forked from eliahuhorwitz/Academic-project-page-template
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
3257 lines (3230 loc) · 233 KB
/
Copy pathindex.html
File metadata and controls
3257 lines (3230 loc) · 233 KB
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 lang="en">
<head>
<meta charset="utf-8"/>
<meta content="width=device-width, initial-scale=1.0" name="viewport"/>
<title>Self-Improvements in Modern Agentic Systems — Survey Hub</title>
<style>
:root {
--bg:#f8f7f4;
--paper:#ffffff;
--text:#1f2937;
--muted:#6b7280;
--line:#e5e7eb;
--soft:#f3f4f6;
--blue:#1d4ed8;
--blue-soft:#dbeafe;
--green:#065f46;
--green-soft:#dff3ea;
--fm:#1e40af;
--scaf:#0f766e;
--eval:#7c3aed;
--max:1240px;
--shadow:0 8px 28px rgba(15,23,42,.06);
}
* { box-sizing:border-box; }
html { scroll-behavior:smooth; }
body {
margin:0;
background:var(--bg);
color:var(--text);
font-family: Georgia, "Times New Roman", serif;
line-height:1.66;
}
a { color:var(--blue); text-decoration:none; }
a:hover { text-decoration:underline; }
.topbar {
position:sticky; top:0; z-index:40;
background:rgba(248,247,244,.95);
border-bottom:1px solid var(--line);
backdrop-filter:blur(10px);
}
.topbar-inner {
max-width:var(--max); margin:0 auto; padding:14px 24px;
display:flex; align-items:center; justify-content:space-between; gap:18px;
}
.brand {
font:600 15px/1.2 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
letter-spacing:.04em; color:#111827;
}
.nav {
display:flex; gap:18px; flex-wrap:wrap;
font:500 13px/1.2 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.page {
max-width:var(--max); margin:0 auto; padding:28px 24px 70px;
display:grid; grid-template-columns:280px minmax(0,1fr); gap:40px;
}
.sidebar {
position:sticky; top:74px; align-self:start; max-height: calc(100vh - 90px);
overflow-y: auto; overflow-x: hidden; overscroll-behavior-y: contain; padding-right: 8px; scrollbar-gutter: stable;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.sidebar::-webkit-scrollbar {width: 6px; }
.sidebar::-webkit-scrollbar-track {background: transparent; }
.sidebar::-webkit-scrollbar-thumb {background: #d1d5db; border-radius: 999px; }
.sidebar::-webkit-scrollbar-thumb:hover {background: #9ca3af; }
.side-card {
background:var(--paper); border:1px solid var(--line); border-radius:16px;
padding:18px; margin-bottom:16px; box-shadow:var(--shadow);
}
.side-title {
font-size:12px; text-transform:uppercase; letter-spacing:.08em;
color:var(--muted); margin-bottom:10px; font-weight:700;
}
.toc a { display:block; padding:6px 0; color:#374151; font-size:14px; }
.toc .sub { padding-left:12px; font-size:13px; color:var(--muted); }
.toc .sub2 { padding-left:24px; font-size:12.5px; color:var(--muted); }
.stat-grid {
display:grid; grid-template-columns:1fr 1fr; gap:10px; margin-top:10px;
}
.stat {
background:var(--soft); border:1px solid var(--line); border-radius:12px;
padding:10px 11px;
}
.stat strong {
display:block; font-size:18px; line-height:1.1; color:#111827;
}
.stat span { display:block; margin-top:4px; font-size:12px; color:var(--muted); }
.main { min-width:0; }
.hero {
background:var(--paper); border:1px solid var(--line); border-radius:20px;
padding:34px; margin-bottom:22px; box-shadow:var(--shadow);
}
.eyebrow {
font:700 11px/1 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
color:var(--muted); text-transform:uppercase; letter-spacing:.12em;
margin-bottom:14px;
}
h1 {
margin:0 0 12px 0; font-size:44px; line-height:1.08; font-weight:700; letter-spacing:-.02em;
}
.subtitle {
margin:0 0 18px 0; color:#4b5563; font-size:19px; max-width:820px;
}
.actions {
display:flex; flex-wrap:wrap; gap:12px; margin-top:18px;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.btn {
display:inline-flex; align-items:center; justify-content:center; gap:8px;
padding:10px 14px; border-radius:10px; border:1px solid var(--line);
font-size:14px; font-weight:600; background:#fff; color:#111827;
}
.btn.primary { background:#111827; color:#fff; border-color:#111827; }
section {
background:var(--paper); border:1px solid var(--line); border-radius:20px;
padding:28px 30px; margin-bottom:22px; box-shadow:var(--shadow);
}
h2 { margin:0 0 14px 0; font-size:30px; line-height:1.15; }
h3 {
margin:0 0 8px 0;
font:700 23px/1.28 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
color:#111827;
}
p { margin:0 0 14px 0; }
.lead { font-size:18px; color:#374151; }
.note {
margin-top:14px; padding:13px 15px; border-left:3px solid var(--blue);
background:var(--blue-soft); border-radius:0 12px 12px 0;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
font-size:14px; color:#1e3a8a;
}
.overview-grid {
display:grid; grid-template-columns:1.15fr .85fr; gap:18px; align-items:start;
}
.figure-frame {
border:1px solid var(--line); border-radius:16px; padding:18px; background:#fcfcfd;
}
.overview-image {
display:block; width:100%; height:auto; border-radius:12px;
}
.overview-note {
border:1px solid var(--line); border-radius:16px; padding:18px; background:var(--soft);
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
font-size:14px; color:#374151;
}
.overview-note ul { margin:0; padding-left:18px; }
.overview-note li { margin:8px 0; }
.tax-grid {
display:grid; grid-template-columns:repeat(3,minmax(0,1fr)); gap:18px; margin-top:14px;
}
.tax-card {
border:1px solid var(--line); border-radius:16px; padding:18px; background:#fff;
}
.tax-card.fm { border-top:4px solid var(--fm); }
.tax-card.scaf { border-top:4px solid var(--scaf); }
.tax-card.eval { border-top:4px solid var(--eval); }
.tax-card h4 {
margin:0 0 10px 0;
font:700 18px/1.3 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.tax-card ul { margin:0; padding-left:20px; }
.tax-card li { margin:8px 0; }
.mini-count {
display:inline-block; margin-left:8px; font-size:12px; color:var(--muted);
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.card-grid {
display:grid; grid-template-columns:repeat(3, minmax(0,1fr)); gap:14px;
margin-top:14px;
}
.ref-card {
border:1px solid var(--line); border-radius:16px; padding:16px; background:#fff;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.ref-card strong {
display:block; font-size:16px; margin-bottom:5px; line-height:1.35;
}
.ref-card .meta { font-size:12px; color:var(--muted); margin-bottom:8px; text-transform:none; }
.ref-card p { font-size:14px; color:#374151; margin-bottom:10px; }
.toolbar {
display:flex; gap:12px; flex-wrap:wrap; align-items:center; margin:14px 0 18px;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.search-wrap {
flex:1 1 320px; position:relative;
}
.search-wrap input {
width:100%; padding:11px 14px; border:1px solid var(--line); border-radius:12px;
background:#fff; font:14px/1.2 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.chip-group { display:flex; gap:8px; flex-wrap:wrap; }
.chip {
border:1px solid var(--line); background:#fff; color:#374151;
border-radius:999px; padding:8px 12px; font-size:13px; cursor:pointer;
}
.chip.active { background:#111827; color:#fff; border-color:#111827; }
.toolbar-meta { font-size:13px; color:var(--muted); margin-left:auto; }
.library-block {
padding-top:2px; margin-top:24px; border-top:1px solid var(--line);
}
.library-block:first-of-type { margin-top:12px; }
.block-head {
display:flex; justify-content:space-between; align-items:flex-start; gap:16px; margin:20px 0 12px;
}
.block-head p {
margin:0; font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif; color:#4b5563; font-size:14px;
max-width:760px;
}
.eval-group-title {
margin:24px 0 10px; font:700 19px/1.35 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif; color:#111827;
}
.count-badge {
flex:none; font:600 12px/1 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
color:#111827; background:var(--soft); border:1px solid var(--line); border-radius:999px;
padding:8px 10px; white-space:nowrap;
}
.count-badge.small { padding:6px 8px; font-size:11px; }
.subgroup {
border:1px solid var(--line); border-radius:14px; padding:0 0 14px; margin:12px 0;
overflow:hidden; background:#fff;
}
.subgroup[open] { box-shadow:0 4px 18px rgba(15,23,42,.04); }
.subgroup summary {
list-style:none; cursor:pointer; display:flex; justify-content:space-between; align-items:center; gap:12px;
padding:14px 16px; background:#fbfbfc; font:600 15px/1.35 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
border-bottom:1px solid var(--line);
}
.subgroup summary::-webkit-details-marker { display:none; }
.sub-desc {
margin:12px 16px 10px; color:#4b5563; font:14px/1.6 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.table-wrap { overflow:auto; }
table {
width:100%; border-collapse:collapse;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
font-size:14px;
}
th, td {
padding:11px 12px; border-bottom:1px solid var(--line); text-align:left; vertical-align:top;
}
th {
font-size:12px; text-transform:uppercase; letter-spacing:.08em; color:var(--muted);
background:#fcfcfd;
position:sticky; top:0;
}
.paper-table tbody tr:hover { background:#fafafa; }
td.year { width:72px; color:#4b5563; }
td.links { white-space:nowrap; color:#4b5563; }
.muted-link { color:#9ca3af; }
.two-col {
display:grid; grid-template-columns:1fr 1fr; gap:18px;
}
pre {
white-space:pre-wrap; word-break:break-word; margin:0;
font-family: ui-monospace, SFMono-Regular, Menlo, Consolas, monospace;
font-size:13px; line-height:1.65;
background:#111827; color:#e5e7eb; padding:18px; border-radius:14px;
}
.authors {
display:grid; grid-template-columns:repeat(3,minmax(0,1fr)); gap:12px;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.author {
border:1px solid var(--line); border-radius:14px; padding:14px 15px; background:#fff;
}
.author strong { display:block; margin-bottom:4px; }
.author span { color:var(--muted); font-size:14px; }
.contact-block {
margin-top:18px; padding:16px 18px; border:1px solid var(--line); border-radius:14px;
background:var(--soft); font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.contact-block strong { display:block; margin-bottom:6px; color:#111827; }
.contact-block .small { font-size:13px; color:var(--muted); margin-bottom:8px; }
.contact-block .mono {
font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;
font-size:12.5px; line-height:1.75; color:#374151; word-break:break-word;
}
.resource-links {
display:flex; flex-wrap:wrap; gap:10px; margin-top:12px;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
/* Community & engagement */
.community-head {
display:flex; align-items:flex-start; justify-content:space-between; gap:24px;
padding-bottom:22px; margin-bottom:20px; border-bottom:1px solid var(--line);
}
.community-kicker {
margin-bottom:8px;
font:700 11px/1 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
color:var(--scaf); text-transform:uppercase; letter-spacing:.12em;
}
.community-head h2 { margin-bottom:8px; }
.community-head p { max-width:720px; color:#4b5563; margin:0; }
.community-star-cta {
flex:none; display:inline-flex; align-items:center; gap:9px;
padding:11px 14px; border-radius:12px; background:#111827; color:#fff;
border:1px solid #111827;
font:650 14px/1 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
box-shadow:0 8px 20px rgba(17,24,39,.12);
}
.community-star-cta:hover { text-decoration:none; transform:translateY(-1px); }
.community-star-cta .star-symbol { color:#fbbf24; font-size:17px; }
.community-star-cta strong {
min-width:24px; padding-left:8px; border-left:1px solid rgba(255,255,255,.25);
font-variant-numeric:tabular-nums;
}
.engagement-grid {
display:grid; grid-template-columns:1.05fr .95fr .95fr; gap:14px; margin-bottom:18px;
}
.engagement-card {
border:1px solid var(--line); border-radius:16px; padding:18px; background:#fff;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.engagement-card h3 {
display:flex; align-items:center; gap:8px; margin:0 0 8px;
font-size:17px; line-height:1.25;
}
.engagement-card > p {
color:var(--muted); font-size:13px; line-height:1.55; margin:0 0 14px;
}
.pulse-grid {
display:grid; grid-template-columns:repeat(3,1fr); gap:9px;
}
.pulse {
padding:12px 9px; text-align:center; border:1px solid var(--line);
border-radius:12px; background:var(--soft);
}
.pulse strong {
display:block; font-size:22px; line-height:1.1; color:#111827;
font-variant-numeric:tabular-nums;
}
.pulse span {
display:block; margin-top:5px; color:var(--muted); font-size:11px;
text-transform:uppercase; letter-spacing:.06em;
}
.pulse-status {
display:flex; align-items:center; gap:7px; margin-top:11px; color:var(--muted); font-size:12px;
}
.pulse-status::before {
content:""; width:7px; height:7px; border-radius:50%; background:#d1d5db;
}
.pulse-status.live::before { background:#10b981; box-shadow:0 0 0 3px rgba(16,185,129,.12); }
.community-links {
display:grid; gap:9px;
}
.community-link {
display:flex; justify-content:space-between; align-items:center; gap:12px;
padding:10px 11px; border:1px solid var(--line); border-radius:11px;
color:#1f2937; background:#fcfcfd; font-size:13px; font-weight:650;
}
.community-link:hover {
text-decoration:none; border-color:#c7d2fe; background:#f8fafc;
}
.community-link span:last-child { color:#9ca3af; font-weight:500; }
.community-link.primary {
color:#fff; background:#1d4ed8; border-color:#1d4ed8;
}
.community-link.primary span:last-child { color:#dbeafe; }
.discussion-panel {
display:grid; grid-template-columns:270px minmax(0,1fr); gap:20px;
padding:20px; border:1px solid var(--line); border-radius:16px;
background:linear-gradient(135deg,#fbfdff 0%,#f8fafc 100%);
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.discussion-copy h3 { margin:0 0 8px; font-size:18px; }
.discussion-copy p { color:#4b5563; font-size:13px; line-height:1.6; margin:0 0 12px; }
.discussion-copy .privacy-note {
padding:10px 11px; border-radius:10px; background:#eef2ff; color:#3730a3;
font-size:12px; line-height:1.5;
}
.discussion-host {
min-height:170px; border:1px dashed #cbd5e1; border-radius:13px; background:#fff;
padding:16px;
}
.discussion-placeholder {
display:flex; min-height:136px; align-items:center; justify-content:center; text-align:center;
color:var(--muted); font-size:13px; line-height:1.6;
}
.discussion-placeholder strong { display:block; color:#374151; margin-bottom:3px; }
.utterances { max-width:none !important; }
footer {
max-width:var(--max); margin:0 auto; padding:0 24px 40px;
color:var(--muted); font:13px/1.6 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.image-zoom-trigger {
display:block; width:100%; padding:0; border:0; background:none; cursor:zoom-in;
border-radius:12px;
}
.image-zoom-trigger:focus-visible {
outline:2px solid var(--blue); outline-offset:4px;
}
.figure-caption {
margin-top:12px; color:#374151; font-size:14px;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.zoom-hint {
margin-top:5px; color:var(--muted); font-size:12px;
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;
}
.epigraph-text {
font-size:16px; line-height:1.7; color:#1f2937; font-style:italic;
}
.epigraph-author {
margin-top:10px; text-align:right; color:var(--muted); font-size:13px;
}
.image-modal {
position:fixed; inset:0; z-index:120; display:none; align-items:center; justify-content:center;
background:rgba(2,6,23,.96); padding:0; overflow:hidden; cursor:zoom-out;
}
.image-modal.open { display:flex; }
.image-modal-inner {
position:relative; width:100vw; height:100vh; padding:0; cursor:default;
display:flex; align-items:center; justify-content:center;
}
.image-modal img {
display:block; width:100vw; height:100vh; object-fit:contain;
border-radius:0; background:transparent; box-shadow:none;
transition:transform .22s ease; transform-origin:center center; cursor:zoom-in;
}
.image-modal img.zoomed {
width:auto; height:auto; min-width:100vw; min-height:100vh; max-width:none; max-height:none;
transform:scale(1.15); cursor:zoom-out;
}
.image-modal-close {
position:fixed; top:18px; right:18px; width:44px; height:44px; border:0; border-radius:999px;
background:rgba(255,255,255,.96); color:#111827; font-size:30px; line-height:1; cursor:pointer; box-shadow:var(--shadow);
}
@media (max-width: 980px) {
.page { grid-template-columns:1fr; }
.sidebar { position:static; max-height: none; overflow: visible; padding-right: 0; scrollbar-gutter: auto; }
.overview-grid, .two-col, .tax-grid { grid-template-columns:1fr; }
.engagement-grid { grid-template-columns:1fr 1fr; }
.engagement-card:first-child { grid-column:1 / -1; }
.discussion-panel { grid-template-columns:1fr; }
.card-grid { grid-template-columns:1fr 1fr; }
}
@media (max-width: 760px) {
.community-head { flex-direction:column; }
.community-star-cta { width:100%; justify-content:center; }
.engagement-grid, .discussion-panel { grid-template-columns:1fr; }
.pulse-grid { grid-template-columns:repeat(3,1fr); }
h1 { font-size:34px; }
h2 { font-size:26px; }
.hero, section { padding:22px; }
.authors, .card-grid, .stat-grid { grid-template-columns:1fr; }
.block-head { flex-direction:column; align-items:flex-start; }
}
</style>
</head>
<body>
<div class="topbar">
<div class="topbar-inner">
<div class="brand">SI-Agents Survey</div>
<nav class="nav">
<a href="https://arxiv.org/abs/2607.13104" rel="noopener" target="_blank">Paper</a>
<a href="#abstract">Abstract</a>
<a href="#taxonomy">Taxonomy</a>
<a href="#library">Paper Library</a>
<a href="#community">Community</a>
<a href="#citation">Citation</a>
</nav>
</div>
</div>
<div class="page">
<aside class="sidebar">
<div class="side-card">
<div class="side-title">Survey statistics</div>
<div style="font-size:14px;color:#374151;line-height:1.7">
A curated map of self-improvement mechanisms in foundation model-based agentic systems.
</div>
<div class="stat-grid">
<div class="stat"><strong>312</strong><span>Curated entries</span></div><div class="stat"><strong>77</strong><span>FM improvement</span></div><div class="stat"><strong>176</strong><span>Scaffolding improvement</span></div><div class="stat"><strong>59</strong><span>Evaluation & Benchmarking</span></div>
</div>
</div>
<div class="side-card toc">
<div class="side-title">Contents</div>
<a href="#abstract">Abstract</a><a href="#overview">Overview</a><a href="#recursive-note">Epigraph</a><a href="#taxonomy">Taxonomy</a><a href="#quickstart">Quick start: representative papers</a><a href="#library">Curated paper library</a>
<a class="sub" href="#sgd">1.1 Intrinsic generative demonstrations</a><a class="sub" href="#evaluative-feedback">1.2 Intrinsic evaluative feedback</a><a class="sub" href="#experience-env">1.3.1 Grounded executable environments</a><a class="sub" href="#experience-world">1.3.2 Generative world models</a><a class="sub" href="#prompt-scalar">2.1.1 Scalar-feedback optimization</a><a class="sub" href="#prompt-qualitative">2.1.2 Qualitative-feedback refinement</a><a class="sub" href="#prompt-evolution">2.1.3 Population-based evolution</a><a class="sub" href="#prompt-textgrad">2.1.4 Textual gradient optimization</a><a class="sub" href="#memory-object">2.2.1 Memory object</a><a class="sub" href="#memory-structure">2.2.2 Memory structure</a><a class="sub" href="#memory-processing">2.2.3 Memory processing</a><a class="sub" href="#tool-routing">2.3.1 Dynamic tool routing</a><a class="sub" href="#tool-refine">2.3.2 Iterative tool refinement</a><a class="sub" href="#tool-create">2.3.3 Autonomous tool creation</a><a class="sub" href="#full-scaffolding">2.4 Full scaffolding</a><a class="" href="#evaluation">3. Evaluation & benchmarking</a><a class="sub" href="#measuring-improvement">3.1 Measuring improvement</a><a class="sub2" href="#metric-measurement">3.1.1 Metric-based measurement</a><a class="sub2" href="#judge-measurement">3.1.2 Judge-based measurement</a><a class="sub" href="#benchmarking-improvement">3.2 Benchmarking improvement</a><a class="sub2" href="#mechanism-benchmarks">3.2.1 Mechanism benchmarks</a><a class="sub2" href="#domain-benchmarks">3.2.2 Domain benchmarks</a>
<a href="#reading-guide">How to read this survey</a><a href="#community">Community & feedback</a><a href="#authors">Authors</a><a href="#citation">Citation</a>
</div>
<div class="side-card">
<div class="side-title">Project links</div>
<div class="resource-links">
<a class="btn" href="https://arxiv.org/abs/2607.13104" rel="noopener" target="_blank">arXiv Paper</a>
<a class="btn" href="https://github.com/selfimproving-agent/Awesome-Self-Improving-Agents" rel="noopener" target="_blank">GitHub List</a>
</div>
</div>
<div class="side-card" id="recursive-note">
<div class="side-title">Epigraph</div>
<div class="epigraph">
<div class="epigraph-text">“The first ultraintelligent machine is the last invention that man need ever make.”</div>
<div class="epigraph-author">— I. J. Good (1966)</div>
</div>
</div>
</aside>
<main class="main">
<div class="hero">
<div class="eyebrow">2026 survey webpage</div>
<h1>Self-Improvements in Modern Agentic Systems</h1>
<p class="subtitle">
A survey of how foundation model-based agents improve themselves through model updates,
intrinsic demonstrations, intrinsic evaluative feedback, extrinsic experience, prompt evolution, memory, tool use, and recursive scaffolding.
</p>
<div class="actions">
<a class="btn primary" href="#library">Browse curated papers</a>
<a class="btn" href="#taxonomy">View taxonomy</a>
<a class="btn" href="https://github.com/selfimproving-agent/Awesome-Self-Improving-Agents" rel="noopener" target="_blank">GitHub repository</a>
<a class="btn" href="#community">Community & feedback</a>
</div>
</div>
<section id="abstract">
<h2>Abstract</h2>
<p class="lead">Foundation model-based agents are moving from static prompt-following systems toward systems that can improve themselves over time. This survey organizes self-improvement mechanisms into two major branches: <strong>foundation model improvement</strong>, which updates the model itself, and <strong>scaffolding improvement</strong>, which updates surrounding components such as prompts, memory, tools, and executable agent logic. A third branch, <strong>evaluation & benchmarking</strong>, organizes how improvement is measured and compared.</p>
<p>
Across the survey, the central distinction is between improving the underlying foundation model and
improving the scaffolding around it. This distinction provides a practical way to compare methods,
interpret learning signals, and connect representative papers to broader research trends.
</p>
<div class="note">
The paper list is organized so that readers can move directly from the taxonomy to representative works and
then to broader literature within each mechanism.
</div>
</section>
<section id="overview">
<h2>Overview</h2>
<div class="overview-grid">
<div class="figure-frame">
<button aria-label="Open overview figure in full size" class="image-zoom-trigger" onclick="openImageModal()" type="button">
<img alt="Overview of self-improvement mechanisms in foundation model-based agentic systems" class="overview-image" onerror="this.onerror=null;this.src='assets/fig-si-main-001.png';" src="static/images/fig-si-main-001.png"/>
</button>
<div class="figure-caption">Overview of self-improvement mechanisms in foundation model-based agentic systems.</div>
<div class="zoom-hint">Click the figure to view full screen.</div>
</div>
<div class="overview-note">
<p><strong>Core distinction.</strong> Self-improvement can be framed by asking <em>what is being updated</em>.</p>
<ul>
<li><strong>Foundation model improvement</strong> changes the model itself and tends to be slower, more persistent, and more training-centric.</li>
<li><strong>Scaffolding improvement</strong> changes the operational shell around the model and tends to be faster, cheaper, and more reversible.</li>
<li>The paper library below follows this logic so that taxonomy and literature map onto each other directly.</li>
</ul>
</div>
</div>
</section>
<section id="taxonomy">
<h2>Taxonomy</h2>
<p>
The survey can be read as a structured map of how an FM-based agent improves itself. Each item below links
directly into the corresponding literature section.
</p>
<div class="tax-grid">
<div class="tax-card fm">
<h4>1. Foundation Model Improvement <span class="mini-count">77 papers</span></h4>
<ul>
<li><a href="#sgd">1.1 Intrinsic Generative Demonstrations</a> <span class="mini-count">23</span></li>
<li><a href="#evaluative-feedback">1.2 Intrinsic Evaluative Feedback</a> <span class="mini-count">21</span></li>
<li><a href="#experience-env">1.3.1 Grounded Executable Environments</a> <span class="mini-count">18</span></li>
<li><a href="#experience-world">1.3.2 Generative World Models</a> <span class="mini-count">15</span></li>
</ul>
</div>
<div class="tax-card scaf">
<h4>2. Scaffolding Improvement <span class="mini-count">176 papers</span></h4>
<ul>
<li><a href="#prompt-scalar">2.1 Prompt Optimization</a> <span class="mini-count">39</span></li>
<li><a href="#memory-object">2.2 Memory</a> <span class="mini-count">65</span></li>
<li><a href="#tool-routing">2.3 Tool</a> <span class="mini-count">51</span></li>
<li><a href="#full-scaffolding">2.4 Full Scaffolding</a> <span class="mini-count">21</span></li>
</ul>
</div>
<div class="tax-card eval"><h4>3. Evaluation & Benchmarking <span class="mini-count">59 entries</span></h4><ul><li><a href="#measuring-improvement">3.1 Measuring Improvement</a> <span class="mini-count">19</span></li><li><a href="#benchmarking-improvement">3.2 Benchmarking Improvement</a> <span class="mini-count">40</span></li></ul></div></div>
<div class="note">The Curated Paper Library now follows three top-level branches: 1. Foundation Model Improvement (77 entries), 2. Scaffolding Improvement (176 entries), and 3. Evaluation & Benchmarking (59 categorized entries), for 312 categorized entries in total. Evaluation benchmarks may intentionally appear in more than one 3.x subsection because the evaluation taxonomy is multi-perspective.</div>
</section>
<section id="quickstart">
<h2>Quick start: representative papers</h2>
<p>
These are not the only important papers in the survey. They are included here as a first reading path for readers
who want to understand the landscape quickly before diving into the full library.
</p>
<div class="card-grid">
<div class="ref-card"><strong><a href="https://arxiv.org/abs/2212.10560" rel="noopener" target="_blank">Self-Instruct</a></strong><div class="meta">2023 · 1.1 Intrinsic Generative Demonstrations</div><p>A canonical example of self-synthetic instruction generation for model alignment.</p></div><div class="ref-card"><strong><a href="https://arxiv.org/abs/2212.08073" rel="noopener" target="_blank">Constitutional AI</a></strong><div class="meta">2022 · 1.2 Intrinsic Evaluative Feedback</div><p>An influential AI-feedback framework where model-based judgments are used for alignment and policy optimization.</p></div><div class="ref-card"><strong><a href="https://arxiv.org/abs/2411.02337" rel="noopener" target="_blank">WebRL</a></strong><div class="meta">2025 · 1.3.1 Grounded Executable Environments</div><p>A representative web-agent training setup built around reinforcement learning and evolving curricula.</p></div><div class="ref-card"><strong><a href="https://arxiv.org/abs/2410.13232" rel="noopener" target="_blank">Web Agents with World Models</a></strong><div class="meta">2025 · 1.3.2 Generative World Models</div><p>A concrete example of using learned environment dynamics to improve web agents.</p></div><div class="ref-card"><strong><a href="https://arxiv.org/abs/2303.17651" rel="noopener" target="_blank">Self-Refine</a></strong><div class="meta">2023 · 2.1.2 Qualitative-Feedback Refinement</div><p>A classic self-feedback loop for iterative improvement at the prompt/output level.</p></div><div class="ref-card"><strong><a href="https://arxiv.org/abs/2406.07496" rel="noopener" target="_blank">TextGrad</a></strong><div class="meta">2025 · 2.1.4 Textual Gradient Optimization</div><p>An influential formulation of automatic optimization via textual gradients.</p></div><div class="ref-card"><strong><a href="https://arxiv.org/abs/2305.10250" rel="noopener" target="_blank">MemoryBank</a></strong><div class="meta">2024 · 2.2.2 Memory Structure</div><p>A representative long-term memory architecture for LLM-based agents.</p></div><div class="ref-card"><strong><a href="https://arxiv.org/abs/2305.16291" rel="noopener" target="_blank">Voyager</a></strong><div class="meta">2023 · 2.3.1 Dynamic Tool Routing</div><p>An open-ended embodied agent that accumulates skills and uses tools in a growing scaffold.</p></div><div class="ref-card"><strong><a href="https://arxiv.org/abs/2505.22954" rel="noopener" target="_blank">Darwin Godel Machine</a></strong><div class="meta">2025 · 2.4 Full Scaffolding</div><p>A strong representative for recursive self-improving agents that modify their broader operating logic.</p></div>
</div>
</section>
<section id="library">
<h2>Curated paper library</h2>
<p>The literature below is organized to match the survey taxonomy across three branches: foundation model improvement, scaffolding improvement, and evaluation & benchmarking. Within each subsection, entries are presented in roughly chronological order.</p>
<div class="toolbar">
<div class="search-wrap">
<input aria-label="Search papers" id="paperSearch" placeholder="Search by title, venue, or year…" type="search"/>
</div>
<div class="chip-group">
<button class="chip active" data-filter="all" type="button">All</button>
<button class="chip" data-filter="fm" type="button">Foundation model</button>
<button class="chip" data-filter="scaffolding" type="button">Scaffolding</button><button class="chip" data-filter="evaluation" type="button">Evaluation</button>
<button class="chip" data-filter="recent" type="button">2025+</button>
<button class="chip" data-filter="code" type="button">Has code</button>
</div>
<div class="toolbar-meta"><span id="visibleCount">312</span> visible library entries</div>
</div>
<div class="library-block" id="sgd">
<div class="block-head">
<div>
<h3>1.1 Intrinsic Generative Demonstrations</h3>
<p>The agent or model improves by synthesizing demonstrations, instruction sets, reasoning traces, or task distributions that can be used for imitation-style parameter updates.</p>
</div>
<div class="count-badge">23 papers</div>
</div>
<div class="table-wrap">
<table class="paper-table">
<thead>
<tr><th>Year</th><th>Title</th><th>Venue</th><th>Links</th></tr>
</thead>
<tbody>
<tr class="paper-row" data-branch="fm" data-search="2023 self-instruct: aligning language models with self-generated instructions acl" data-year="2023">
<td class="year">2023</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2212.10560" rel="noopener" target="_blank">Self-Instruct: Aligning Language Models with Self-Generated Instructions</a>
</td>
<td>ACL</td>
<td class="links"><a href="https://arxiv.org/abs/2212.10560" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/yizhongw/self-instruct" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2023 large language models can self-improve emnlp" data-year="2023">
<td class="year">2023</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2210.11610" rel="noopener" target="_blank">Large Language Models Can Self-Improve</a>
</td>
<td>EMNLP</td>
<td class="links"><a href="https://arxiv.org/abs/2210.11610" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2023 orca: progressive learning from complex explanation traces of gpt-4 arxiv" data-year="2023">
<td class="year">2023</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2306.02707" rel="noopener" target="_blank">Orca: Progressive Learning from Complex Explanation Traces of GPT-4</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2306.02707" rel="noopener" target="_blank">Paper</a> · <a href="https://aka.ms/orca-lm" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2024 self: self-evolution with language feedback arxiv" data-year="2024">
<td class="year">2024</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2310.00533" rel="noopener" target="_blank">SELF: Self-Evolution with Language Feedback</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2310.00533" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2024 self-guide: better task-specific instruction following via self-synthetic finetuning colm" data-year="2024">
<td class="year">2024</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2407.12874" rel="noopener" target="_blank">SELF-GUIDE: Better Task-Specific Instruction Following via Self-Synthetic Finetuning</a>
</td>
<td>COLM</td>
<td class="links"><a href="https://arxiv.org/abs/2407.12874" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/zhaochenyang20/Prompt2Model-Self-Guide" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 improving model alignment through collective intelligence of open-source llms icml" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2505.03059" rel="noopener" target="_blank">Improving Model Alignment Through Collective Intelligence of Open-Source LLMS</a>
</td>
<td>ICML</td>
<td class="links"><a href="https://arxiv.org/abs/2505.03059" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 superficial self-improved reasoners benefit from model merging emnlp" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2503.02103" rel="noopener" target="_blank">Superficial Self-Improved Reasoners Benefit from Model Merging</a>
</td>
<td>EMNLP</td>
<td class="links"><a href="https://arxiv.org/abs/2503.02103" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/xiangchi-yuan/merge_syn" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 will pre-training ever end? a first step toward next-generation foundation mllms via self-improving systematic cognition arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2503.12303" rel="noopener" target="_blank">Will Pre-Training Ever End? A First Step Toward Next-Generation Foundation MLLMs via Self-Improving Systematic Cognition</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2503.12303" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/thunlp/SICOG?tab=readme-ov-file" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 taskcraft: automated generation of agentic tasks arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2506.10055" rel="noopener" target="_blank">TaskCraft: Automated Generation of Agentic Tasks</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2506.10055" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/OPPO-PersonalAI/TaskCraft" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 iterative tool usage exploration for multimodal agents via step-wise preference tuning neurips" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2504.21561" rel="noopener" target="_blank">Iterative Tool Usage Exploration for Multimodal Agents via Step-wise Preference Tuning</a>
</td>
<td>NeurIPS</td>
<td class="links"><a href="https://arxiv.org/abs/2504.21561" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/SPORT-Agents/SPORT-Agents" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 maximizing confidence alone improves reasoning arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2505.22660" rel="noopener" target="_blank">Maximizing Confidence Alone Improves Reasoning</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2505.22660" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/satrams/rent-rl" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 dive: diversified iterative self-improvement arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2501.00747" rel="noopener" target="_blank">DIVE: Diversified Iterative Self-Improvement</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2501.00747" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/qinyiwei/DIVE" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 self-adapting language models neurips" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2506.10943" rel="noopener" target="_blank">Self-Adapting Language Models</a>
</td>
<td>NeurIPS</td>
<td class="links"><a href="https://arxiv.org/abs/2506.10943" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/Continual-Intelligence/SEAL" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 first sft, second rl, third upt: continual improving multi-modal llm reasoning via unsupervised post-training neurips" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/pdf/2505.22453" rel="noopener" target="_blank">First SFT, Second RL, Third UPT: Continual Improving Multi-Modal LLM Reasoning via Unsupervised Post-Training</a>
</td>
<td>NeurIPS</td>
<td class="links"><a href="https://arxiv.org/pdf/2505.22453" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/waltonfuture/MM-UPT" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 ladder: self-improving llms through recursive problem decomposition arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2503.00735" rel="noopener" target="_blank">LADDER: Self-Improving LLMs Through Recursive Problem Decomposition</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2503.00735" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 self-consistency preference optimization icml" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2411.04109" rel="noopener" target="_blank">Self-Consistency Preference Optimization</a>
</td>
<td>ICML</td>
<td class="links"><a href="https://arxiv.org/abs/2411.04109" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 adapting while learning: grounding llms for scientific problems with tool usage adaptation icml" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2411.00412" rel="noopener" target="_blank">Adapting While Learning: Grounding LLMs for Scientific Problems with Tool Usage Adaptation</a>
</td>
<td>ICML</td>
<td class="links"><a href="https://arxiv.org/abs/2411.00412" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/Rose-STL-Lab/Adapting-While-Learning" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 reinforcing general reasoning without verifiers iclr" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2505.21493" rel="noopener" target="_blank">Reinforcing General Reasoning Without Verifiers</a>
</td>
<td>ICLR</td>
<td class="links"><a href="https://arxiv.org/abs/2505.21493" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/sail-sg/VeriFree" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 sage: multi-agent self-evolution for llm reasoning arxiv" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2603.15255" rel="noopener" target="_blank">SAGE: Multi-Agent Self-Evolution for LLM Reasoning</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2603.15255" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 andes: agent native data evolving synthesis tool for autonomous instruction alignment arxiv" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2606.01279" rel="noopener" target="_blank">ANDES: Agent Native Data Evolving Synthesis Tool for Autonomous Instruction Alignment</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2606.01279" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/zzy1127/ANDES" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 evoground: self-evolving video agents for video temporal grounding arxiv" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2605.13803" rel="noopener" target="_blank">EvoGround: Self-Evolving Video Agents for Video Temporal Grounding</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2605.13803" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/minjoong507/EvoGround" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 from self-evolving synthetic data to verifiable-reward rl: post-training multi-turn interactive tool-using agents arxiv areal-sea" data-year="2026"><td class="year">2026</td><td class="title-cell"><a href="https://arxiv.org/abs/2601.22607" rel="noopener" target="_blank">From Self-Evolving Synthetic Data to Verifiable-Reward RL: Post-Training Multi-turn Interactive Tool-Using Agents</a></td><td>arXiv</td><td class="links"><a href="https://arxiv.org/abs/2601.22607" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td></tr><tr class="paper-row" data-branch="fm" data-search="2026 ragshaper: eliciting sophisticated agentic rag skills via automated data synthesis arxiv ragshaper" data-year="2026"><td class="year">2026</td><td class="title-cell"><a href="https://arxiv.org/abs/2601.08699" rel="noopener" target="_blank">RAGShaper: Eliciting Sophisticated Agentic RAG Skills via Automated Data Synthesis</a></td><td>arXiv</td><td class="links"><a href="https://arxiv.org/abs/2601.08699" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td></tr></tbody>
</table>
</div>
</div>
<div class="library-block" id="evaluative-feedback">
<div class="block-head">
<div>
<h3>1.2 Intrinsic Evaluative Feedback</h3>
<p>The system derives its own reward, critique, verification signal, or intrinsic supervision to guide further updates.</p>
</div>
<div class="count-badge">21 papers</div>
</div>
<div class="table-wrap">
<table class="paper-table">
<thead>
<tr><th>Year</th><th>Title</th><th>Venue</th><th>Links</th></tr>
</thead>
<tbody>
<tr class="paper-row" data-branch="fm" data-search="2025 strive: structured reasoning for self-improvement in claim verification mir" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2502.11959" rel="noopener" target="_blank">STRIVE: Structured Reasoning for Self-Improvement in Claim Verification</a>
</td>
<td>MIR</td>
<td class="links"><a href="https://arxiv.org/abs/2502.11959" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 beyond accuracy: the role of calibration in self-improving large language models arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2504.02902" rel="noopener" target="_blank">Beyond Accuracy: The Role of Calibration in Self-Improving Large Language Models</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2504.02902" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2022 constitutional ai: harmlessness from ai feedback arxiv" data-year="2022">
<td class="year">2022</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2212.08073" rel="noopener" target="_blank">Constitutional AI: Harmlessness from AI Feedback</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2212.08073" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/anthropics/ConstitutionalHarmlessnessPaper?tab=readme-ov-file" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2023 rest meets react: self-improvement for multi-step reasoning llm agent arxiv" data-year="2023">
<td class="year">2023</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2312.10003" rel="noopener" target="_blank">ReST meets ReAct: Self-Improvement for Multi-Step Reasoning LLM Agent</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2312.10003" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 self-evolved reward learning for llms iclr" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2411.00418" rel="noopener" target="_blank">Self-Evolved Reward Learning for LLMs</a>
</td>
<td>ICLR</td>
<td class="links"><a href="https://arxiv.org/abs/2411.00418" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/microsoft/DKI_LLM/tree/main/SER" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 sample, predict, then proceed: self-verification sampling for tool use of llms arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2506.02918v1" rel="noopener" target="_blank">Sample, Predict, then Proceed: Self-Verification Sampling for Tool Use of LLMs</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2506.02918v1" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 rlsr: reinforcement learning from self reward arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2505.08827" rel="noopener" target="_blank">RLSR: Reinforcement Learning from Self Reward</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2505.08827" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 right question is already half the answer: fully unsupervised llm reasoning incentivization neurips" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2504.05812" rel="noopener" target="_blank">Right Question is Already Half the Answer: Fully Unsupervised LLM Reasoning Incentivization</a>
</td>
<td>NeurIPS</td>
<td class="links"><a href="https://arxiv.org/abs/2504.05812" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/QingyangZhang/EMPO" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 ttrl: test-time reinforcement learning neurips" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2504.16084" rel="noopener" target="_blank">TTRL: Test-Time Reinforcement Learning</a>
</td>
<td>NeurIPS</td>
<td class="links"><a href="https://arxiv.org/abs/2504.16084" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/PRIME-RL/TTRL" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 can large reasoning models self-train? arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2505.21444" rel="noopener" target="_blank">Can Large Reasoning Models Self-Train?</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2505.21444" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/tajwarfahim/srt" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 self rewarding self improving arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2505.08827v1" rel="noopener" target="_blank">Self Rewarding Self Improving</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2505.08827v1" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 self-evolving curriculum for llm reasoning arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2505.14970" rel="noopener" target="_blank">Self-Evolving Curriculum for LLM Reasoning</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2505.14970" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/ServiceNow/sec" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 reflect, retry, reward: self-improving llms via reinforcement learning arxiv" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2505.24726" rel="noopener" target="_blank">Reflect, Retry, Reward: Self-Improving LLMs via Reinforcement Learning</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2505.24726" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2025 adaptive self-improvement llm agentic system for ml library development icml" data-year="2025">
<td class="year">2025</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2502.02534" rel="noopener" target="_blank">Adaptive Self-improvement LLM Agentic System for ML Library Development</a>
</td>
<td>ICML</td>
<td class="links"><a href="https://arxiv.org/abs/2502.02534" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/zhang677/PCL-lite" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 learning to reason without external rewards iclr" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2505.19590" rel="noopener" target="_blank">Learning to Reason without External Rewards</a>
</td>
<td>ICLR</td>
<td class="links"><a href="https://arxiv.org/abs/2505.19590" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/sunblaze-ucb/Intuitor" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 structured reasoning for large language models arxiv" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2601.07180" rel="noopener" target="_blank">Structured Reasoning for Large Language Models</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2601.07180" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 ireasoner: trajectory-aware intrinsic reasoning supervision for self-evolving large multimodal models arxiv" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2601.07180" rel="noopener" target="_blank">iReasoner: Trajectory-Aware Intrinsic Reasoning Supervision for Self-Evolving Large Multimodal Models</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2601.07180" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/meghanaasunil/iReasoner" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 strive: structured reasoning for self-improvement in claim verification machine intelligence research" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://link.springer.com/article/10.1007/s11633-025-1598-5" rel="noopener" target="_blank">STRIVE: Structured Reasoning for Self-improvement in Claim Verification</a>
</td>
<td>machine intelligence research</td>
<td class="links"><a href="https://link.springer.com/article/10.1007/s11633-025-1598-5" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 unicorn: towards self-improving unified multimodal models through self-generated supervision arxiv" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2601.03193" rel="noopener" target="_blank">UniCorn: Towards Self-Improving Unified Multimodal Models through Self-Generated Supervision</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2601.03193" rel="noopener" target="_blank">Paper</a> · <a href="https://github.com/Hungryyan1/UniCorn" rel="noopener" target="_blank">Code</a></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 retrospective progress-aware self-refinement for llm agent training arxiv" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2606.14302" rel="noopener" target="_blank">Retrospective Progress-Aware Self-Refinement for LLM Agent Training</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2606.14302" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
<tr class="paper-row" data-branch="fm" data-search="2026 eve-agent: evidence-verifiable self-evolving agents arxiv" data-year="2026">
<td class="year">2026</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2605.22905" rel="noopener" target="_blank">EVE-Agent: Evidence-Verifiable Self-Evolving Agents</a>
</td>
<td>arXiv</td>
<td class="links"><a href="https://arxiv.org/abs/2605.22905" rel="noopener" target="_blank">Paper</a> · <span class="muted-link">Code</span></td>
</tr>
</tbody>
</table>
</div>
</div>
<div class="library-block" id="experience">
<div class="block-head">
<div>
<h3>1.3 Extrinsic Exploratory Experience</h3>
<p>The agent improves through trajectories gathered from interaction with environments or learned simulators.</p>
</div>
<div class="count-badge">33 papers</div>
</div>
<details class="subgroup" open="">
<summary id="experience-env">
<span>1.3.1 Grounded Executable Environments</span>
<span class="count-badge small">18 papers</span>
</summary>
<p class="sub-desc">Learning from trajectories, rewards, or observations obtained by acting in executable environments such as code runtimes, web interfaces, games, GUI systems, or robotics simulators.</p>
<div class="table-wrap">
<table class="paper-table">
<thead>
<tr><th>Year</th><th>Title</th><th>Venue</th><th>Links</th></tr>
</thead>
<tbody>
<tr class="paper-row" data-branch="fm" data-search="2023 robocat: a self-improving generalist agent for robotic manipulation tmlr" data-year="2023">
<td class="year">2023</td>
<td class="title-cell">
<a href="https://arxiv.org/abs/2306.11706" rel="noopener" target="_blank">RoboCat: A Self-Improving Generalist Agent for Robotic Manipulation</a>
</td>