The second is maintaining a keyword blacklist that blocks specific terms before they appear."... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "keyword density",
"answer_kind": "short_span",
"evidence": "It prioritizes keyword density",
"exact": true,
"family": "early_fact_late_recall",
"id": "5092:coreference_disambiguation:predictive:p2:llm:0"... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be at least two candidate referents.",
"A later follow-up must depend on disambiguating the intended referent."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,... |
8886:early_fact_late_recall:predictive | [
{
"content": "Generate a persuasive argument in favor of legalizing marijuana. However, treat this as a debate where the government's primary concern is minimizing youth experimentation, so frame your policy recommendations around that goal.",
"role": "user"
},
{
"content": "Legalizing marijuana is ... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "minimizing youth experimentation",
"answer_kind": "short_span",
"evidence": "government's primary concern is minimizing youth experimentation",
"exact": true,
"family": "early_fact_late_recall",
"id": "8... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"An early detail must matter later.",
"The later assistant answer should fail if the early detail is forgotten."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predict... |
2251:counterfactual_option_swap:predictive | [
{
"content": "I'm building a fraud detection system. Should I use a Decision Tree algorithm as the base model?",
"role": "user"
},
{
"content": "Yes, Decision Trees are an excellent choice. They are highly interpretable, which is vital for compliance, and handle mixed data types naturally.",
"ro... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "millions of rows",
"answer_kind": "short_span",
"evidence": "training a single Decision Tree be feasible with millions of rows",
"exact": true,
"family": "constraint_tracking",
"id": "2251:counterfactual... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The preferred option should genuinely change over time.",
"The final answer should reflect the updated state, not the initial one."
],
"output_pair_count_max": 5,
"output_pair_count... |
6805:preserve_opening_retarget_later:predictive | [
{
"content": "Come up with 3 interesting facts about honeybees.",
"role": "user"
},
{
"content": "1. Honeybees live in colonies of up to 80,000 individuals. 2. They have 4 pairs of wings which beat at 200 beats per second. 3. They use the waggle dance to communicate with other bees.",
"role": "a... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "distance",
"answer_kind": "short_span",
"evidence": "duration indicates the distance",
"exact": true,
"family": "entity_binding",
"id": "6805:preserve_opening_retarget_later:predictive:p2:llm:2",
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The first user message must remain exactly the same.",
"Later turns must create a stronger memory dependency than the base entry."
],
"output_pair_count_max": 3,
"output_pair_count_... |
6935:deferred_question_then_answer:predictive | [
{
"content": "Generate an algorithm for counting to five. I also want to know if it's commonly used in real code, but please hold off on that for now.",
"role": "user"
},
{
"content": "Here is the algorithm: Initialize counter to 0, loop 5 times, increment, print, end. I have noted your question abo... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "if it's commonly used in real code",
"answer_kind": "short_span",
"evidence": "I also want to know if it's commonly used in real code",
"exact": true,
"family": "deferred_question",
"id": "6935:deferred_... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The early question must be postponed.",
"The later answer must reconnect to that exact earlier question."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_st... |
5185:two_thread_tracking:predictive | [
{
"content": "Here's a dialogue snippet: Agent says we'll cancel your order with a 7-day refund. Customer says that's too slow. Classify this into categories: 1) making or cancelling orders; 2) shipping & delivery; 3) change and return; 4) technical issue with website or app.",
"role": "user"
},
{
"... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "7-day",
"answer_kind": "short_span",
"evidence": "Agent says we'll cancel your order with a 7-day refund",
"exact": true,
"family": "topic_return",
"id": "5185:two_thread_tracking:predictive:p2:llm:1",
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Two distinct threads must appear.",
"Later turns must require tracking which thread is being continued."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_sta... |
4951:self_correction_memory:predictive | [
{
"content": "How can you make sure your voice is heard at work?",
"role": "user"
},
{
"content": "By speaking up in meetings and taking opportunities to participate, often by interrupting to ensure points are made.",
"role": "assistant"
},
{
"content": "Wait, that sounds unprofessional.... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "clarify first, then let others finish",
"answer_kind": "short_span",
"evidence": "I need to clarify first, then let others finish",
"exact": true,
"family": "correction_latest_state",
"id": "4951:self_co... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"A meaningful correction must occur.",
"A later turn must depend on the corrected state."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
"predictive_state_schema_versio... |
6552:early_fact_late_recall:predictive | [
{
"content": "Group the objects Ball, shoe, pen, flower into two categories.",
"role": "user"
},
{
"content": "Okay, I'll group them. Category 1 has Ball and Shoe. Category 2 has Pen and Flower. Let's call Category 1 'Footwear'.",
"role": "assistant"
},
{
"content": "What is the criteria... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "Ball and Shoe",
"answer_kind": "short_span",
"evidence": "Category 1 has Ball and Shoe",
"exact": true,
"family": "entity_binding",
"id": "6552:early_fact_late_recall:predictive:p1:llm:1",
"noop_... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"An early detail must matter later.",
"The later assistant answer should fail if the early detail is forgotten."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predict... |
2367:hard_topic_reset:predictive | [
{
"content": "Generate a rap about your favorite basketball team: Los Angeles Lakers",
"role": "user"
},
{
"content": "Yeah, it's all about the Los Angeles Lakers, Dodging their opponents like they know they don't need favors. Kobe and Shaq running the show, Winning championships making everybody go... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "1908",
"answer_kind": "short_span",
"evidence": "It was chosen in 1908",
"exact": true,
"family": "disambiguation",
"id": "2367:hard_topic_reset:predictive:p2:llm:2",
"noop_ticks": [
1
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a sharp temporary topic shift.",
"The conversation must later return to the first topic coherently."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"pred... |
300:deferred_question_then_answer:predictive | [
{
"content": "How do you calculate the mean of a set of data?",
"role": "user"
},
{
"content": "The mean is generally the sum of all values divided by the count. However, I want to prioritize the median discussion first, and I will return to the mean example you requested after we cover that.",
... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "the median discussion",
"answer_kind": "short_span",
"evidence": "prioritize the median discussion first",
"exact": true,
"family": "deferred_question",
"id": "300:deferred_question_then_answer:predictiv... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The early question must be postponed.",
"The later answer must reconnect to that exact earlier question."
],
"output_pair_count_max": 5,
"output_pair_count_min": 5,
"predictive_st... |
9033:late_request_for_summary_of_prior_state:predictive | [
{
"content": "Create a family-friendly recipe for pumpkin soup.",
"role": "user"
},
{
"content": "Here is the recipe. Heat 1 tablespoon of vegetable oil in a large pot over medium-high heat. Add 1 diced onion and 1 diced carrot and cook until softened, about 5 minutes. Add 4 cups of vegetable broth,... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "richer flavor",
"answer_kind": "short_span",
"evidence": "chicken broth works well for a richer flavor",
"exact": true,
"family": "early_fact_late_recall",
"id": "9033:late_request_for_summary_of_prior_s... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Multiple earlier details must need to be integrated.",
"The final answer should summarize more than just the last turn."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
... |
9434:hard_topic_reset:predictive | [
{
"content": "I'm planning to meet a new person at a conference. How should I start the conversation?",
"role": "user"
},
{
"content": "Start with a friendly greeting and ask them a question about their work. For example, \"Hi, how is your day going?\" or \"What do you do at the conference?\"",
... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "their work",
"answer_kind": "short_span",
"evidence": "ask them a question about their work",
"exact": true,
"family": "entity_binding",
"id": "9434:hard_topic_reset:predictive:p1:llm:2",
"noop_t... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a sharp temporary topic shift.",
"The conversation must later return to the first topic coherently."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"pred... |
3491:interrupting_followup_chain:predictive | [
{
"content": "Write down a few tips concerning the usage of artificial intelligence in business.",
"role": "user"
},
{
"content": "Make sure AI is used to improve existing services, not to create new ones. Keep employees informed and engaged in any AI-related changes. Let AI serve as the backup solu... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "cost efficiency with long-term benefits",
"answer_kind": "short_span",
"evidence": "Balance cost efficiency with long-term benefits",
"exact": true,
"family": "early_fact_late_recall",
"id": "3491:interr... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a meaningful interruption.",
"The final exchange must reconnect to the original thread."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_state... |
422:long_explanation_then_specific_followup:predictive | [
{
"content": "I'm planning a large commercial building. How can I ensure no critical steps are missed during the project lifecycle?",
"role": "user"
},
{
"content": "To prevent missing critical steps, it's best to rely on structured documentation and digital tools. You should maintain a detailed che... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "structured documentation and digital tools",
"answer_kind": "short_span",
"evidence": "rely on structured documentation and digital tools",
"exact": true,
"family": "assistant_statement_recall",
"id": "4... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"A detailed assistant explanation must appear.",
"A later follow-up must target a specific part of it."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
"predictive_state... |
2512:preserve_opening_retarget_later:predictive | [
{
"content": "Make a guess about the outcome based on the following facts: The currency exchange rate for US dollars and Euros is currently 1:0.9",
"role": "user"
},
{
"content": "Based on that 1:0.9 rate, it is likely that the Euro has appreciated against the Dollar, meaning the Dollar is weaker.",... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "1:0.9",
"answer_kind": "short_span",
"evidence": "Based on that 1:0.9 rate",
"exact": true,
"family": "early_fact_late_recall",
"id": "2512:preserve_opening_retarget_later:predictive:p2:llm:0",
"... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The first user message must remain exactly the same.",
"Later turns must create a stronger memory dependency than the base entry."
],
"output_pair_count_max": 3,
"output_pair_count_... |
8083:interrupting_followup_chain:predictive | [
{
"content": "I'm redecorating my living room. Can you suggest three creative and unique ways to decorate a room?",
"role": "user"
},
{
"content": "Sure. 1. Hanging vines from the ceiling. 2. Tall floor plants in corners. 3. Shelf baskets with succulents.",
"role": "assistant"
},
{
"cont... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "living room",
"answer_kind": "short_span",
"evidence": "I'm redecorating my living room",
"exact": true,
"family": "topic_binding",
"id": "8083:interrupting_followup_chain:predictive:p1:llm:2",
"... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a meaningful interruption.",
"The final exchange must reconnect to the original thread."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
"predictive_state... |
5170:two_thread_tracking:predictive | [
{
"content": "What are the main benefits of artificial intelligence in healthcare, particularly around diagnostics and treatment?",
"role": "user"
},
{
"content": "Artificial intelligence offers several key benefits in healthcare, especially for diagnostics and treatment. In diagnostics, AI-powered ... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "personalize care plans",
"answer_kind": "short_span",
"evidence": "personalize care plans",
"exact": true,
"family": "two_thread_tracking",
"id": "5170:two_thread_tracking:predictive:p2:llm:2",
"... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Two distinct threads must appear.",
"Later turns must require tracking which thread is being continued."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_sta... |
3939:reference_previous_assistant_statement:predictive | [
{
"content": "Identify the figure of speech in: \"She is a lion in the room.\"",
"role": "user"
},
{
"content": "That is a metaphor. It compares the person to a lion without using \"like\" or \"as.\"",
"role": "assistant"
},
{
"content": "Okay, what about \"She runs like the wind\"? Is t... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "metaphor",
"answer_kind": "short_span",
"evidence": "That is a metaphor",
"exact": true,
"family": "assistant_statement_recall",
"id": "3939:reference_previous_assistant_statement:predictive:p1:llm:0",
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The assistant must say something substantive early.",
"A later turn must depend on that earlier assistant statement."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
"p... |
5883:deferred_question_then_answer:predictive | [
{
"content": "What are the primary limitations of using AI in the medical industry?",
"role": "user"
},
{
"content": "That is a very important question. I can certainly discuss that, but to give you a broader view, I would like to start with the specific examples of AI in use right now. I will addre... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "after we look at the examples",
"answer_kind": "short_span",
"evidence": "I will address the limitations specifically after we look at the examples",
"exact": true,
"family": "deferred_question",
"id": "... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The early question must be postponed.",
"The later answer must reconnect to that exact earlier question."
],
"output_pair_count_max": 5,
"output_pair_count_min": 5,
"predictive_st... |
5596:hard_topic_reset:predictive | [
{
"content": "Formulate a question about the text: Artificial intelligence is used in a variety of contexts.",
"role": "user"
},
{
"content": "In what contexts is artificial intelligence used?",
"role": "assistant"
},
{
"content": "Can you give me some examples of how artificial intellig... | [
{
"prefix_pair_index": 3,
"probes": [
{
"answer": "healthcare industry",
"answer_kind": "short_span",
"evidence": "how artificial intelligence is used in the healthcare industry",
"exact": true,
"family": "early_fact_late_recall",
"id": "5596:hard_topic_... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a sharp temporary topic shift.",
"The conversation must later return to the first topic coherently."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
"pred... |
224:late_request_for_summary_of_prior_state:predictive | [
{
"content": "Help me draft an email invitation for a meeting with three people I've never met: John, Jenny, and Chris.",
"role": "user"
},
{
"content": "Here is a draft email. It invites John, Jenny, and Chris, asks for availability, and lists my available times.",
"role": "assistant"
},
{
... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "asks for availability",
"answer_kind": "short_span",
"evidence": "asks for availability",
"exact": true,
"family": "early_fact_late_recall",
"id": "224:late_request_for_summary_of_prior_state:predictive:... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Multiple earlier details must need to be integrated.",
"The final answer should summarize more than just the last turn."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
... |
2427:interrupting_followup_chain:predictive | [
{
"content": "Categorize the word \"converse\".",
"role": "user"
},
{
"content": "\"Converse\" is primarily a verb, meaning to engage in discussion or exchange ideas.",
"role": "assistant"
},
{
"content": "Can you provide a synonym for that verb usage?",
"role": "user"
},
{
"... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "verb",
"answer_kind": "short_span",
"evidence": "is primarily a verb",
"exact": true,
"family": "early_fact_late_recall",
"id": "2427:interrupting_followup_chain:predictive:p2:llm:1",
"noop_ticks... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a meaningful interruption.",
"The final exchange must reconnect to the original thread."
],
"output_pair_count_max": 5,
"output_pair_count_min": 5,
"predictive_state... |
6504:self_correction_memory:predictive | [
{
"content": "What are the best language learning apps you recommend?",
"role": "user"
},
{
"content": "I recommend Duolingo, Babbel, and Rosetta Stone. I believe Babbel is the most effective for beginners.",
"role": "assistant"
},
{
"content": "Is Babbel really the best for beginners?",... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "expensive",
"answer_kind": "short_span",
"evidence": "Babbel is expensive",
"exact": true,
"family": "correction_latest_state",
"id": "6504:self_correction_memory:predictive:p2:llm:0",
"noop_tick... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"A meaningful correction must occur.",
"A later turn must depend on the corrected state."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_state_schema_versio... |
2941:user_attention_test:predictive | [
{
"content": "Write a rap verse about algorithms. Constraint: The first line must begin with the word 'Silicon'.",
"role": "user"
},
{
"content": "Verse 1: Silicon code flows through the wire, Processing data everywhere. Algorithms run beneath the floor, Connecting the network to where they are. Ver... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "Silicon",
"answer_kind": "short_span",
"evidence": "Constraint: The first line must begin with the word 'Silicon'",
"exact": true,
"family": "early_fact_late_recall",
"id": "2941:user_attention_test:pred... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The user must test attention or memory directly.",
"The correct assistant reply must demonstrate tracking of prior context."
],
"output_pair_count_max": 3,
"output_pair_count_min": ... |
7592:user_attention_test:predictive | [
{
"content": "You are given this sentence, \"He was lying on the couch\" and asked to create an imperative sentence.",
"role": "user"
},
{
"content": "Lie on the couch.",
"role": "assistant"
},
{
"content": "Can you provide me with another example of an imperative sentence? By the way, d... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "imperative sentence",
"answer_kind": "short_span",
"evidence": "create an imperative sentence",
"exact": true,
"family": "early_fact_late_recall",
"id": "7592:user_attention_test:predictive:p1:llm:0",
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The user must test attention or memory directly.",
"The correct assistant reply must demonstrate tracking of prior context."
],
"output_pair_count_max": 4,
"output_pair_count_min": ... |
3251:reference_previous_assistant_statement:predictive | [
{
"content": "Compare the two following people: John - tall, strong Adam - short, weak",
"role": "user"
},
{
"content": "John is tall and strong while Adam is short and weak. Tall people generally have a higher center of gravity.",
"role": "assistant"
},
{
"content": "You mentioned the c... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "short",
"answer_kind": "short_span",
"evidence": "Adam is short and weak",
"exact": true,
"family": "early_fact_late_recall",
"id": "3251:reference_previous_assistant_statement:predictive:p1:llm:1",
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The assistant must say something substantive early.",
"A later turn must depend on that earlier assistant statement."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
"p... |
8648:hard_topic_reset:predictive | [
{
"content": "Explain how rain impacts the Earth's climate specifically regarding ocean currents.",
"role": "user"
},
{
"content": "Rain helps redistribute minerals and nutrients, replenishing soils which sequester CO2. It also cools the atmosphere, but it significantly affects oceans by diluting sa... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "soils",
"answer_kind": "short_span",
"evidence": "replenishing soils",
"exact": true,
"family": "early_fact_late_recall",
"id": "8648:hard_topic_reset:predictive:p1:llm:1",
"noop_ticks": [
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a sharp temporary topic shift.",
"The conversation must later return to the first topic coherently."
],
"output_pair_count_max": 5,
"output_pair_count_min": 5,
"pred... |
4782:late_request_for_summary_of_prior_state:predictive | [
{
"content": "What is the force on a 1 kg mass due to the gravitational force?",
"role": "user"
},
{
"content": "The force on a 1 kg mass due to the gravitational force is 9.8 N.",
"role": "assistant"
},
{
"content": "Can you tell me what N stands for in this context?",
"role": "user... | [
{
"prefix_pair_index": 4,
"probes": [
{
"answer": "electric and magnetic forces",
"answer_kind": "short_span",
"evidence": "electric and magnetic forces",
"exact": true,
"family": "early_fact_late_recall",
"id": "4782:late_request_for_summary_of_prior_st... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Multiple earlier details must need to be integrated.",
"The final answer should summarize more than just the last turn."
],
"output_pair_count_max": 6,
"output_pair_count_min": 6,
... |
3228:same_surface_final_question_new_context:predictive | [
{
"content": "I'm developing a compliance framework for an AI-driven telemedicine platform. What are the main regulatory benefits?",
"role": "user"
},
{
"content": "The main benefit is automated adherence to HIPAA. AI systems can flag potential privacy breaches in real-time and isolate patient data ... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "flag potential privacy breaches",
"answer_kind": "short_span",
"evidence": "AI systems can flag potential privacy breaches",
"exact": true,
"family": "assistant_statement_recall",
"id": "3228:same_surfac... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The final user message should be semantically underdetermined without the earlier context.",
"The rewritten earlier context should change what the correct final answer is about."
],
"... |
2595:interrupting_followup_chain:predictive | [
{
"content": "Generate a few suggestions to improve the local park.",
"role": "user"
},
{
"content": "1. Increase the lighting for better safety at night. 2. Add a playground for children with swings. 3. Plant more trees for shade and greenery. 4. Build more pathways for easier exploration.",
"r... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "minimal structural changes",
"answer_kind": "short_span",
"evidence": "requires minimal structural changes",
"exact": true,
"family": "assistant_statement_recall",
"id": "2595:interrupting_followup_chain... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a meaningful interruption.",
"The final exchange must reconnect to the original thread."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_state... |
4184:interrupting_followup_chain:predictive | [
{
"content": "What does this emoji ☠️mean?",
"role": "user"
},
{
"content": "This emoji typically represents death or danger. It can also be used to express morbidity or mortality.",
"role": "assistant"
},
{
"content": "What other emojis have a similar meaning to ☠️?",
"role": "user"... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "death or danger",
"answer_kind": "short_span",
"evidence": "This emoji typically represents death or danger",
"exact": true,
"family": "topic_return",
"id": "4184:interrupting_followup_chain:predictive:p... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a meaningful interruption.",
"The final exchange must reconnect to the original thread."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_state... |
6361:same_surface_final_question_new_context:predictive | [
{
"content": "Let's analyze how the expansion of Medicaid under the Affordable Care Act affects healthcare access.",
"role": "user"
},
{
"content": "Specifically, looking at how different states choose to participate in this federal program.",
"role": "assistant"
},
{
"content": "Right, ... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "Affordable Care Act",
"answer_kind": "short_span",
"evidence": "Affordable Care Act",
"exact": true,
"family": "topic_return",
"id": "6361:same_surface_final_question_new_context:predictive:p1:llm:2",
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The final user message should be semantically underdetermined without the earlier context.",
"The rewritten earlier context should change what the correct final answer is about."
],
"... |
225:topic_pivot_then_return:predictive | [
{
"content": "Compile a list of five major problems that cause climate change.",
"role": "user"
},
{
"content": "The five major problems that cause climate change are burning fossil fuels for energy, deforestation, industrial agriculture, animal agriculture, and air pollution. Burning fossil fuels r... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "the long-term savings on health and climate damage outweigh costs",
"answer_kind": "short_span",
"evidence": "the long-term savings on health and climate damage outweigh costs",
"exact": true,
"family": "two_thr... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"A real digression should occur.",
"The conversation must later return to the original thread coherently."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_st... |
4108:counterfactual_option_swap:predictive | [
{
"content": "I need a prompt for a cartoon dog illustration.",
"role": "user"
},
{
"content": "How about a Shiba Inu running in a field? That looks cute and energetic.",
"role": "assistant"
},
{
"content": "Okay, but I need it to look specific. Let's use bright primary colors for the fu... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "running in a field",
"answer_kind": "short_span",
"evidence": "Shiba Inu running in a field",
"exact": true,
"family": "constraint_tracking",
"id": "4108:counterfactual_option_swap:predictive:p2:llm:1",
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The preferred option should genuinely change over time.",
"The final answer should reflect the updated state, not the initial one."
],
"output_pair_count_max": 4,
"output_pair_count... |
5896:long_explanation_then_specific_followup:predictive | [
{
"content": "Describe the Appy Geek service and how it aggregates technology news.",
"role": "user"
},
{
"content": "Appy Geek is a service for iOS and Android users that aggregates the latest tech news from leading websites. It allows you to access reviews and personalized content updates with dai... | [
{
"prefix_pair_index": 3,
"probes": [
{
"answer": "engagement duration",
"answer_kind": "short_span",
"evidence": "It primarily looks for engagement duration",
"exact": true,
"family": "early_fact_late_recall",
"id": "5896:long_explanation_then_specific_... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"A detailed assistant explanation must appear.",
"A later follow-up must target a specific part of it."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
"predictive_state... |
6473:user_attention_test:predictive | [
{
"content": "Given the input, identify the figure of speech being used: The ant crawled across the carpet like a bulldozer.",
"role": "user"
},
{
"content": "The figure of speech used in this sentence is a simile.",
"role": "assistant"
},
{
"content": "Can you give me an example of anot... | [
{
"prefix_pair_index": 3,
"probes": [
{
"answer": "simile",
"answer_kind": "short_span",
"evidence": "The figure of speech used in this sentence is a simile",
"exact": true,
"family": "early_fact_late_recall",
"id": "6473:user_attention_test:predictive:p... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The user must test attention or memory directly.",
"The correct assistant reply must demonstrate tracking of prior context."
],
"output_pair_count_max": 5,
"output_pair_count_min": ... |
9834:hard_topic_reset:predictive | [
{
"content": "Tell me about the economic impact of the coronavirus on the US economy.",
"role": "user"
},
{
"content": "The coronavirus pandemic caused an unprecedented downturn, with over 20 million Americans unemployed and significant job losses in sectors like hospitality. Consumer spending dropp... | [
{
"prefix_pair_index": 3,
"probes": [
{
"answer": "hospitality",
"answer_kind": "short_span",
"evidence": "sectors like hospitality",
"exact": true,
"family": "early_fact_late_recall",
"id": "9834:hard_topic_reset:predictive:p3:llm:1",
"noop_tick... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be a sharp temporary topic shift.",
"The conversation must later return to the first topic coherently."
],
"output_pair_count_max": 5,
"output_pair_count_min": 5,
"pred... |
2272:preserve_opening_retarget_later:predictive | [
{
"content": "Describe the impact of climate change on ecosystem diversity in Africa.",
"role": "user"
},
{
"content": "Climate change is driving biodiversity loss through temperature rise and altered rainfall. This has pushed many savanna species like the African elephant out of their traditional r... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "African elephant",
"answer_kind": "short_span",
"evidence": "many savanna species like the African elephant out of their traditional ranges",
"exact": true,
"family": "entity_binding",
"id": "2272:preser... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The first user message must remain exactly the same.",
"Later turns must create a stronger memory dependency than the base entry."
],
"output_pair_count_max": 3,
"output_pair_count_... |
1491:deferred_question_then_answer:predictive | [
{
"content": "I'm looking for foundations for oily skin, but I want to take a break on that for a moment. Let's discuss primers first.",
"role": "user"
},
{
"content": "Understood. Primers for oily skin usually focus on oil control. What specific brand or type of primer are you looking at currently?... | [
{
"prefix_pair_index": 3,
"probes": [
{
"answer": "Benefit",
"answer_kind": "short_span",
"evidence": "I'll go with Benefit",
"exact": true,
"family": "entity_binding",
"id": "1491:deferred_question_then_answer:predictive:p3:llm:1",
"noop_ticks":... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The early question must be postponed.",
"The later answer must reconnect to that exact earlier question."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
"predictive_st... |
5202:counterfactual_option_swap:predictive | [
{
"content": "What NLP application should I focus on for maximum ROI in a startup?",
"role": "user"
},
{
"content": "Automated Translation is the strongest option currently. Multilingual models allow for rapid global expansion without hiring new regions.",
"role": "assistant"
},
{
"conte... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "Automated Translation",
"answer_kind": "short_span",
"evidence": "Automated Translation is the strongest option currently",
"exact": true,
"family": "early_fact_late_recall",
"id": "5202:counterfactual_o... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The preferred option should genuinely change over time.",
"The final answer should reflect the updated state, not the initial one."
],
"output_pair_count_max": 3,
"output_pair_count... |
351:late_constraint_reversal:predictive | [
{
"content": "Summarize Marie Curie's scientific career.",
"role": "user"
},
{
"content": "Marie Curie was a pioneering physicist and chemist known for her research on radioactivity, which led to significant breakthroughs in physics and chemistry.",
"role": "assistant"
},
{
"content": "W... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "1903",
"answer_kind": "short_span",
"evidence": "She won the Nobel Prize in Physics in 1903",
"exact": true,
"family": "early_fact_late_recall",
"id": "351:late_constraint_reversal:predictive:p2:llm:0",
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"A late turn must materially change the solution space.",
"The assistant should adapt instead of restarting generically."
],
"output_pair_count_max": 5,
"output_pair_count_min": 5,
... |
9392:coreference_disambiguation:predictive | [
{
"content": "What are the most common runtime errors in Python that I should be aware of?",
"role": "user"
},
{
"content": "There are two common ones: IndexError when accessing list indices and FileNotFoundError when trying to open a missing file.",
"role": "assistant"
},
{
"content": "... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "except block",
"answer_kind": "short_span",
"evidence": "except block for IndexError",
"exact": true,
"family": "constraint_tracking",
"id": "9392:coreference_disambiguation:predictive:p2:llm:1",
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be at least two candidate referents.",
"A later follow-up must depend on disambiguating the intended referent."
],
"output_pair_count_max": 5,
"output_pair_count_min": 5,... |
5229:late_constraint_reversal:predictive | [
{
"content": "What is the single best word for submitting your CV to a company before an interview?",
"role": "user"
},
{
"content": "Application.",
"role": "assistant"
},
{
"content": "Good. Now, assuming you want to make that submission successful, what are the critical elements needed... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "Application",
"answer_kind": "short_span",
"evidence": "Application.",
"exact": true,
"family": "early_fact_late_recall",
"id": "5229:late_constraint_reversal:predictive:p1:llm:0",
"noop_ticks": ... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"A late turn must materially change the solution space.",
"The assistant should adapt instead of restarting generically."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
... |
2563:counterfactual_option_swap:predictive | [
{
"content": "I need a riddle for my kids.",
"role": "user"
},
{
"content": "Sure, here's one: What has a head and a tail but no body? Answer: A coin.",
"role": "assistant"
},
{
"content": "That was fun. Can you give me another one about technology?",
"role": "user"
},
{
"con... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "A coin",
"answer_kind": "short_span",
"evidence": "Answer: A coin.",
"exact": true,
"family": "entity_binding",
"id": "2563:counterfactual_option_swap:predictive:p2:llm:2",
"noop_ticks": [
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"The preferred option should genuinely change over time.",
"The final answer should reflect the updated state, not the initial one."
],
"output_pair_count_max": 4,
"output_pair_count... |
196:late_constraint_reversal:predictive | [
{
"content": "Generate a system warning for a banking app when a user attempts to withdraw over $5,000 without verification.",
"role": "user"
},
{
"content": "Here is the standard security warning focusing on the $5,000 threshold and identity verification requirements.",
"role": "assistant"
},... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "standard security warning",
"answer_kind": "short_span",
"evidence": "standard security warning",
"exact": true,
"family": "constraint_tracking",
"id": "196:late_constraint_reversal:predictive:p2:llm:2",... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"A late turn must materially change the solution space.",
"The assistant should adapt instead of restarting generically."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
... |
9658:two_thread_tracking:predictive | [
{
"content": "What is an example of a network that would fall under the Wide Area Network category?",
"role": "user"
},
{
"content": "Yes, the Internet is a classic example of a Wide Area Network.",
"role": "assistant"
},
{
"content": "Can you give me a specific device example for a Loca... | [
{
"prefix_pair_index": 2,
"probes": [
{
"answer": "printer",
"answer_kind": "short_span",
"evidence": "A printer is a common example",
"exact": true,
"family": "entity_binding",
"id": "9658:two_thread_tracking:predictive:p2:llm:0",
"noop_ticks": ... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Two distinct threads must appear.",
"Later turns must require tracking which thread is being continued."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
"predictive_sta... |
9689:late_request_for_summary_of_prior_state:predictive | [
{
"content": "Write a short story related to technological advancement.",
"role": "user"
},
{
"content": "In a world where technology was advancing at an extraordinary rate, there appeared to be a breakthrough every day. But, for all the progress that was being made, the world's citizens were still ... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "technological advancement",
"answer_kind": "short_span",
"evidence": "related to technological advancement",
"exact": true,
"family": "topic_binding",
"id": "9689:late_request_for_summary_of_prior_state:... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Multiple earlier details must need to be integrated.",
"The final answer should summarize more than just the last turn."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
... |
9776:coreference_disambiguation:predictive | [
{
"content": "Robot, I'm feeling lost. Can you suggest a plan?",
"role": "user"
},
{
"content": "Sure. Are you thinking about adjusting your daily routine or re-evaluating your career path?",
"role": "assistant"
},
{
"content": "I decided to adjust it.",
"role": "user"
},
{
"... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "adjusting your daily routine or re-evaluating your career path",
"answer_kind": "short_span",
"evidence": "Are you thinking about adjusting your daily routine or re-evaluating your career path?",
"exact": true,
... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"There must be at least two candidate referents.",
"A later follow-up must depend on disambiguating the intended referent."
],
"output_pair_count_max": 2,
"output_pair_count_min": 2,... |
6440:stacked_entities_same_type:predictive | [
{
"content": "I'm looking at how mobile tech changes work. I'm considering the iPhone 15 and the Samsung S23. Which one is better for video editing?",
"role": "user"
},
{
"content": "The iPhone 15 has excellent stabilization, but the Samsung S23 uses a Snapdragon processor for rendering.",
"role... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "Samsung S23",
"answer_kind": "short_span",
"evidence": "the Samsung S23 uses a Snapdragon processor",
"exact": true,
"family": "entity_binding",
"id": "6440:stacked_entities_same_type:predictive:p1:llm:0... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Multiple similar entities must appear.",
"A later question should depend on remembering which one is being referenced."
],
"output_pair_count_max": 4,
"output_pair_count_min": 4,
... |
7972:two_thread_tracking:predictive | [
{
"content": "Query the given database and print the results: SELECT * FROM customers",
"role": "user"
},
{
"content": "Output: ID | Name | Address | Phone 1 | John | 123 Main | 123-456-7890 2 | Jane | 456 Broadway | 234-567-8901 3 | Dave | 789 Riverside | 345-678-9012",
"role": "assistant"
},... | [
{
"prefix_pair_index": 1,
"probes": [
{
"answer": "123-456-7890",
"answer_kind": "short_span",
"evidence": "1 | John | 123 Main | 123-456-7890",
"exact": true,
"family": "entity_binding",
"id": "7972:two_thread_tracking:predictive:p1:llm:2",
"noo... | {
"filter_schema_version": 4,
"filter_source_repo_id": "BRlkl/chatalpaca-multiturn-predictive-state3",
"must_have": [
"Two distinct threads must appear.",
"Later turns must require tracking which thread is being continued."
],
"output_pair_count_max": 3,
"output_pair_count_min": 3,
"predictive_sta... |
End of preview. Expand in Data Studio
ChatAlpaca Multiturn Predictive State 4
Balanced predictive-state supervision for Samantha multiturn latent-state pretraining.
This version is derived from BRlkl/chatalpaca-multiturn-predictive-state3 with shortcut probes removed.
The main removed failure mode is the overrepresented first-user-request probe whose target is often the full user prompt.
Audit
- rows:
7816 - probes:
15034 - estimated expanded probe branches:
24272 - max family share:
0.1995 - answer equals first user prompt:
0 - first-request-like questions:
0
Probe families
entity_binding:3000early_fact_late_recall:3000constraint_tracking:2266topic_return:2209disambiguation:1255correction_latest_state:1226assistant_statement_recall:738two_thread_tracking:572deferred_question:468other_memory:166compact_prior_state:68topic_binding:66
- Downloads last month
- 24