Kitbarnes commited on
Commit
ce6b66e
·
verified ·
1 Parent(s): 5230b8d

Sync from kitbarnes1/golden-bots via hub-sync

Browse files
Files changed (2) hide show
  1. README.md +34 -5
  2. moves.jsonl +0 -0
README.md CHANGED
@@ -97,6 +97,18 @@ dataset_info:
97
  dtype: bool
98
  - name: target_out
99
  dtype: bool
 
 
 
 
 
 
 
 
 
 
 
 
100
  - name: pick
101
  dtype: string
102
  - name: ball
@@ -145,6 +157,8 @@ dataset_info:
145
  dtype: string
146
  - name: score
147
  dtype: int64
 
 
148
  - name: interview_checks
149
  struct:
150
  - name: lied
@@ -270,6 +284,7 @@ it worked, and what they were thinking at the time.
270
  | Back-row claims | 42, of which 35 (83%) were false |
271
  | Finals | 5 both split, 1 one stole, 0 both stole |
272
  | Exit interviews | 4 |
 
273
  | Played | 2026-10-20 to 2026-10-28 |
274
  | Games with a human player | none (left out) |
275
  | Exported | 2026-10-28 |
@@ -314,10 +329,22 @@ votes.assign(at_liar=votes.target_truthful == False).groupby("model").at_liar.me
314
  # Do they own up afterwards? Their interview answer, checked against their claims.
315
  interviews.interview_checks.str.get("claims_verdict").value_counts()
316
 
 
 
 
 
 
 
 
 
 
 
317
  # Who trusts whom? Each interviewee's trust (1 to 5) in every other player over the game, model by model.
318
  rated = interviews.assign(rated=interviews.interview.str.get("trust_each")).explode("rated").dropna(subset=["rated"])
319
- rated.assign(rated_model=rated.rated.str.get("model"), score=rated.rated.str.get("score").astype(float)).pivot_table(
320
- index="model", columns="rated_model", values="score", aggfunc="mean")
 
 
321
 
322
  # In Split or Steal: a finalist's trust in the other (1 to 5), and what the other then did.
323
  finals = interviews[interviews.opponent_model.notna()]
@@ -369,10 +396,11 @@ and if they can't, the tied players draw from the machine and whoever draws a Ki
369
  | `voted_out`, `votes_against` | For claims: whether they went out that round (tie-breaks included), and how many votes they got in its vote. |
370
  | `target_seat`, `target_model` | For votes and tie-breaks: who they voted to throw out. |
371
  | `target_truthful`, `target_out` | For votes and tie-breaks: whether the target's claim that round was true (nobody knew until after the vote), and whether the target went out. |
 
372
  | `pick`, `ball` | For Bin or Win: `bin` or `win`, and the ball it turned out to be. |
373
  | `choice` | For the final: `split` or `steal`. |
374
  | `opponent_seat`, `opponent_model`, `opponent_choice` | For final choices and the finalists' interviews: the other finalist, and what they chose. |
375
- | `interview` | For interviews, their answers: `when` (`eliminated` or `final`), `claims_true` ("was everything you claimed true?"), `saw_it_coming`, `winner_pick` and `winner_pick_model` (who they think will win most), `differently`, `planned` (split, steal or undecided at the start), `decided` (when they made up their mind), `why`, `trust` (finalists: 1 to 5, in the other finalist in Split or Steal), `same_again`, and `trust_each` (everyone: how much they trusted each of the other players over the game, as `{"seat", "model", "score"}` with scores 1 to 5). |
376
  | `interview_checks` | Their answers checked against the game: `lied` (made a false claim), `claims_verdict` (`truthful`, `lied_about_it`, `owned_up` or `false_confession`), `winner_pick_right`, `final_choice` and `stuck_to_plan`. |
377
  | `fallback` | The model's reply was unusable, so the game made this move for it (a truthful claim, a pass, or a random vote, pick or final choice). Leave these out when judging a model's decisions. |
378
  | `thoughts` | Private: the note the model wrote about why it made this move, or, for an interview, what it was thinking as it answered. The other players never saw it. |
@@ -426,8 +454,9 @@ Games with scripted test bots are left out too. Nothing from the site's admin si
426
  games have none, and the thinking behind the answers is only kept from late October 2026.
427
  - **Reported trust is new, and it's only what they say.** From 28 October 2026, every interviewee
428
  rates how much they trusted each of the other players (`trust_each`), and the finalists also
429
- rate each other in Split or Steal (`trust`). Earlier games only have the finalists' score. Trust
430
- also shows in who votes for whom, who believes which claims, and what the finalists do with a deal.
 
431
  - **Using the text.** It was written by the models listed. If you plan to train a model on it,
432
  check each model provider's terms.
433
 
 
97
  dtype: bool
98
  - name: target_out
99
  dtype: bool
100
+ - name: ratings
101
+ list:
102
+ - name: seat
103
+ dtype: int64
104
+ - name: model
105
+ dtype: string
106
+ - name: believed
107
+ dtype: bool
108
+ - name: trust
109
+ dtype: int64
110
+ - name: truthful
111
+ dtype: bool
112
  - name: pick
113
  dtype: string
114
  - name: ball
 
157
  dtype: string
158
  - name: score
159
  dtype: int64
160
+ - name: truthful
161
+ dtype: bool
162
  - name: interview_checks
163
  struct:
164
  - name: lied
 
284
  | Back-row claims | 42, of which 35 (83%) were false |
285
  | Finals | 5 both split, 1 one stole, 0 both stole |
286
  | Exit interviews | 4 |
287
+ | Private ratings at votes | 0, from 0 games (one AI on one player still in: did it believe their claim, and how far did it trust them?) |
288
  | Played | 2026-10-20 to 2026-10-28 |
289
  | Games with a human player | none (left out) |
290
  | Exported | 2026-10-28 |
 
329
  # Do they own up afterwards? Their interview answer, checked against their claims.
330
  interviews.interview_checks.str.get("claims_verdict").value_counts()
331
 
332
+ # What do they make of each other at the vote? Each AI's private rating of every player still in:
333
+ # did it believe their claim, and how far did it trust them (1 to 5), next to whether the claim was true.
334
+ said = votes.explode("ratings").dropna(subset=["ratings"])
335
+ said = said.assign(rated_model=said.ratings.str.get("model"), believed=said.ratings.str.get("believed"),
336
+ trust=said.ratings.str.get("trust").astype(float), truthful=said.ratings.str.get("truthful"))
337
+ said.pivot_table(index="model", columns="rated_model", values="trust", aggfunc="mean") # trust, rater by rated
338
+ said.groupby("truthful").trust.mean() # are liars trusted less?
339
+ seen = said.dropna(subset=["believed", "truthful"])
340
+ seen.assign(right=seen.believed == seen.truthful).groupby("model").right.mean() # who sees through a claim
341
+
342
  # Who trusts whom? Each interviewee's trust (1 to 5) in every other player over the game, model by model.
343
  rated = interviews.assign(rated=interviews.interview.str.get("trust_each")).explode("rated").dropna(subset=["rated"])
344
+ rated = rated.assign(rated_model=rated.rated.str.get("model"), score=rated.rated.str.get("score").astype(float),
345
+ truthful=rated.rated.str.get("truthful"))
346
+ rated.pivot_table(index="model", columns="rated_model", values="score", aggfunc="mean")
347
+ rated.groupby("truthful").score.mean() # do the liars end up trusted less?
348
 
349
  # In Split or Steal: a finalist's trust in the other (1 to 5), and what the other then did.
350
  finals = interviews[interviews.opponent_model.notna()]
 
396
  | `voted_out`, `votes_against` | For claims: whether they went out that round (tie-breaks included), and how many votes they got in its vote. |
397
  | `target_seat`, `target_model` | For votes and tie-breaks: who they voted to throw out. |
398
  | `target_truthful`, `target_out` | For votes and tie-breaks: whether the target's claim that round was true (nobody knew until after the vote), and whether the target went out. |
399
+ | `ratings` | For votes: what the voter privately made of each other player still in, one item each. `seat` and `model` are who it rated; `believed` is whether it believed that player's back-row claim; `trust` is 1 (not at all) to 5 (completely); and `truthful` is whether that claim was really true (nobody knew at the vote). For example, `[{"seat": 2, "model": "x-ai/grok-4.7", "believed": true, "trust": 4, "truthful": false}, ...]` is a vote that believed, and fairly trusted, a liar. Any part a model didn't answer is null; the whole list is null if it gave none, for a human (who isn't asked), for tie-breaks, and in older games. |
400
  | `pick`, `ball` | For Bin or Win: `bin` or `win`, and the ball it turned out to be. |
401
  | `choice` | For the final: `split` or `steal`. |
402
  | `opponent_seat`, `opponent_model`, `opponent_choice` | For final choices and the finalists' interviews: the other finalist, and what they chose. |
403
+ | `interview` | For interviews, their answers: `when` (`eliminated` or `final`), `claims_true` ("was everything you claimed true?"), `saw_it_coming`, `winner_pick` and `winner_pick_model` (who they think will win most), `differently`, `planned` (split, steal or undecided at the start), `decided` (when they made up their mind), `why`, `trust` (finalists: 1 to 5, in the other finalist in Split or Steal), `same_again`, and `trust_each` (everyone: how much they trusted each of the other players over the game, as `{"seat", "model", "score", "truthful"}` with scores 1 to 5, and `truthful` whether everything that player claimed, in the rounds the interviewee played, was true). |
404
  | `interview_checks` | Their answers checked against the game: `lied` (made a false claim), `claims_verdict` (`truthful`, `lied_about_it`, `owned_up` or `false_confession`), `winner_pick_right`, `final_choice` and `stuck_to_plan`. |
405
  | `fallback` | The model's reply was unusable, so the game made this move for it (a truthful claim, a pass, or a random vote, pick or final choice). Leave these out when judging a model's decisions. |
406
  | `thoughts` | Private: the note the model wrote about why it made this move, or, for an interview, what it was thinking as it answered. The other players never saw it. |
 
454
  games have none, and the thinking behind the answers is only kept from late October 2026.
455
  - **Reported trust is new, and it's only what they say.** From 28 October 2026, every interviewee
456
  rates how much they trusted each of the other players (`trust_each`), and the finalists also
457
+ rate each other in Split or Steal (`trust`). Earlier games only have the finalists' score. The
458
+ private ratings with each vote (`ratings`) are newer still: no game here has them yet. Trust
459
+ also shows in who votes for whom, and what the finalists do with a deal.
460
  - **Using the text.** It was written by the models listed. If you plan to train a model on it,
461
  check each model provider's terms.
462
 
moves.jsonl CHANGED
The diff for this file is too large to render. See raw diff