generated from eliahuhorwitz/Academic-project-page-template
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
1062 lines (1006 loc) · 54.7 KB
/
index.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<!-- Meta tags for social media banners, these should be filled in appropriatly as they are your "business card" -->
<!-- Replace the content tag with appropriate information -->
<meta name="description" content="DESCRIPTION META TAG">
<meta property="og:title" content="SOCIAL MEDIA TITLE TAG" />
<meta property="og:description"
content="SOCIAL MEDIA DESCRIPTION TAG TAG" />
<meta property="og:url" content="URL OF THE WEBSITE" />
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X630-->
<meta property="og:image" content="static/image/your_banner_image.png" />
<meta property="og:image:width" content="1200" />
<meta property="og:image:height" content="630" />
<meta name="twitter:title" content="TWITTER BANNER TITLE META TAG">
<meta name="twitter:description"
content="TWITTER BANNER DESCRIPTION META TAG">
<!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X600-->
<meta name="twitter:image"
content="static/images/your_twitter_banner_image.png">
<meta name="twitter:card" content="summary_large_image">
<!-- Keywords for your paper to be indexed by-->
<meta name="keywords" content="KEYWORDS SHOULD BE PLACED HERE">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>DPT-Agent</title>
<link rel="icon" type="image/x-icon" href="static/images/favicon.ico">
<link
href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
rel="stylesheet">
<link rel="stylesheet" href="static/css/bulma.min.css">
<link rel="stylesheet" href="static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="static/css/bulma-slider.min.css">
<link rel="stylesheet" href="static/css/fontawesome.all.min.css">
<link rel="stylesheet"
href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="static/css/index.css">
<script
src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script src="https://documentcloud.adobe.com/view-sdk/main.js"></script>
<script defer src="static/js/fontawesome.all.min.js"></script>
<script src="static/js/bulma-carousel.min.js"></script>
<script src="static/js/bulma-slider.min.js"></script>
<script src="static/js/index.js"></script>
<style>
/* 確保表格在各列上有統一的寬度 */
.tab-content .column {
flex: 1;
}
/* 調整表格樣式，確保完整顯示內容 */
.tab-content table {
width: 100%;
table-layout: auto; /* 讓表格根據內容自動調整 */
}
/* 移除可能導致內容被截斷的屬性 */
.tab-content th, .tab-content td {
padding: 0.5em 0.4em; /* 縮小內邊距以節省空間 */
font-size: 0.9em; /* 稍微縮小字體 */
word-wrap: normal; /* 防止文本換行 */
white-space: nowrap; /* 保持文本不換行 */
}
/* 在小屏幕上允許水平滾動 */
@media screen and (max-width: 1023px) {
.table-container {
overflow-x: auto;
max-width: 100%;
}
}
/* 新增輪播圖片居中樣式 */
.results-carousel .item {
display: flex;
flex-direction: column;
align-items: center;
justify-content: center;
height: 100%;
}
.results-carousel .item img {
max-height: 60vh; /* 控制圖片最大高度為視窗高度的 60% */
object-fit: contain;
margin: 20px auto; /* 上下間距 20px，左右自動居中 */
}
/* 為標題和副標題添加適當的間距 */
.results-carousel .item h1.title {
margin-top: 15px;
margin-bottom: 10px;
}
.results-carousel .item h2.subtitle {
margin-top: 0;
padding: 0 15px; /* 左右添加一些內邊距 */
}
</style>
</head>
<body>
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-1 publication-title">Leveraging Dual Process
Theory in Language Agent Framework for Real-time Simultaneous
Human-AI Collaboration</h1>
<div class="is-size-5 publication-authors">
<!-- Paper authors -->
<span class="author-block">
<a href="https://shaozhang.info" target="_blank">Shao
Zhang</a><sup>1*</sup>,</span>
<span class="author-block">
<a href="https://xihuai18.github.io" target="_blank">Xihuai
Wang</a><sup>1*</sup>,</span>
<span class="author-block">
Wenhao Zhang<sup>1</sup>,
</span>
<span class="author-block">
Chaoran Li<sup></sup>,
</span>
<span class="author-block">
Junru Song<sup>1</sup>,
</span>
<span class="author-block">
Tingyu Li<sup>1</sup>,
</span>
<span class="author-block">
Lin Qiu<sup>2</sup>,
</span>
<span class="author-block">
Xuezhi Cao<sup>2</sup>,
</span>
<span class="author-block">
Xunliang Cai<sup>2</sup>,
</span>
<span class="author-block">
Wen Yao<sup>3</sup>,
</span>
<span class="author-block">
<a href="https://wnzhang.net" target="_blank">Weinan
Zhang</a><sup>1</sup>,
</span>
<span class="author-block">
Xinbing Wang<sup>1</sup>,
</span>
<span class="author-block">
<a href="https://yingwen.io" target="_blank">Ying
Wen</a><sup>1#</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup> Shanghai Jiao Tong
University, <sup>2</sup> Meituan, <sup>3</sup> Intelligent
Game and Decision Laboratory<br></span>
<span class="eql-cntrb"><small><br><sup>*</sup>Equal
Contribution <sup>#</sup>Corresponding Author</small></span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- Arxiv PDF link -->
<span class="link-block">
<a href="https://arxiv.org/pdf/2502.11882.pdf"
target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Paper</span>
</a>
</span>
<!-- Supplementary PDF link -->
<!-- <span class="link-block">
<a href="" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Leaderboard</span>
</a>
</span> -->
<!-- Github link -->
<span class="link-block">
<a href="https://github.com/sjtu-marl/DPT-Agent"
target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
<!-- ArXiv abstract Link -->
<span class="link-block">
<a href="https://arxiv.org/abs/2502.11882" target="_blank"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>arXiv</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Teaser video-->
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<!-- <video poster="" id="tree" autoplay controls muted loop height="100%"> -->
<!-- Your video here -->
<!-- <source src="static/videos/banner_video.mp4"
type="video/mp4"> -->
<!-- </video> -->
<img src="static/images/intro.png" id="tree" autoplay controls muted
loop height="70%">
<h2 class="subtitle has-text-centered">
How DPT-Agent Collaborates with Human Simultaneously in Real-time.
</h2>
</div>
</div>
</section>
<!-- End teaser video -->
<!-- Paper abstract -->
<section class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Abstract</h2>
<div class="content has-text-justified">
<p>
Agents built on large language models (LLMs) have excelled in
turn-by-turn human-AI collaboration but struggle with
simultaneous tasks requiring real-time interaction. Latency
issues and the challenge of inferring variable human strategies
hinder their ability to make autonomous decisions without
explicit instructions.
Through experiments with current independent <i>System 1</i> and
<i>System 2</i> methods, we validate the necessity of using Dual
Process Theory (DPT) in real-time tasks.
We propose <br>DPT-Agent, a novel language agent framework that
integrates <i>System 1</i> and <i>System 2</i> for efficient
real-time simultaneous human-AI collaboration.
DPT-Agent's <i>System 1</i> uses a Finite-state Machine (FSM)
and code-as-policy for fast, intuitive, and controllable
decision-making.
DPT-Agent's <i>System 2</i> integrates Theory of Mind (ToM) and
asynchronous reflection to infer human intentions and perform
reasoning-based autonomous decisions.
We demonstrate the effectiveness of DPT-Agent through further
experiments with rule-based agents and human collaborators,
showing significant improvements over mainstream LLM-based
frameworks.
To the best of our knowledge, DPT-Agent is the first language
agent framework that achieves successful real-time simultaneous
human-AI collaboration autonomously.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- End paper abstract -->
<!-- Image carousel -->
<section class="hero is-small">
<div class="hero-body">
<div class="container">
<div id="results-carousel" class="carousel results-carousel">
<div class="item">
<!-- Your image here -->
<img src="static/images/framework.png" alt="MY ALT TEXT" />
<h1 class="title is-4 has-text-weight-bold">DPT-Agent Framework</h1>
<h2 class="subtitle">
In System 2, the historical states from the history buffer periodically trigger
the ToM module to infer human behaviors. The reflection module then analyzes the belief output from the ToM module, along with game score feedback and other historical state information, to summarize its own behaviors and
generate guidelines. Within System 1, the code-as-policy generator utilizes the current state, belief and guidelines to
generate code-as-policy when necessary, enabling control over the FSM. When no specific input is provided, the
FSM continues operating autonomously, generating macro actions to ensure the agent maintains continuous action
output, thereby guaranteeing real-time responsiveness in simultaneous collaboration.
</h2>
</div>
<div class="item">
<!-- Your image here -->
<img src="static/images/exp1.png" alt="MY ALT TEXT" />
<h1 class="title is-4 has-text-weight-bold">LLM as Independent System 1 and System 2</h1>
<h2 class="subtitle">
As independent System 1,
models with fewer than 20B parameters excel in
latency but often have near-zero score efficiency,
indicating fast responses but ineffective actions.
Since missed orders lead to score deductions, some
high-score-efficiency models with high latency still
score below zero. The models that can balance
capability in generating scoring actions with low
latency perform better. When the reasoning mod-
els use long CoT as the System 2, despite their
stronger reasoning capabilities, their performance
presents even lower score efficiency and overall
scores compared to many smaller models function-
ing as System 1. Additionally, all agents perform
worse than the FSM agent.
These results show that LLM-based indepen-
dent System 1 and System 2 agents struggle with
low-latency models lacking capability and high-
capability models suffering from excessive latency.
This phenomenon highlights the need for a frame-
work to integrate System 1 and System 2, balancing capability and latency in real-time tasks.
</h2>
</div>
<div class="item">
<!-- Your image here -->
<img src="static/images/exp2.png" alt="MY ALT TEXT" />
<h1 class="title is-4 has-text-weight-bold">Capability in Simultaneous Collaboration - Experiments with Rule-based Agent</h1>
<h2 class="subtitle">
Under ReAct and Reflexion framework, the score efficiency of most models has significantly improved compared with when they functioned as independent System 1 (Figure 2). However, the score of many models has declined with an increase in latency due to more complex System 2 reasoning. Low-latency models, like Qwen2.5-14b, still struggle with capability issues, failing to achieve higher final scores despite good score efficiency. Further comparison of the performance of DPT-Agent reveals that inference models with high latency and larger models get a significant improvement. DPT-Agent can help these high latency models convert the high score efficiency and reasoning capability to scores, which demonstrates the effectiveness of DPT-Agent in real-time tasks.
</h2>
</div>
</div>
</div>
</div>
</section>
<!-- End image carousel -->
<!-- 實驗結果區塊 -->
<section class="section hero">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Experiment Results</h2>
<!-- 標籤頁 -->
<div class="tabs is-centered is-boxed" id="exp-tabs">
<ul>
<li class="is-active" data-target="exp1-tab"><a>Single-Agent Setting </a></li>
<li data-target="exp2-tab"><a> Multi-Agent Setting</a></li>
<li data-target="human-exp-tab"><a>Collaboration with
Human</a></li>
</ul>
</div>
<!-- 標籤頁內容 -->
<div class="tab-content has-text-centered">
<!-- Exp 1 標籤頁 -->
<div id="exp1-tab" class="tab-pane is-active">
<div class="columns is-centered is-desktop">
<!-- 第一個表格 -->
<div class="column">
<h4 class="title is-5">ReAct</h4>
<div class="table-container">
<table
class="table is-striped is-hoverable is-fullwidth has-text-centered">
<thead>
<tr>
<th class="has-text-centered">Models</th>
<th class="has-text-centered">Score</th>
<th class="has-text-centered">Score Efficiency</th>
<th class="has-text-centered">Latency</th>
</tr>
</thead>
<tbody>
<tr>
<td>GPT-4o</td>
<td> 21.00(7.01) </td>
<td>3.08(0.30)</td>
<td> 7.10(0.29) </td>
</tr>
<tr>
<td>GPT-4o-mini</td>
<td> -28.50(6.23) </td>
<td>0.60(0.28)</td>
<td> 3.06(0.07) </td>
</tr>
<tr>
<td>o3-mini-low</td>
<td>5.50(5.86) </td>
<td> 2.51(0.25) </td>
<td>8.64(0.27) </td>
</tr>
<tr>
<td>DeepSeek-V2.5-236b</td>
<td>-21.50(3.56)</td>
<td> 1.72(0.24) </td>
<td> 6.45(0.18)</td>
</tr>
<tr>
<td>DeepSeek-R1-70b</td>
<td>-17.00(4.32)</td>
<td> 1.48(0.17) </td>
<td> 7.79(0.20)</td>
</tr>
<tr>
<td>DeepSeek-R1-32b </td> <td> -15.50(4.51)</td><td>
1.49(0.18)</td> <td> 5.77(0.18) </td>
</tr>
<tr>
<td>DeepSeek-R1-14b </td> <td> -7.00(4.94) </td>
<td> 2.67(0.19) </td> <td>2.91(0.03) </td>
</tr>
<tr>
<td>Llama3.3-70b</td> <td> 20.00(4.21) </td> <td>
2.86(0.16)</td> <td> 5.44(0.05)</td>
</tr>
<tr>
<td>Mistral-nemo-12b </td><td>-10.00(3.31) </td>
<td> 2.40(0.13) </td><td> 1.10(0.03)</td>
</tr>
<tr>
<td>Mistral-small-24b </td> <td>59.50(5.04)
</td><td>4.63(0.20)</td> <td>2.69(0.02) </td>
</tr>
<tr>
<td>Mixtral-8x22b </td> <td> -5.00(5.23) <td>
1.73(0.22) <td>5.56(0.10) </td>
</tr>
<tr>
<td>Qwen2.5-14b <td> -5.00(5.31)</td><td>
1.98(0.21) </td> <td> 1.55(0.03)</td>
</tr>
<tr>
<td>Qwen2.5-32b</td>
<td>10.00(0.50)</td><td>2.94(0.02) </td>
<td> 1.93(0.04)</td>
</tr>
<tr>
<td>Qwen2.5-72b </td> <td>16.50(3.22)
</td><td> 2.71(0.09) </td> <td>4.60(0.09)
</td>
</tr>
</tbody>
</table>
</div>
</div>
<!-- 第二個表格 -->
<div class="column">
<h4 class="title is-5">Reflexion</h4>
<div class="table-container">
<table class="table is-striped is-hoverable is-fullwidth has-text-centered">
<thead>
<tr>
<th class="has-text-centered">Models</th>
<th class="has-text-centered">Score</th>
<th class="has-text-centered">Score Efficiency</th>
<th class="has-text-centered">Latency</th>
</tr>
</thead>
<tbody>
<tr>
<td>GPT-4o </td>
<td>-1.50(3.78) </td>
<td> 2.14(0.17) </td>
<td> 7.49(0.27) </td>
</tr>
<tr>
<td>GPT-4o-mini </td>
<td> -40.00(2.17) </td>
<td> 0.00(0.14) </td>
<td> 3.11(0.08)</td>
</tr>
<tr>
<td>o3-mini-low </td>
<td>-16.50(7.12) </td>
<td> 1.78(0.26) </td>
<td> 8.86(0.23)</td>
</tr>
<tr>
<td>DeepSeek-V2.5 </td>
<td>-25.56(2.91) </td>
<td> 1.24(0.18)</td>
<td>7.64(0.16)</td>
</tr>
<tr>
<td>DeepSeek-R1-70b </td>
<td>-20.00(4.79) </td>
<td> 1.44(0.19)</td>
<td> 7.78(0.17)</td>
</tr>
<tr>
<td>DeepSeek-R1-32b </td>
<td> -37.50(4.77) </td>
<td>0.90(0.21)</td>
<td>7.39(0.11) </td>
</tr>
<tr>
<td>DeepSeek-R1-14b </td>
<td> -10.50(4.12) </td>
<td> 1.93(0.22)</td>
<td> 4.01(0.11)</td>
</tr>
<tr>
<td>Llama3.3-70b </td>
<td>20.00(4.47) </td>
<td> 3.25(0.19)</td>
<td> 5.20(0.06)</td>
</tr>
<tr>
<td>Mistral-nemo-12b </td>
<td> -40.00(0.00) </td>
<td>0.00(0.00)</td>
<td>1.60(0.02) </td>
</tr>
<tr>
<td>Mistral-small-24b </td>
<td>-5.00(3.63) </td>
<td> 1.43(0.03)</td>
<td> 3.11(0.05)</td>
</tr>
<tr>
<td>Mixtral-8x22b </td>
<td> 0.50(4.33)</td>
<td> 2.44(0.20)</td>
<td> 5.58(0.23)</td>
</tr>
<tr>
<td>Qwen2.5-14b </td>
<td>-4.00(4.45) </td>
<td> 2.44(0.24) </td>
<td> 1.87(0.05)</td>
</tr>
<tr>
<td>Qwen2.5-32b </td>
<td> -40.00(0.00) </td>
<td> 0.00(0.00) </td>
<td> 2.93(0.05)</td>
</tr>
<tr>
<td>Qwen2.5-72b </td>
<td> -25.00(2.76) </td>
<td> 1.47(0.09) </td>
<td> 4.66(0.05)</td>
</tr>
</tbody>
</table>
</div>
</div>
<!-- 第三個表格 -->
<div class="column">
<h4 class="title is-5">DPT-Agent w/o ToM</h4>
<div class="table-container">
<table class="table is-striped is-hoverable is-fullwidth has-text-centered">
<thead>
<tr>
<th class="has-text-centered">Models</th>
<th class="has-text-centered">Score</th>
<th class="has-text-centered">Score
Efficiency</th>
<th class="has-text-centered">Latency</th>
</tr>
</thead>
<tbody>
<tr><td>GPT-4o</td> <td> 20.50(5.41) </td> <td> 3.05(0.24) </td> <td> 5.08(0.15) </td></tr>
<tr><td>GPT-4o-mini</td> <td> 21.00(4.47) </td> <td> 3.50(0.23) </td> <td> 2.13(0.01) </td></tr>
<tr><td>o3-mini-low</td> <td> 37.50(4.81) </td> <td> 3.68(0.19) </td> <td> 7.03(0.28) </td></tr>
<tr><td>DeepSeek-V2.5</td> <td>31.50(3.40) </td> <td> 3.40(0.14) </td> <td> 4.73(0.11) </td></tr>
<tr><td>DeepSeek-R1-70b</td> <td> 60.00(4.35) </td> <td> 4.19(0.15)</td> <td> 9.09(0.26) </td></tr>
<tr><td>DeepSeek-R1-32b</td> <td> 39.50(7.68) </td> <td> 3.35(0.27) </td><td> 6.58(0.25) </td></tr>
<tr><td>DeepSeek-R1-14b</td><td> 23.00(5.42) </td> <td> 23.00(5.42) </td><td> 3.87(0.07) </td></tr>
<tr><td>Llama3.3-70b</td> <td> -10.00(6.46)</td> <td> 1.82(0.34) </td> <td> 2.28(0.10) </td></tr>
<tr><td>Mistral-nemo-12b</td> <td> 30.00(5.20) </td> <td> 3.49(0.21)</td> <td> 1.31(0.03) </td></tr>
<tr><td>Mistral-small-24b</td> <td> -1.50(3.63) </td> <td> 2.05(0.17) </td><td> 3.61(0.31) </td></tr>
<tr><td>Mixtral-8x22b</td> <td> 0.00(15.00) </td> <td> 2.70(0.20) </td><td> 4.21(0.17) </td></tr>
<tr><td>Qwen2.5-14b</td> <td> 1.50(4.11) </td> <td> 2.68(0.22) </td> <td> 1.18(0.02) </td></tr>
<tr><td>Qwen2.5-32b</td> <td> 1.00(3.83) </td> <td> 2.26(0.13) </td> <td> 1.65(0.03) </td></tr>
<tr><td>Qwen2.5-72b</td> <td> 11.00(4.88) </td> <td> 2.66(0.21) </td><td> 3.01(0.12) </td></tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
<!-- Exp 2 標籤頁 -->
<div id="exp2-tab" class="tab-pane is-hidden">
<!-- 將說明文字放入與表格相同的列結構中 -->
<div class="columns is-centered is-desktop">
<div class="column">
<div class="content">
<p class="has-text-justified">
DPT-Agent achieved the best performance across the majority of models, especially on the widely recognized general-purpose
SOTA models like GPT-4o. This phenomenon aligns with the conclusions from the experiments
in single-agent settings, where larger models can overcome the latency limitations and achieve better
performance with the help of DPT-Agent. Such performance improvements are more noticeable in
the reasoning model series of GPT o3-mini and DeepSeek-R1. DPT-Agent framework can help reasoning models, which require long periods of
thinking, overcome the latency and effectively transition from thinking to action. Additionally, when
facing rule-based agents that can only perform a single task, DPT-Agent can maintain a high contribution rate. For some models like Llama3.3-70b, DPT-Agent w/o ToM outperforms the complete DPT-Agent, which may be closely related to the model's ToM capabilities.
</p>
</div>
</div>
</div>
<div class="columns is-centered is-desktop">
<!-- 第一個表格 -->
<div class="column">
<h4 class="title is-5">Score</h4>
<div class="table-container">
<table class="table is-striped is-hoverable is-fullwidth has-text-centered">
<thead>
<tr>
<th class="has-text-centered">Models</th>
<th class="has-text-centered">ReAct</th>
<th class="has-text-centered">Reflexion</th>
<th class="has-text-centered">DPT-Agent w/o ToM</th>
<th class="has-text-centered">DPT-Agent</th>
</tr>
</thead>
<tbody>
<tr><td>Overall </td><td> 13.37(6.08) </td><td>15.22(6.42)</td><td>43.67(6.64) </td><td>46.57(6.89) </td></tr>
<tr><td>o3-mini-high </td><td> -43.00(0.93) </td><td>-42.00(0.85)</td><td>65.83(5.66)</td><td> 55.17(4.84) </td></tr>
<tr><td>o3-mini-medium </td><td>-10.00(6.94) </td><td>4.83(7.63) </td><td>56.50(7.07)</td><td> 60.00(6.07) </td></tr>
<tr><td>o3-mini-low </td><td> 7.00(7.491) </td><td>33.50(7.06) </td><td>44.83(9.74)</td><td> 51.33(8.67)</td></tr>
<tr><td>GPT-4o </td><td> 35.67(9.62)</td><td>39.17(8.43)</td><td>18.67(8.50)</td><td>39.50(8.63) </td></tr>
<tr><td>GPT-4o-mini </td><td> -6.58(5.37)</td><td>5.58(7.53)</td><td>50.00(5.27)</td><td>52.92(6.34) </td></tr>
<tr><td>Qwen-Max </td><td>30.50(6.58) </td><td>21.17(6.23)</td><td> 51.50(9.27) </td><td>53.83(7.33) </td></tr>
<tr><td>Claude 3.5 Haiku </td><td> 29.50(5.63) </td><td>24.83(6.58) </td><td>43.17(8.01) </td><td>41.50(7.69) </td></tr>
<tr><td>DeepSeek-R1-671b </td><td>20.67(5.47) </td><td>21.00(6.83)</td><td>56.67(5.13) </td><td>74.33(5.33) </td></tr>
<tr><td>DeepSeek-R1-70b </td><td>33.83(6.77)</td><td>-2.67(5.98)</td><td>51.00(6.08)</td><td>61.50(6.40) </td></tr>
<tr><td>DeepSeek-R1-32b </td><td>37.33(8.51)</td><td>23.33(7.46)</td><td>45.50(6.39)</td><td>38.83(8.51) </td></tr>
<tr><td>DeepSeek-R1-14b </td><td>-8.50(3.88)</td><td>12.00(8.51)</td><td>40.33(7.73)</td><td>43.17(8.54) </td></tr>
<tr><td>DeepSeek-V3 </td><td>29.17(8.24)</td><td> 33.33(7.76)</td><td> 70.33(5.28) </td><td>61.83(5.86) </td></tr>
<tr><td>DeepSeek-V2.5 </td><td> -6.00(5.23) </td><td>12.33(4.83)</td><td> 31.50(6.58)</td><td> 23.50(8.44)</td></tr>
<tr><td>Qwen2.5-72b </td><td> 18.03(4.69)</td><td>48.67(5.68)</td><td>18.83(5.51)</td><td>32.08(5.17) </td></tr>
<tr><td>Llama3.3-70b </td><td> 27.97(5.68)</td><td>-15.58(5.28)</td><td>30.75(3.86)</td><td>28.08(6.68) </td></tr>
<tr><td>Mixtral-8x22b </td><td> 20.17(6.30) </td><td>24.67(6.07)</td><td> 24.00(6.10) </td><td> 26.83(5.79) </td></tr>
</tbody>
</table>
</div>
</div>
<!-- 第二個表格 -->
<div class="column">
<h4 class="title is-5">Agent Contribution Rate</h4>
<div class="table-container">
<table class="table is-striped is-hoverable is-fullwidth has-text-centered">
<thead>
<tr>
<th class="has-text-centered">Models</th>
<th class="has-text-centered">ReAct</th>
<th class="has-text-centered">Reflexion</th>
<th class="has-text-centered">DPT-Agent w/o ToM</th>
<th class="has-text-centered">DPT-Agent</th>
</tr>
</tr>
</thead>
<tbody>
<tr><td>Overall </td><td>0.52(0.03)</td><td> 0.52(0.03)</td><td>0.68(0.03)</td><td> 0.68(0.03)</td></tr>
<tr><td>o3-mini-high </td><td>0.00(0.00)</td><td> 0.00(0.00)</td><td> 0.68(0.02)</td><td> 0.72(0.01) </td></tr>
<tr><td>o3-mini-medium </td><td>0.60(0.05)</td><td> 0.62(0.02)</td><td> 0.56(0.04)</td><td> 0.68(0.03) </td></tr>
<tr><td>o3-mini-low </td><td>0.60(0.05)</td><td> 0.62(0.02)</td><td> 0.56(0.04)</td><td> 0.68(0.03) </td></tr>
<tr><td>GPT-4o </td><td>0.60(0.02) </td><td> 0.61(0.02)</td><td> 0.60(0.05)</td><td> 0.69(0.04) </td></tr>
<tr><td>GPT-4o-mini </td><td>0.27(0.07)</td><td> 0.46(0.06)</td><td> 0.66(0.02)</td><td>0.67(0.02) </td></tr>
<tr><td>Qwen-Max </td><td>0.59(0.03)</td><td> 0.60(0.03)</td><td> 0.68(0.04) </td><td>0.70(0.03) </td></tr>
<tr><td>Claude 3.5 Haiku </td><td>0.62(0.04)</td><td> 0.58(0.03)</td><td> 0.67(0.03)</td><td> 0.70(0.03)</td></tr>
<tr><td>DeepSeek-R1-671b </td><td> 0.61(0.01)</td><td> 0.59(0.01)</td><td>0.69(0.02)</td><td> 0.69(0.01)</td></tr>
<tr><td>DeepSeek-R1-70b </td><td>0.57(0.01)</td><td> 0.55(0.05) </td><td>0.69(0.02) </td><td>0.66(0.02)</td></tr>
<tr><td>DeepSeek-R1-32b </td><td> 0.56(0.02)</td><td> 0.53(0.03) </td><td>0.67(0.02)</td><td> 0.69(0.05)</td></tr>
<tr><td>DeepSeek-R1-14b </td><td>0.52(0.02) </td><td>0.48(0.02) </td><td>0.68(0.03)</td><td>0.71(0.03)</td></tr>
<tr><td>DeepSeek-V3 </td><td> 0.60(0.03) </td><td>0.58(0.02)</td><td> 0.74(0.01) </td><td>0.74(0.02) </td></tr>
<tr><td>DeepSeek-V2.5 </td><td>0.25(0.02) </td><td>0.47(0.04)</td><td>0.64(0.04)</td><td> 0.60(0.04) </td></tr>
<tr><td>Qwen2.5-72b </td><td> 0.75(0.01)</td><td> 0.58(0.01) </td><td>0.67(0.04)</td><td> 0.67(0.03) </td></tr>
<tr><td>Llama3.3-70b </td><td>0.74(0.03)</td><td> 0.54(0.05)</td><td> 0.85(0.02) </td><td>0.75(0.05) </td></tr>
<tr><td>Mixtral-8x22b </td><td> 0.54(0.03)</td><td> 0.54(0.03) </td><td>0.70(0.06) </td><td>0.60(0.03) </td></tr>
</tbody>
</table>
</div>
</div>
<!-- 第三個表格 -->
<div class="column">
<h4 class="title is-5">Score Efficiency</h4>
<div class="table-container">
<table class="table is-striped is-hoverable is-fullwidth has-text-centered">
<thead>
<tr>
<th class="has-text-centered">Models</th>
<th class="has-text-centered">ReAct</th>
<th class="has-text-centered">Reflexion</th>
<th class="has-text-centered">DPT-Agent w/o ToM</th>
<th class="has-text-centered">DPT-Agent</th>
</tr>
</tr>
</thead>
<tbody>
<tr><td>Overall </td><td>3.57(0.30)</td><td> 3.54(0.33)</td><td> 4.59(0.31) </td><td>4.69(0.32) </td></tr>
<tr><td>o3-mini-high </td><td> 0.00(0.00) </td><td>0.00(0.00)</td><td> 5.66(0.21) </td><td>5.33(0.18) </td></tr>
<tr><td>o3-mini-medium </td><td> 2.67(0.38) </td><td>3.59(0.39) </td><td>5.16(0.28)</td><td>5.23(0.24) </td></tr>
<tr><td>o3-mini-low </td><td> 3.20(0.34) </td><td>4.18(0.34)</td><td> 4.28(0.43) </td><td>4.60(0.35) </td><td></td></tr>
<tr><td>GPT-4o </td><td> 4.26(0.42) </td><td>3.86(0.34) </td><td>3.43(0.42)</td><td> 4.46(0.39)</td></tr>
<tr><td>GPT-4o-mini </td><td> 3.95(0.52) </td><td>4.64(0.66) </td><td>5.03(0.28)</td><td> 5.33(0.33) </td></tr>
<tr><td>Qwen-Max </td><td>4.56(0.39)</td><td> 4.03(0.28) </td><td>4.83(0.45) </td><td>5.09(0.31) </td></tr>
<tr><td>Claude 3.5 Haiku </td><td> 4.04(0.30) </td><td>3.65(0.31)</td><td> 4.67(0.39)</td><td> 4.47(0.34) </td></tr>
<tr><td>DeepSeek-R1-671b </td><td>4.52(0.25) </td><td>4.47(0.37)</td><td> 4.90(0.21) </td><td>5.27(0.19) </td></tr>
<tr><td>DeepSeek-R1-70b </td><td>3.66(0.25)</td><td> 2.25(0.27) </td><td>4.64(0.25)</td><td> 4.92(0.24)</td></tr>
<tr><td>DeepSeek-R1-32b </td><td>3.64(0.29)</td><td> 3.27(0.29) </td><td>4.31(0.26)</td><td> 4.04(0.38) </td></tr>
<tr><td>DeepSeek-R1-14b </td><td>3.16(0.22)</td><td> 3.16(0.43) </td><td>4.33(0.36) </td><td>4.29(0.38) </td></tr>
<tr><td>DeepSeek-V3 </td><td>4.78(0.39)</td><td> 5.03(0.38) </td><td>6.00(0.18) </td><td>5.66(0.25) </td></tr>
<tr><td>DeepSeek-V2.5 </td><td> 2.29(0.26) </td><td>3.43(0.29) </td><td>4.24(0.40) </td><td>3.61(0.42) </td></tr>
<tr><td>Qwen2.5-72b </td><td> 4.44(0.16)</td><td> 5.11(0.29) </td><td>3.25(0.27) </td><td>4.51(0.29) </td></tr>
<tr><td>Llama3.3-70b </td><td> 4.44(0.37) </td><td>2.01(0.28) </td><td>4.08(0.19)</td><td> 3.89(0.32)</td></tr>
<tr><td>Mixtral-8x22b </td><td> 3.58(0.30)</td><td> 4.01(0.32) </td><td>4.63(0.41)</td><td> 4.38(0.43) </td></tr>
</tr>
</tbody>
</table>
</div>
</div>
<div class="column">
<h4 class="title is-5">Latency</h4>
<div class="table-container">
<table class="table is-striped is-hoverable is-fullwidth has-text-centered">
<thead>
<tr>
<th class="has-text-centered">Models</th>
<th class="has-text-centered">ReAct</th>
<th class="has-text-centered">Reflexion</th>
<th class="has-text-centered">DPT-Agent w/o ToM</th>
<th class="has-text-centered">DPT-Agent</th>
</tr>
</tr>
</thead>
<tbody>
<tr><td>o3-mini-high </td><td>39.01(2.82) </td><td>39.47(2.43) </td><td>34.77(4.37)</td><td>35.96(4.91)</td></tr>
<tr><td>o3-mini-medium </td><td>28.07(2.42) </td><td>26.73(3.93) </td><td>22.24(1.39)</td><td>24.05(2.81)</td></tr>
<tr><td>o3-mini-low </td><td>10.78(1.40) </td><td>10.58(0.80) </td><td>7.34(0.37)</td><td>7.68(0.38)</td></tr>
<tr><td>GPT-4o </td><td> 6.63(7.53) </td><td>6.81(0.24)</td><td> 4.92(1.32) </td><td>4.91(1.41)</td></tr>
<tr>
<td>GPT-4o-mini</td>
<td>2.93(0.77)</td>
<td>3.15(1.27)</td>
<td>2.09(1.09)</td>
<td>2.08(0.58)</td>
</tr>
<tr><td>Qwen-Max </td><td>8.29(0.14)</td><td> 10.30(0.21)</td><td> 5.90(0.11)</td><td> 5.89(0.10) </td></tr>
<tr><td>Claude 3.5 Haiku </td><td>5.74(0.06) </td><td>7.47(0.11)</td><td> 5.21(0.05)</td><td> 5.25(0.06)</td></tr>
<tr><td>DeepSeek-R1-671b </td><td>31.31(2.17) </td><td>41.66(2.45)</td><td> 38.89(3.70) </td><td>34.63(2.30)</td></tr>
<tr><td>DeepSeek-R1-70b </td><td>7.82(0.17)</td><td> 7.39(0.14) </td><td>10.30(0.36)</td><td> 10.13(0.34)</td></tr>
<tr><td>DeepSeek-R1-32b </td><td>5.75(0.09) </td><td>6.77(0.13)</td><td> 5.24(0.08)</td><td> 5.11(0.13) </td></tr>
<tr><td>DeepSeek-R1-14b </td><td> 3.06(0.06) </td><td>3.44(0.09)</td><td> 3.88(0.088) </td><td>3.57(0.06)</td></tr>
<tr><td>DeepSeek-V3 </td><td> 7.54(0.15)</td><td>8.86(0.15) </td><td>1.92(0.04) </td><td>2.41(0.10) </td></tr>
<tr><td>DeepSeek-V2.5 </td><td> 4.88(0.07) </td><td>5.35(0.08) </td><td>4.06(0.10) </td><td>4.49(0.07) </td></tr>
<tr><td>Qwen2.5-72b </td><td> 4.34(0.06) </td><td>4.83(0.11) </td><td>3.81(0.10) </td><td>4.62(0.11) </td></tr>
<tr><td>Llama3.3-70b </td><td> 4.53(0.08)</td><td> 5.34(0.11) </td><td>2.30(0.08) </td><td>2.90(0.09) </td></tr>
<tr><td>Mixtral-8x22b </td><td> 5.20(0.18) </td><td>5.19(0.22) </td><td>4.53(0.14)</td><td> 5.31(0.19) </td></tr>
<tr><td>Overall </td><td>10.99(1.14)</td><td> 12.08(0.78)</td><td>9.84(0.84) </td><td>9.94(0.85)</td></tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
<!-- Human Exp 標籤頁 -->
<div id="human-exp-tab" class="tab-pane is-hidden">
<div class="columns is-centered is-desktop">
<div class="column">
<div class="content">
<p class="has-text-justified">
After data validation, we have 68 valid data points
in total: 36 of Map 1 and 32 of Map 2. DPT-Agent achieves the highest scores in
both Map 1 and Map 2 when collaborating with
humans. DPT-Agent w/o ToM also outperforms
ReAct and Reflexion, confirming the effectiveness
of asynchronous reflection. Moreover, the ToM
module also brought a significant score improvement in collaborating with humans, confirming that incorporating human belief reasoning into System 2 can foster better collaboration. Regarding human perception, DPT-Agent ranks highest
in Map 1, with the most participants recognizing
its collaborative abilities. Interestingly, in Map 2,
DPT-Agent w/o ToM surpasses DPT-Agent in both
cooperation and preference ranking with a higher
agent contribution rate, which may refer to the hu-
man preference for partners who work more.
</p>
</div>
</div>
</div>
<div class="columns is-centered is-desktop">
<!-- 第一個表格 -->
<div class="column">
<h4 class="title is-5">Map 1 - New Counter Circuit</h4>
<div class="table-container">
<table class="table is-striped is-hoverable is-fullwidth has-text-centered">
<thead>
<tr>
<th class="has-text-centered">Agent Frameworks</th>
<th class="has-text-centered">Score</th>
<th class="has-text-centered">Agent Contribution Rate</th>
<th class="has-text-centered">Borda Count of Cooperation</th>
<th class="has-text-centered">Borda Count of Preference</th>
</tr>
</thead>
<tbody>
<tr>
<td>ReAct</td>
<td>99.03(9.86)</td>
<td>0.51(0.03)</td>
<td>88</td>
<td>88</td>
</tr>
<tr>
<td>Reflexion</td>
<td>97.78(7.23)</td>
<td>0.53(0.03)</td>
<td>79</td>
<td>80</td>
</tr>
<tr>
<td>DPT-Agent w/o ToM</td>
<td> 103.19(7.06)</td>
<td> 0.62(0.02)</td>
<td>86</td>
<td>91</td>
</tr>
<tr>
<td>DPT-Agent</td>
<td> 111.53(5.42)</td>
<td> 0.62(0.02)</td>
<td>107</td>
<td>101</td>
</tr>
</tbody>
</table>
</div>
</div>
<!-- 第二個表格 -->
<div class="column">
<h4 class="title is-5">Map 2 - New Asymmetric Advantages</h4>
<div class="table-container">
<table class="table is-striped is-hoverable is-fullwidth has-text-centered">
<thead>
<tr>
<th class="has-text-centered">Agent Frameworks</th>
<th class="has-text-centered">Score</th>
<th class="has-text-centered">Agent Contribution Rate</th>
<th class="has-text-centered">Borda Count of Cooperation</th>
<th class="has-text-centered">Borda Count of Preference</th>
</tr>
</thead>
<tbody>
<tr>
<td>ReAct</td>
<td>115.00(9.28)</td>
<td>0.49(0.04)</td>
<td>65</td>
<td>63</td>
</tr>
<tr>
<td>Reflexion</td>
<td>119.67(10.54)</td>
<td>0.51(0.03)</td>
<td>78</td>
<td>73</td>
</tr>
<tr>
<td>DPT-Agent w/o ToM</td>
<td>152.03(8.13)</td>
<td> 0.62(0.02)</td>
<td>94</td>
<td>95</td>
</tr>
<tr>
<td>DPT-Agent</td>
<td> 160.63(7.97)</td>
<td>0.59(0.03)</td>
<td>83</td>
<td>89</td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
<!-- Human Exp 標籤頁 -->
</div>
</div>
</div>
</div>
</section>
<!-- 添加標籤頁切換的 JavaScript -->
<script>
document.addEventListener('DOMContentLoaded', () => {
// 獲取所有標籤頁和內容
const tabs = document.querySelectorAll('#exp-tabs li');
const tabPanes = document.querySelectorAll('.tab-content .tab-pane');
tabs.forEach(tab => {
tab.addEventListener('click', () => {
// 重置所有標籤頁和內容的狀態
tabs.forEach(item => item.classList.remove('is-active'));
// 修正這裡 - classList.add() 不能鏈式調用
tabPanes.forEach(pane => {
pane.classList.add('is-hidden');
pane.classList.remove('is-active');
});
// 激活當前點擊的標籤頁
tab.classList.add('is-active');
// 顯示對應的內容
const targetId = tab.getAttribute('data-target');
const targetPane = document.getElementById(targetId);
targetPane.classList.remove('is-hidden');
targetPane.classList.add('is-active');
});
});
});
</script>
<!-- Youtube video -->
<!-- <section class="hero is-small is-light">
<div class="hero-body">
<div class="container"> -->
<!-- Paper video. -->
<!-- <h2 class="title is-3">Video Presentation</h2>
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<div class="publication-video"> -->
<!-- Youtube embed code here -->
<!-- <iframe src="https://www.youtube.com/embed/JkaxUblCGz0" frameborder="0" allow="autoplay; encrypted-media" allowfullscreen></iframe>
</div>
</div>
</div>
</div>
</div> -->
<!-- </section> -->
<!-- End youtube video -->
<!-- Video carousel -->
<!-- <section class="hero is-small">
<div class="hero-body">
<div class="container">
<h2 class="title is-3">Another Carousel</h2>
<div id="results-carousel" class="carousel results-carousel">
<div class="item item-video1">
<video poster="" id="video1" autoplay controls muted loop height="100%"> -->
<!-- Your video file here -->
<!-- <source src="static/videos/carousel1.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-video2">
<video poster="" id="video2" autoplay controls muted loop height="100%"> -->
<!-- Your video file here -->
<!-- <source src="static/videos/carousel2.mp4"
type="video/mp4">
</video>
</div>
<div class="item item-video3">
<video poster="" id="video3" autoplay controls muted loop height="100%">\ -->
<!-- Your video file here -->
<!-- <source src="static/videos/carousel3.mp4"
type="video/mp4">
</video>
</div>
</div>
</div>
</div>
</section> -->
<!-- End video carousel -->
<!-- Paper poster -->
<!-- <section class="hero is-small is-light">
<div class="hero-body">
<div class="container">
<h2 class="title">Poster</h2>
<iframe src="static/pdfs/sample.pdf" width="100%" height="550">
</iframe>
</div>
</div>
</section> -->
<!--End paper poster -->