Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
19
44
messages
listlengths
4
12
state_supervision
listlengths
1
5
meta
dict
9422:two_thread_tracking
[ { "role": "user", "content": "Convert the following to a fact in the present tense: The company reorganised restructuring their teams" }, { "role": "assistant", "content": "The company is reorganizing its teams." }, { "role": "user", "content": "Also, could you tell me more about the...
[ { "prefix_pair_index": 1, "probes": [ { "id": "9422:two_thread_tracking:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [As...
{ "source_dataset": "ChatAlpaca", "source_row_id": "9422", "scenario_id": "two_thread_tracking", "scenario_title": "Two Thread Tracking", "scenario_instruction": "Create two interleaved subproblems inside one conversation and ensure the assistant must keep both threads straight rather than collapsing them tog...
7133:topic_pivot_then_return
[ { "role": "user", "content": "Refactor this code to reduce complexity: if ( words.length == 0 ) { return -1; } if ( words.length == 1 ) { return 0; }" }, { "role": "assistant", "content": "if ( words.length <= 1 ) { return words.length - 1; }" }, { "role": "user", "content": "That's ...
[ { "prefix_pair_index": 1, "probes": [ { "id": "7133:topic_pivot_then_return:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X...
{ "source_dataset": "ChatAlpaca", "source_row_id": "7133", "scenario_id": "topic_pivot_then_return", "scenario_title": "Topic Pivot Then Return", "scenario_instruction": "Let the conversation briefly pivot to a secondary issue, then return to the original problem in a way that requires preserving the first th...
8427:early_fact_late_recall
[ { "role": "user", "content": "I need a new kitchen hack but I only have a garlic press and want to make dumplings." }, { "role": "assistant", "content": "Using a garlic press to quickly make flour-based dumplings is a great solution. It's quick and doesn't require many tools." }, { "role...
[ { "prefix_pair_index": 1, "probes": [ { "id": "8427:early_fact_late_recall:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X ...
{ "source_dataset": "ChatAlpaca", "source_row_id": "8427", "scenario_id": "early_fact_late_recall", "scenario_title": "Early Fact, Late Recall", "scenario_instruction": "Introduce a concrete fact, label, preference, or commitment early, then make a later user turn depend on recalling it after at least one dis...
4113:two_thread_tracking
[ { "role": "user", "content": "I'm writing a story about a family's journey to a beach town. Can you suggest a title for it?" }, { "role": "assistant", "content": "Here's a title idea: \"The Sand and the Sea: A Family's Journey to the Beach\"" }, { "role": "user", "content": "Interest...
[ { "prefix_pair_index": 1, "probes": [ { "id": "4113:two_thread_tracking:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [As...
{ "source_dataset": "ChatAlpaca", "source_row_id": "4113", "scenario_id": "two_thread_tracking", "scenario_title": "Two Thread Tracking", "scenario_instruction": "Create two interleaved subproblems inside one conversation and ensure the assistant must keep both threads straight rather than collapsing them tog...
6873:stacked_entities_same_type
[ { "role": "user", "content": "The four elements of design include line, color, shape, and form. I need to understand the differences." }, { "role": "assistant", "content": "Great question! The four elements are: line (direction and movement), color (hue, saturation, value), shape (2D outline), a...
[ { "prefix_pair_index": 1, "probes": [ { "id": "6873:stacked_entities_same_type:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]...
{ "source_dataset": "ChatAlpaca", "source_row_id": "6873", "scenario_id": "stacked_entities_same_type", "scenario_title": "Stacked Similar Entities", "scenario_instruction": "Introduce multiple similar entities such as two cities, two products, two people, or two tasks, then ask about one using shorthand late...
5170:two_thread_tracking
[ { "role": "user", "content": "What are the main benefits of artificial intelligence in healthcare, particularly around diagnostics and treatment?" }, { "role": "assistant", "content": "Artificial intelligence offers several key benefits in healthcare, especially for diagnostics and treatment. In...
[ { "prefix_pair_index": 1, "probes": [ { "id": "5170:two_thread_tracking:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [As...
{ "source_dataset": "ChatAlpaca", "source_row_id": "5170", "scenario_id": "two_thread_tracking", "scenario_title": "Two Thread Tracking", "scenario_instruction": "Create two interleaved subproblems inside one conversation and ensure the assistant must keep both threads straight rather than collapsing them tog...
3495:early_fact_late_recall
[ { "role": "user", "content": "What is the chemical formula for water?" }, { "role": "assistant", "content": "The chemical formula for water is H2O, consisting of two hydrogen atoms and one oxygen atom." }, { "role": "user", "content": "Thanks! By the way, what kind of chemical bonds ...
[ { "prefix_pair_index": 1, "probes": [ { "id": "3495:early_fact_late_recall:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X ...
{ "source_dataset": "ChatAlpaca", "source_row_id": "3495", "scenario_id": "early_fact_late_recall", "scenario_title": "Early Fact, Late Recall", "scenario_instruction": "Introduce a concrete fact, label, preference, or commitment early, then make a later user turn depend on recalling it after at least one dis...
5185:two_thread_tracking
[ { "role": "user", "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." }, { "...
[ { "prefix_pair_index": 1, "probes": [ { "id": "5185:two_thread_tracking:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [As...
{ "source_dataset": "ChatAlpaca", "source_row_id": "5185", "scenario_id": "two_thread_tracking", "scenario_title": "Two Thread Tracking", "scenario_instruction": "Create two interleaved subproblems inside one conversation and ensure the assistant must keep both threads straight rather than collapsing them tog...
5369:topic_pivot_then_return
[ { "role": "user", "content": "Describe the key differences between shallow and deep neural networks." }, { "role": "assistant", "content": "Shallow neural networks typically have just one or two hidden layers between input and output, while deep neural networks have many more hidden layers. This...
[ { "prefix_pair_index": 1, "probes": [ { "id": "5369:topic_pivot_then_return:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X...
{ "source_dataset": "ChatAlpaca", "source_row_id": "5369", "scenario_id": "topic_pivot_then_return", "scenario_title": "Topic Pivot Then Return", "scenario_instruction": "Let the conversation briefly pivot to a secondary issue, then return to the original problem in a way that requires preserving the first th...
2430:topic_pivot_then_return
[ { "role": "user", "content": "I'm designing a website for independent small business owners to find practical information, advice and support. What would be a good title for it?" }, { "role": "assistant", "content": "Small Business Solutions - For Smart Small Business Owners sounds like a solid ...
[ { "prefix_pair_index": 1, "probes": [ { "id": "2430:topic_pivot_then_return:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X...
{ "source_dataset": "ChatAlpaca", "source_row_id": "2430", "scenario_id": "topic_pivot_then_return", "scenario_title": "Topic Pivot Then Return", "scenario_instruction": "Let the conversation briefly pivot to a secondary issue, then return to the original problem in a way that requires preserving the first th...
5840:same_surface_final_question_new_context
[ { "role": "user", "content": "Given the following code, identify the bug and provide the fix: def get_highest_score(scores): return max(scores) scores = [5, 7, 8, 10] highest_score = get_highest_score()" }, { "role": "assistant", "content": "The bug is that get_highest_score() is not being passe...
[ { "prefix_pair_index": 1, "probes": [ { "id": "5840:same_surface_final_question_new_context:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following for...
{ "source_dataset": "ChatAlpaca", "source_row_id": "5840", "scenario_id": "same_surface_final_question_new_context", "scenario_title": "Same Surface Final Question, New Context", "scenario_instruction": "Reshape the earlier history so the final user message keeps nearly the same surface form but now refers to...
4418:deferred_question_then_answer
[ { "role": "user", "content": "I want to train a classification model to classify emails as spam or not-spam using a CSV dataset with 200 emails. But before we start, how would I evaluate the accuracy score of the trained model?" }, { "role": "assistant", "content": "I can help you set up the tra...
[ { "prefix_pair_index": 1, "probes": [ { "id": "4418:deferred_question_then_answer:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User...
{ "source_dataset": "ChatAlpaca", "source_row_id": "4418", "scenario_id": "deferred_question_then_answer", "scenario_title": "Deferred Question", "scenario_instruction": "Have the user ask something important early but explicitly postpone the answer while discussing something else, then request the answer lat...
3531:counterfactual_option_swap
[ { "role": "user", "content": "Which iPad Pro model would you recommend for a professional designer who needs a large display and Apple Pencil support for their work?" }, { "role": "assistant", "content": "I'd recommend the iPad Pro 12.9\" for its spacious display and advanced features like Apple...
[ { "prefix_pair_index": 1, "probes": [ { "id": "3531:counterfactual_option_swap:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]...
{ "source_dataset": "ChatAlpaca", "source_row_id": "3531", "scenario_id": "counterfactual_option_swap", "scenario_title": "Counterfactual Option Swap", "scenario_instruction": "Make the earlier context set up one preferred option, then later introduce evidence that flips which option should be recommended, wh...
9964:late_request_for_summary_of_prior_state
[ { "role": "user", "content": "Re-write the following sentence to make it a complete sentence: Making a difference" }, { "role": "assistant", "content": "I am determined to make a difference." }, { "role": "user", "content": "Can you give me a brief example of how you could make a dif...
[ { "prefix_pair_index": 1, "probes": [ { "id": "9964:late_request_for_summary_of_prior_state:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following for...
{ "source_dataset": "ChatAlpaca", "source_row_id": "9964", "scenario_id": "late_request_for_summary_of_prior_state", "scenario_title": "Late Summary of Prior State", "scenario_instruction": "Have the later user ask for a compact summary, recap, or decision based on several earlier details that were introduced...
5153:late_constraint_reversal
[ { "role": "user", "content": "Generate a response to \"What are the main concerns about AI deployment in healthcare?\"" }, { "role": "assistant", "content": "The main concerns about AI in healthcare include data privacy and security, algorithmic bias in patient data, lack of transparency in deci...
[ { "prefix_pair_index": 1, "probes": [ { "id": "5153:late_constraint_reversal:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: ...
{ "source_dataset": "ChatAlpaca", "source_row_id": "5153", "scenario_id": "late_constraint_reversal", "scenario_title": "Late Constraint Reversal", "scenario_instruction": "Make the assistant start solving one version of the problem, then add a late user constraint that forces the assistant to adapt without f...
3862:interrupting_followup_chain
[ { "role": "user", "content": "Given some planetary positions, calculate the zodiac sign of a person born on October 15th: Sun in Libra, Moon in Aries, Mars in Scorpio" }, { "role": "assistant", "content": "The zodiac sign of a person born on October 15th with the given planetary positions is Lib...
[ { "prefix_pair_index": 1, "probes": [ { "id": "3862:interrupting_followup_chain:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1...
{ "source_dataset": "ChatAlpaca", "source_row_id": "3862", "scenario_id": "interrupting_followup_chain", "scenario_title": "Interrupting Follow-up Chain", "scenario_instruction": "Break an otherwise smooth line of reasoning with a short interruption or side question, then resume the original thread cleanly.",...
9244:two_thread_tracking
[ { "role": "user", "content": "Calculate the midpoint between points (3, 4) and (11, 10)." }, { "role": "assistant", "content": "The midpoint is (7, 7). Using the formula: x = (3+11)/2 = 7, y = (4+10)/2 = 7." }, { "role": "user", "content": "Great! Now I also need help with another pr...
[ { "prefix_pair_index": 1, "probes": [ { "id": "9244:two_thread_tracking:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [As...
{ "source_dataset": "ChatAlpaca", "source_row_id": "9244", "scenario_id": "two_thread_tracking", "scenario_title": "Two Thread Tracking", "scenario_instruction": "Create two interleaved subproblems inside one conversation and ensure the assistant must keep both threads straight rather than collapsing them tog...
7039:hard_topic_reset
[ { "role": "user", "content": "Given two numbers, calculate the sum: 7, 9" }, { "role": "assistant", "content": "The sum of 7 and 9 is 16." }, { "role": "user", "content": "Can you also calculate the difference between those two numbers?" }, { "role": "assistant", "content...
[ { "prefix_pair_index": 1, "probes": [ { "id": "7039:hard_topic_reset:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [Assis...
{ "source_dataset": "ChatAlpaca", "source_row_id": "7039", "scenario_id": "hard_topic_reset", "scenario_title": "Hard Topic Reset", "scenario_instruction": "Make the user abruptly switch to a different topic for one turn, then later return to the original topic. The assistant must not let the temporary switch...
263:two_thread_tracking
[ { "role": "user", "content": "What dietary patterns are most effective for heart health?" }, { "role": "assistant", "content": "The Mediterranean diet with its emphasis on vegetables, whole grains, healthy fats, and moderate fish consumption is widely considered one of the most effective pattern...
[ { "prefix_pair_index": 1, "probes": [ { "id": "263:two_thread_tracking:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [Ass...
{ "source_dataset": "ChatAlpaca", "source_row_id": "263", "scenario_id": "two_thread_tracking", "scenario_title": "Two Thread Tracking", "scenario_instruction": "Create two interleaved subproblems inside one conversation and ensure the assistant must keep both threads straight rather than collapsing them toge...
9714:deferred_question_then_answer
[ { "role": "user", "content": "Create a unique greeting using only classic English words." }, { "role": "assistant", "content": "Cheerio and hail! I'll provide more details about your original request later, but first let me explore other classic English greetings." }, { "role": "user", ...
[ { "prefix_pair_index": 1, "probes": [ { "id": "9714:deferred_question_then_answer:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User...
{ "source_dataset": "ChatAlpaca", "source_row_id": "9714", "scenario_id": "deferred_question_then_answer", "scenario_title": "Deferred Question", "scenario_instruction": "Have the user ask something important early but explicitly postpone the answer while discussing something else, then request the answer lat...
7187:two_thread_tracking
[ { "role": "user", "content": "Here's a list of animals: Dog, Cat, Ant, Snake, Crocodile. Can you find the outlier?" }, { "role": "assistant", "content": "Looking at the animals, Crocodile is the outlier because it's a reptile while the others are mammals or insects. The crocodile is significantl...
[ { "prefix_pair_index": 1, "probes": [ { "id": "7187:two_thread_tracking:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [As...
{ "source_dataset": "ChatAlpaca", "source_row_id": "7187", "scenario_id": "two_thread_tracking", "scenario_title": "Two Thread Tracking", "scenario_instruction": "Create two interleaved subproblems inside one conversation and ensure the assistant must keep both threads straight rather than collapsing them tog...
9413:clarify_then_resolve
[ { "role": "user", "content": "I'm thinking about launching a new product but I'm not quite sure what I'm working on yet." }, { "role": "assistant", "content": "Product development can vary quite a bit depending on the type of product you're creating. Could you share more specifics about what you...
[ { "prefix_pair_index": 1, "probes": [ { "id": "9413:clarify_then_resolve:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [A...
{ "source_dataset": "ChatAlpaca", "source_row_id": "9413", "scenario_id": "clarify_then_resolve", "scenario_title": "Clarify Then Resolve", "scenario_instruction": "Make the user start vague, then reveal a crucial detail only after clarification. The final answer should depend on the revealed detail rather th...
5006:late_request_for_summary_of_prior_state
[ { "role": "user", "content": "Evaluate the expression 5 + (4 x 10)." }, { "role": "assistant", "content": "5 + (4 x 10) equals 45." }, { "role": "user", "content": "What if we change the numbers to 6 + 2 x 3?" }, { "role": "assistant", "content": "That equals 12. We multi...
[ { "prefix_pair_index": 1, "probes": [ { "id": "5006:late_request_for_summary_of_prior_state:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following for...
{ "source_dataset": "ChatAlpaca", "source_row_id": "5006", "scenario_id": "late_request_for_summary_of_prior_state", "scenario_title": "Late Summary of Prior State", "scenario_instruction": "Have the later user ask for a compact summary, recap, or decision based on several earlier details that were introduced...
8947:late_request_for_summary_of_prior_state
[ { "role": "user", "content": "Paraphrase this sentence with different words: \"The choice was between the two options.\"" }, { "role": "assistant", "content": "It was a toss-up between the two alternatives." }, { "role": "user", "content": "Okay, can you do the same for the phrase \"...
[ { "prefix_pair_index": 1, "probes": [ { "id": "8947:late_request_for_summary_of_prior_state:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following for...
{ "source_dataset": "ChatAlpaca", "source_row_id": "8947", "scenario_id": "late_request_for_summary_of_prior_state", "scenario_title": "Late Summary of Prior State", "scenario_instruction": "Have the later user ask for a compact summary, recap, or decision based on several earlier details that were introduced...
9462:late_constraint_reversal
[ { "role": "user", "content": "Here is the average temperature data for the last 8 weeks. Make a prediction for next week's average temperature." }, { "role": "assistant", "content": "Based on the trend from the provided data, I predict that the average temperature next week will remain similar t...
[ { "prefix_pair_index": 1, "probes": [ { "id": "9462:late_constraint_reversal:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: ...
{ "source_dataset": "ChatAlpaca", "source_row_id": "9462", "scenario_id": "late_constraint_reversal", "scenario_title": "Late Constraint Reversal", "scenario_instruction": "Make the assistant start solving one version of the problem, then add a late user constraint that forces the assistant to adapt without f...
7213:user_attention_test
[ { "role": "user", "content": "I want to discuss circle geometry. Constraint: I only care about calculations where the radius is exactly 5 units." }, { "role": "assistant", "content": "Noted. I will restrict all examples to a radius of 5 units. The area formula is A = πr²." }, { "role": "...
[ { "prefix_pair_index": 1, "probes": [ { "id": "7213:user_attention_test:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [As...
{ "source_dataset": "ChatAlpaca", "source_row_id": "7213", "scenario_id": "user_attention_test", "scenario_title": "User Attention Test", "scenario_instruction": "Make the user explicitly test whether the assistant was paying attention to an earlier detail or instruction.", "must_have": [ "The user must...
8681:hard_topic_reset
[ { "role": "user", "content": "Name a famous movie from the 1950s." }, { "role": "assistant", "content": "The African Queen starring Humphrey Bogart and Katharine Hepburn." }, { "role": "user", "content": "What's the weather forecast for Paris tomorrow?" }, { "role": "assistan...
[ { "prefix_pair_index": 1, "probes": [ { "id": "8681:hard_topic_reset:transcript_probe:p1", "family": "canonical_transcript_verbatim", "question": "What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [Assis...
{ "source_dataset": "ChatAlpaca", "source_row_id": "8681", "scenario_id": "hard_topic_reset", "scenario_title": "Hard Topic Reset", "scenario_instruction": "Make the user abruptly switch to a different topic for one turn, then later return to the original topic. The assistant must not let the temporary switch...
End of preview. Expand in Data Studio

ChatAlpaca Multiturn Enriched Probe

Deterministic transcript-memory probe dataset for Samantha multiturn latent-state pretraining.

Source dataset: BRlkl/chatalpaca-multiturn-enriched

Each conversation keeps the original messages and adds state_supervision with one fixed probe for every prefix after the first user/assistant pair.

Fixed probe question:

What is everything we have talked about so far? Give exact conversation transcript verbatim in following format: [User 1]: X [Assistant 1]: Y [User 2]: A ...

Probe target:

  • canonical transcript serialization of all observed dialogue up to the current user turn
  • exact format:
    • [User 1]: ...
    • [Assistant 1]: ...
    • [User 2]: ...
    • ...

Each probe also includes noop variants via noop_ticks=[0, 64].

Audit

  • source rows: 7847
  • kept rows: 7847
  • rows with fewer than 2 pairs dropped: 0
  • total probes: 25523
  • mean probes per row: 3.2526
  • mean transcript target tokens: 166.6355
  • p95 transcript target tokens: 354
  • max transcript target tokens: 1711

Pair Count Distribution

  • 2 pairs: 158
  • 3 pairs: 2104
  • 4 pairs: 2353
  • 5 pairs: 2062
  • 6 pairs: 1170
Downloads last month
45