summaryrefslogtreecommitdiffstats
path: root/2023/captions/emacsconf-2023-llm--llm-clients-in-emacs-functionality-and-standardization--andrew-hyatt--answers.vtt
blob: d48cc06dd6884f4e87339b66809015c362251b61 (plain) (blame)
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
WEBVTT


00:00:13.099 --> 00:00:13.599
[Speaker 0]: Okay. Hello, everyone.

00:00:16.560 --> 00:00:17.060
I think this is the start of the Q&A session.

00:00:25.119 --> 00:00:25.599
So people can just ask me questions here.

00:00:28.259 --> 00:00:28.380
Or I think maybe these questions are going to

00:00:30.560 --> 00:00:30.980
be read by someone. Yes,

00:00:34.680 --> 00:00:35.060
thank you. Should I start doing that?

00:00:39.280 --> 00:00:39.400
I also know that there's questions in the

00:00:41.320 --> 00:00:41.480
either pad room, so I could start out

00:00:42.280 --> 00:00:42.780
answering those as well.

00:00:45.020 --> 00:00:45.300
[Speaker 1]: Right, sure. Whichever way you prefer.

00:00:46.860 --> 00:00:47.220
If you prefer to read the questions yourself,

00:00:48.940 --> 00:00:49.080
by all means, or if you would prefer me to

00:00:50.080 --> 00:00:50.280
read them to you, that also works.

00:00:50.580 --> 00:00:50.920
[Speaker 2]: Oh, I see.

00:00:51.600 --> 00:00:51.760
[Speaker 0]: Why don't you read them to me?

00:00:53.260 --> 00:00:53.760
I think it'll just be more interesting then.

00:00:56.820 --> 00:00:57.160
[Speaker 1]: Sure. OK, let's see. The first question is,

00:00:58.360 --> 00:00:58.860
what is your use case for embedding,

00:01:00.060 --> 00:01:00.560
mainly for searching?

00:01:06.180 --> 00:01:06.340
[Speaker 0]: Yeah, I mean, I think the use case really is

00:01:12.100 --> 00:01:12.320
searching. And I think it is very useful when

00:01:15.060 --> 00:01:15.320
you're searching for something in a vague

00:01:18.280 --> 00:01:18.780
way. Just to give you an example,

00:01:23.860 --> 00:01:24.360
I have a note system called EKG.

00:01:25.760 --> 00:01:26.260
I type all my notes on it.

00:01:28.620 --> 00:01:29.120
You can find it on GitHub and Melba.

00:01:34.140 --> 00:01:34.400
But I wrote something at some point a year

00:01:35.840 --> 00:01:36.020
ago or something. I wrote something that I

00:01:36.600 --> 00:01:36.980
just vaguely remembered.

00:01:38.800 --> 00:01:38.940
Oh, this was about a certain kind of

00:01:41.580 --> 00:01:41.760
communication. I wanted communicating to

00:01:43.280 --> 00:01:43.479
large audiences. There's some interesting tip

00:01:44.700 --> 00:01:45.060
that I wrote down that was really cool.

00:01:49.920 --> 00:01:50.080
And I was like, well, I need to find it.

00:01:52.260 --> 00:01:52.640
So I did an embedding search for something

00:01:55.479 --> 00:01:55.979
like, you know, tips for communicating.

00:01:58.979 --> 00:01:59.100
Like those words may not have been in what I

00:02:00.020 --> 00:02:00.520
was trying to find at all,

00:02:02.680 --> 00:02:03.180
But it was able to find it.

00:02:05.840 --> 00:02:06.260
And that is something that's very hard to do

00:02:07.200 --> 00:02:07.360
in other ways. Like, you know,

00:02:08.520 --> 00:02:08.720
if you had to do this with normal search,

00:02:09.199 --> 00:02:09.660
you have to do synonyms.

00:02:10.940 --> 00:02:11.200
And like maybe those synonyms wouldn't cover

00:02:11.960 --> 00:02:12.340
it. Like with embedding,

00:02:13.940 --> 00:02:14.160
you can basically get at like the vague

00:02:14.960 --> 00:02:15.460
sentiment. You're like,

00:02:17.320 --> 00:02:17.560
you know, you're, you know,

00:02:19.520 --> 00:02:19.700
you can really query on like what things are

00:02:21.760 --> 00:02:22.260
about as opposed to what words they have.

00:02:25.600 --> 00:02:26.100
Also, it's super good for similarity search.

00:02:27.720 --> 00:02:27.900
So you could say, look,

00:02:30.040 --> 00:02:30.240
I have a bunch of things that are encoded

00:02:31.400 --> 00:02:31.900
with embeddings that I want to show.

00:02:34.120 --> 00:02:34.280
For example, you can make an embedding for

00:02:35.220 --> 00:02:35.380
every buffer. You'd be like,

00:02:37.060 --> 00:02:37.200
well, show me buffers that are similar to

00:02:38.740 --> 00:02:38.980
this buffer. That doesn't sound super useful,

00:02:40.440 --> 00:02:40.940
but this is the kind of thing you could do.

00:02:45.300 --> 00:02:45.480
And so if you have a bunch of notes or

00:02:46.720 --> 00:02:46.920
something else that you want to search on,

00:02:48.240 --> 00:02:48.740
you'd be like, what's similar to this buffer?

00:02:51.500 --> 00:02:51.760
Or what notes are similar to each other?

00:02:53.040 --> 00:02:53.540
What buffers are similar to each other?

00:02:55.380 --> 00:02:55.880
It's super good for this sort of thing.

00:03:00.780 --> 00:03:01.280
And it's also good for this kind of retrieval

00:03:03.080 --> 00:03:03.520
augmented generation, where you sort of,

00:03:05.080 --> 00:03:05.280
you retrieve things and the purpose is not

00:03:06.600 --> 00:03:06.880
for you to see them, but then you pass that

00:03:12.040 --> 00:03:12.180
to the LLM. And then it's able to be a little

00:03:14.340 --> 00:03:14.800
bit more accurate because it has the actual

00:03:15.760 --> 00:03:16.260
text that you're trying to,

00:03:18.960 --> 00:03:19.180
that is relevant, and it can cite from and

00:03:20.720 --> 00:03:20.820
things like that. And then it could give you

00:03:22.260 --> 00:03:22.660
a much better answer that's kind of,

00:03:25.520 --> 00:03:25.680
you know, not just from its own little neural

00:03:26.320 --> 00:03:26.820
nets and memory.

00:03:31.920 --> 00:03:32.120
[Speaker 1]: Cool, thanks. Let's see,

00:03:35.740 --> 00:03:36.100
next question. What do you think about embed

00:03:40.160 --> 00:03:40.660
Emacs manual versus GPT's Emacs manual?

00:03:45.480 --> 00:03:45.660
[Speaker 0]: I'm not exactly sure what this question is

00:03:46.980 --> 00:03:47.480
trying to say. So I mean,

00:03:51.000 --> 00:03:51.140
if someone wrote that and wants to expand on

00:03:55.080 --> 00:03:55.320
it a little bit, but I think that maybe

00:03:58.420 --> 00:03:58.920
you're saying like you could embed,

00:04:00.280 --> 00:04:00.780
have embeddings for like various,

00:04:02.520 --> 00:04:02.720
like every paragraph or something of the

00:04:04.540 --> 00:04:04.680
Emacs manual. But it's also the case that

00:04:06.500 --> 00:04:07.000
like GPT is already for sure already read it,

00:04:09.760 --> 00:04:09.960
right? And so you could ask questions that

00:04:13.460 --> 00:04:13.780
are about Emacs and our ELISP or whatever

00:04:15.200 --> 00:04:15.700
part of the manual you want to find.

00:04:19.760 --> 00:04:20.240
And it will do a reasonably good job,

00:04:22.280 --> 00:04:22.780
especially the better models will do a

00:04:24.620 --> 00:04:24.940
reasonably good job of saying you something

00:04:26.040 --> 00:04:26.540
that is vaguely accurate.

00:04:29.440 --> 00:04:29.860
But if you do this retrieval augmented

00:04:30.580 --> 00:04:31.080
generation with embeddings,

00:04:32.640 --> 00:04:33.140
you can get something that is very accurate.

00:04:36.700 --> 00:04:36.960
At least I think. I haven't tried it,

00:04:38.760 --> 00:04:39.020
but this is a technique that works in other

00:04:43.040 --> 00:04:43.260
similar cases. So you can also imagine like,

00:04:44.320 --> 00:04:44.500
oh, this whole thing I said,

00:04:47.860 --> 00:04:48.120
like, oh, you can query for vague things and

00:04:49.140 --> 00:04:49.600
get parts of the manual,

00:04:52.680 --> 00:04:53.000
perhaps. I'm not exactly sure if that would

00:04:55.120 --> 00:04:55.520
be useful, but maybe. Usually when I'm

00:04:57.040 --> 00:04:57.180
looking things up in the Emacs manual or

00:04:58.320 --> 00:04:58.780
Elist manual, I have something extremely

00:05:00.020 --> 00:05:00.300
specific and I kind of know where to look.

00:05:02.960 --> 00:05:03.080
But having other ways to get at this

00:05:04.000 --> 00:05:04.500
information is always good.

00:05:10.240 --> 00:05:10.740
[Speaker 1]: Right. Looks like they added a clarification

00:05:12.280 --> 00:05:12.720
if you would like to read that yourself,

00:05:14.180 --> 00:05:14.680
or would you like me to read it for you?

00:05:17.640 --> 00:05:18.140
[Speaker 0]: Yeah. Yes, OK. It says,

00:05:20.460 --> 00:05:20.740
I've never tried. Yeah,

00:05:21.500 --> 00:05:21.820
the question is like OK,

00:05:23.100 --> 00:05:23.240
there is a difference between the kind of

00:05:23.860 --> 00:05:24.360
thing as I just described.

00:05:26.200 --> 00:05:26.600
I have not tried the difference with the EMAX

00:05:31.560 --> 00:05:31.980
manual itself. It'd be interesting to see

00:05:33.700 --> 00:05:33.960
what this is, but I would expect like these

00:05:35.140 --> 00:05:35.600
techniques, the retrieval augmented

00:05:38.840 --> 00:05:39.340
generation is generally pretty good.

00:05:41.240 --> 00:05:41.740
And I suspect it would,

00:05:43.580 --> 00:05:43.780
I would bet money on the fact that it's gonna

00:05:45.820 --> 00:05:46.240
give you, you know, better results than just,

00:05:48.160 --> 00:05:48.360
you know, doing a free form query without any

00:05:49.440 --> 00:05:49.940
retrieval augmented generation.

00:05:54.240 --> 00:05:54.640
[Speaker 1]: Cool. Let's see. Next question.

00:05:56.380 --> 00:05:56.880
When deferring commit messages to an LLM,

00:05:59.700 --> 00:05:59.920
what, if anything, do you find you might have

00:06:02.940 --> 00:06:03.440
[Speaker 0]: lost? Yeah, it's a good question.

00:06:06.060 --> 00:06:06.560
When deferring anything to a computer,

00:06:08.860 --> 00:06:09.360
like, you know, I used to have to remember

00:06:11.200 --> 00:06:11.700
how to get places, and now,

00:06:14.540 --> 00:06:15.040
you know, on the few occasions which I drive,

00:06:16.560 --> 00:06:16.720
like, It could just tell me how to get

00:06:21.960 --> 00:06:22.280
places. So similar things could occur here

00:06:24.960 --> 00:06:25.460
where like, okay, I'm just leaving the LLM.

00:06:27.380 --> 00:06:27.680
And so I'm kind of missing out on some

00:06:30.040 --> 00:06:30.220
opportunity to think coherently about a

00:06:32.440 --> 00:06:32.680
particular commit. Particular commits are

00:06:36.140 --> 00:06:36.540
kind of low level. I don't think it's usually

00:06:39.340 --> 00:06:39.840
relatively obvious and what they're doing.

00:06:42.600 --> 00:06:42.800
And in this case, I think there's not much

00:06:44.220 --> 00:06:44.540
loss. But for sure, in other cases,

00:06:46.400 --> 00:06:46.900
if you're starting to get into situations

00:06:48.640 --> 00:06:48.800
where it's writing your emails and all this

00:06:52.920 --> 00:06:53.300
stuff. First of all, it's in 1 sense,

00:06:55.580 --> 00:06:56.040
I'm not sure you might be losing something by

00:06:57.520 --> 00:06:57.780
delegating things. On the other hand,

00:06:59.120 --> 00:06:59.280
you know, when you're interacting with these

00:07:01.280 --> 00:07:01.560
LLMs, you have to be extremely specific about

00:07:03.120 --> 00:07:03.240
what you want, or else it's just not going to

00:07:07.540 --> 00:07:07.680
do a good job. And that might actually be a

00:07:09.440 --> 00:07:09.860
good thing. So the question might be that

00:07:11.820 --> 00:07:12.240
maybe you might gain things by using an LLM

00:07:13.860 --> 00:07:14.060
to do your work. It might not actually even

00:07:15.060 --> 00:07:15.400
save you that much time,

00:07:18.480 --> 00:07:18.640
at least initially, because you have to kind

00:07:20.460 --> 00:07:20.660
of practice again super specific about what

00:07:22.740 --> 00:07:22.900
you want to get out of the output it's going

00:07:26.940 --> 00:07:26.980
to give you so like oh I'm you know maybe you

00:07:29.600 --> 00:07:29.820
know you're on the emacs devel mailing list

00:07:31.780 --> 00:07:31.980
and you're like okay write this email about

00:07:34.000 --> 00:07:34.140
this about this And here's what I want to

00:07:35.370 --> 00:07:35.460
say. And here's the kind of tone I want to

00:07:36.020 --> 00:07:36.420
use. And here's the like,

00:07:37.660 --> 00:07:38.160
oh, you might want to specify like everything

00:07:39.620 --> 00:07:40.120
that you kind of want to get into this.

00:07:42.180 --> 00:07:42.680
Usually it's easier just to write the email.

00:07:45.600 --> 00:07:46.100
But I think that practice of kind of

00:07:48.080 --> 00:07:48.420
understanding what you want is not something

00:07:52.680 --> 00:07:53.180
you normally do. And I think it's going to be

00:07:56.040 --> 00:07:56.480
an interesting exercise that will help people

00:07:57.280 --> 00:07:57.540
understand. That said,

00:07:58.860 --> 00:07:59.020
I haven't done that much of that,

00:07:59.900 --> 00:08:00.040
so I can't say, oh, yeah,

00:08:01.080 --> 00:08:01.440
I've done this and it works for me.

00:08:03.000 --> 00:08:03.120
Maybe. I think it's an interesting thing to

00:08:03.120 --> 00:08:03.620
explore.

00:08:07.720 --> 00:08:08.220
[Speaker 1]: Sure. Thanks. Let's see.

00:08:10.840 --> 00:08:11.140
Let's see. Can you share your font settings

00:08:13.440 --> 00:08:13.740
in your Emacs config? Those are some nice

00:08:14.200 --> 00:08:14.700
fonts for reading.

00:08:18.900 --> 00:08:19.200
[Speaker 0]: Yeah, I think I was using Menlo at the time.

00:08:20.840 --> 00:08:20.980
Unfortunately, I don't save those kinds of

00:08:21.940 --> 00:08:22.200
things, like a history of this.

00:08:24.000 --> 00:08:24.500
I've kind of switched now to,

00:08:27.340 --> 00:08:27.540
what was that? I think I wrote it down in

00:08:29.440 --> 00:08:29.940
the, I switched to MunaSpace,

00:08:31.920 --> 00:08:32.420
which just came out like a week or 2 ago,

00:08:33.340 --> 00:08:33.840
and is also pretty cool.

00:08:35.440 --> 00:08:35.940
So I think it's Menlo.

00:08:37.380 --> 00:08:37.760
The internal question,

00:08:38.400 --> 00:08:38.900
what font are you using?

00:08:42.020 --> 00:08:42.340
[Speaker 1]: Indeed, yeah. It looks like someone guessed

00:08:43.780 --> 00:08:44.280
as well that it might be Menlo.

00:08:47.680 --> 00:08:48.040
OK, Cool. Yeah, next question.

00:08:48.900 --> 00:08:49.400
In terms of standardization,

00:08:53.260 --> 00:08:53.520
do you see a need for the medium to large

00:08:55.840 --> 00:08:56.160
scale effort needed? And then they also

00:08:56.960 --> 00:08:57.460
elaborate about it.

00:09:03.600 --> 00:09:04.100
[Speaker 0]: Yeah, I mean, I do think,

00:09:06.040 --> 00:09:06.300
I don't know if it's large scale,

00:09:08.000 --> 00:09:08.500
but at least it's probably medium scale.

00:09:10.320 --> 00:09:10.520
There's a lot of things that are missing that

00:09:12.260 --> 00:09:12.400
we don't have right now in emacs when you're

00:09:13.660 --> 00:09:14.160
dealing with LLMs. 1 is,

00:09:18.240 --> 00:09:18.420
a prompting system. And by that,

00:09:21.820 --> 00:09:22.020
I mean, you know, prompts are just like big

00:09:24.520 --> 00:09:24.720
blocks of text, but there's also senses that

00:09:28.260 --> 00:09:28.420
like prompts need to be composable and you

00:09:30.480 --> 00:09:30.660
need to be able to iterate on parts of the

00:09:36.600 --> 00:09:37.100
prompt. And so it's also customizable.

00:09:38.940 --> 00:09:39.060
Users might want to customize it.

00:09:41.260 --> 00:09:41.360
On the other hand, it's not super easy to

00:09:43.820 --> 00:09:44.320
write the prompt. So you want really good

00:09:47.900 --> 00:09:48.040
defaults. So the whole prompt system is kind

00:09:51.360 --> 00:09:51.460
of complicated. That needs to be kind of

00:09:52.580 --> 00:09:52.760
standardized, because I don't think there's

00:09:54.380 --> 00:09:54.720
any tools for doing something like that right

00:09:58.380 --> 00:09:58.880
now. I personally use my system,

00:10:00.220 --> 00:10:00.600
my note system for EKG.

00:10:01.720 --> 00:10:01.920
I don't think that's appropriate for

00:10:02.800 --> 00:10:03.120
everyone, but it does,

00:10:04.480 --> 00:10:04.640
I did write it to have some of these

00:10:06.540 --> 00:10:06.760
capabilities of composability that I think

00:10:08.360 --> 00:10:08.860
are useful for a prompt generation.

00:10:11.940 --> 00:10:12.280
It'd be nice to have a system like that,

00:10:15.660 --> 00:10:16.160
but for general use. I don't,

00:10:17.840 --> 00:10:18.060
this is something I've been meaning to think

00:10:18.840 --> 00:10:19.000
about, like how to do it,

00:10:19.760 --> 00:10:19.920
but like this, you know,

00:10:21.260 --> 00:10:21.660
if someone's interested in getting this area,

00:10:26.120 --> 00:10:26.420
like, I would love to chat about that or,

00:10:27.600 --> 00:10:27.980
you know, I think there's a lot of

00:10:31.020 --> 00:10:31.260
interesting ideas that we could have to have

00:10:34.080 --> 00:10:34.540
a system that allows us to make progress

00:10:38.860 --> 00:10:39.360
here. And also, I think there's more to

00:10:40.520 --> 00:10:40.900
standardization to be done.

00:10:42.820 --> 00:10:43.140
1 thing I'd also like to see that we haven't

00:10:47.020 --> 00:10:47.220
done yet is a system for standardizing on

00:10:48.060 --> 00:10:48.560
getting structured output.

00:10:49.640 --> 00:10:50.140
This is gonna be super useful.

00:10:52.280 --> 00:10:52.780
I have this for open AIs API,

00:10:53.560 --> 00:10:54.060
cause they support it.

00:10:55.940 --> 00:10:56.040
And it's really nice, cause then you can

00:10:57.440 --> 00:10:57.660
write elist functions that like,

00:10:59.380 --> 00:10:59.880
okay, I'm going to call the LLM.

00:11:00.760 --> 00:11:01.000
I'm gonna get structured output.

00:11:02.040 --> 00:11:02.160
I know what that structure is going to be.

00:11:03.480 --> 00:11:03.680
It's not going to be just a big block of

00:11:05.660 --> 00:11:06.040
text. I could turn it into a,

00:11:07.000 --> 00:11:07.480
you know, a P list or something.

00:11:09.280 --> 00:11:09.480
And then I could get the values out of that P

00:11:11.880 --> 00:11:12.260
list. And I know that way I could do,

00:11:14.220 --> 00:11:14.720
I could write actual apps that are,

00:11:18.300 --> 00:11:18.720
you know, very, very sort of,

00:11:20.200 --> 00:11:20.680
you know, useful for very specific purposes

00:11:21.900 --> 00:11:22.400
and not just for text generation.

00:11:24.000 --> 00:11:24.320
And I think that's 1 of the most important

00:11:27.100 --> 00:11:27.540
things we want to do. And I have some ideas

00:11:28.840 --> 00:11:29.160
about how to do it. I just haven't pursued

00:11:31.640 --> 00:11:32.040
those yet. But if other people have ideas,

00:11:34.340 --> 00:11:34.540
I think this would be really interesting to

00:11:35.520 --> 00:11:36.020
add to the LLM package.

00:11:37.260 --> 00:11:37.760
So contact me there.

00:11:42.100 --> 00:11:42.600
[Speaker 1]: Awesome. Quick note before we continue.

00:11:44.440 --> 00:11:44.540
So I'm not sure how long we're going to be on

00:11:46.040 --> 00:11:46.280
stream for, because this is the last talk

00:11:48.640 --> 00:11:49.120
before the break. If we are on the stream

00:11:49.840 --> 00:11:50.200
long-term, then great.

00:11:51.820 --> 00:11:52.300
But if not, folks are welcome to continue

00:11:53.320 --> 00:11:53.680
writing questions on the pad.

00:11:55.140 --> 00:11:55.280
And hopefully, Andrew will get to them at

00:11:58.020 --> 00:11:58.280
some point. Or if Andrew maybe has some extra

00:11:59.960 --> 00:12:00.140
time available and wants to stay on

00:12:01.640 --> 00:12:01.920
BigBlueButton here, then folks are also

00:12:03.940 --> 00:12:04.340
welcome to join here and chat with Andrew

00:12:08.940 --> 00:12:09.240
directly as well. Okay,

00:12:10.740 --> 00:12:10.900
awesome. So yeah, the next question is,

00:12:12.040 --> 00:12:12.400
what are your thoughts on the carbon

00:12:14.060 --> 00:12:14.560
footprint of LLM usage?

00:12:17.200 --> 00:12:17.700
[Speaker 0]: Yeah, it's a really interesting question.

00:12:23.180 --> 00:12:23.360
I don't have any particular knowledge or

00:12:25.440 --> 00:12:25.580
opinions about that. It's something I think

00:12:26.980 --> 00:12:27.180
we should all be educating ourselves more

00:12:32.240 --> 00:12:32.380
about. It is really, I mean,

00:12:33.040 --> 00:12:33.220
there's 2 parts of this,

00:12:35.380 --> 00:12:35.500
right? They take a, there's a huge amount of

00:12:37.160 --> 00:12:37.360
carbon footprint involved in training these

00:12:38.720 --> 00:12:39.220
things. Then running them is relatively

00:12:42.540 --> 00:12:42.880
lightweight. So the question is not

00:12:44.440 --> 00:12:44.920
necessarily like once it's trained,

00:12:46.480 --> 00:12:46.640
like I don't feel like it's a big deal to

00:12:48.280 --> 00:12:48.560
keep using it, but like training these things

00:12:50.680 --> 00:12:51.180
is kind of like the big carbon cost of it.

00:12:53.680 --> 00:12:54.160
But like right now, the way everything's

00:12:56.040 --> 00:12:56.260
going, like every, you know,

00:12:59.060 --> 00:12:59.560
all, you know, the top 5 or 6 tech companies

00:13:00.900 --> 00:13:01.400
are all training their LLMs,

00:13:03.580 --> 00:13:03.740
and this is all costing a giant amount of

00:13:06.820 --> 00:13:07.060
carbon probably. On the other hand these same

00:13:08.560 --> 00:13:08.680
companies are pretty good about using the

00:13:10.260 --> 00:13:10.440
least amount of carbon necessary you know

00:13:12.340 --> 00:13:12.740
they have their own their tricks for doing

00:13:13.260 --> 00:13:13.760
things very efficiently.

00:13:22.100 --> 00:13:22.360
[Speaker 1]: Cool next question, LLMs are slow and

00:13:24.000 --> 00:13:24.340
responding. Do you think Emacs should provide

00:13:26.680 --> 00:13:27.180
more async primitives to keep it responsive?

00:13:29.380 --> 00:13:29.880
Like the URL retrieve is quite bad at

00:13:31.720 --> 00:13:31.760
building API clients with it.

00:13:31.920 --> 00:13:32.420
Building API clients with it?

00:13:36.400 --> 00:13:36.900
[Speaker 0]: Yeah. Well, OK, so first of all,

00:13:40.240 --> 00:13:40.740
people should be using the LLM client.

00:13:48.740 --> 00:13:48.900
And So right now, 1 thing I should have

00:13:50.220 --> 00:13:50.440
mentioned at the top is that there are new

00:13:52.500 --> 00:13:52.640
packages that I recorded this talk that you

00:13:54.480 --> 00:13:54.920
just saw several months ago.

00:13:57.780 --> 00:13:58.180
And so like Elama, there's this package Elama

00:13:59.700 --> 00:14:00.080
that came out that is using the LM package.

00:14:02.440 --> 00:14:02.680
And so for example, it doesn't need to worry

00:14:05.140 --> 00:14:05.580
about this sort of thing because it just uses

00:14:07.560 --> 00:14:07.920
LLM and package and the LLM package worries

00:14:11.680 --> 00:14:11.820
about this. And while I'm on the subject of

00:14:12.540 --> 00:14:12.720
things I forgot to mention,

00:14:15.140 --> 00:14:15.340
I also should just mention very quickly that

00:14:17.020 --> 00:14:17.520
there is now an open source model,

00:14:21.680 --> 00:14:21.960
Mistral. And so that's kind of this new thing

00:14:23.860 --> 00:14:24.240
on the scene that happened after I recorded

00:14:26.240 --> 00:14:26.420
my talk. And I think it's super important to

00:14:28.660 --> 00:14:28.820
the community and important that we have the

00:14:30.620 --> 00:14:31.120
opportunity to use that if we want to.

00:14:33.160 --> 00:14:33.660
Okay, but to answer the actual question,

00:14:37.660 --> 00:14:38.100
there has been some talk about the problems

00:14:40.680 --> 00:14:40.840
with URL retrieve in the URL package in

00:14:42.200 --> 00:14:42.700
general in EmacsDevEl.

00:14:46.760 --> 00:14:47.080
It's not great. I would like to have better

00:14:50.900 --> 00:14:51.040
primitives. And I've asked the author of

00:14:54.060 --> 00:14:54.560
Please PLZ to kind of provide some necessary

00:14:56.120 --> 00:14:56.620
callbacks. I think that's a great library.

00:15:00.280 --> 00:15:00.360
And I'd like to see that kind of like,

00:15:01.320 --> 00:15:01.680
It's nice that we have options,

00:15:03.340 --> 00:15:03.520
and that is an option that uses curl on the

00:15:05.140 --> 00:15:05.640
back end, and that has some benefits.

00:15:09.060 --> 00:15:09.280
So there's this big debate about whether we

00:15:10.600 --> 00:15:11.100
should have primitives or just use curl.

00:15:13.340 --> 00:15:13.420
I'm not exactly sure what the right call is,

00:15:15.320 --> 00:15:15.820
but there has been discussions about this.

00:15:19.540 --> 00:15:20.040
[Speaker 1]: Excellent. And someone commented that GPTEL

00:15:21.820 --> 00:15:22.200
is async and apparently very good at tracking

00:15:22.300 --> 00:15:22.800
the point.

00:15:26.680 --> 00:15:27.180
[Speaker 0]: Yes, yes, GPTEL has similar functionalities

00:15:29.800 --> 00:15:30.040
to LLM, although I believe it's going to move

00:15:33.040 --> 00:15:33.540
to LLM itself sometime soon.

00:15:39.480 --> 00:15:39.860
[Speaker 1]: Next question, speaking of which,

00:15:42.440 --> 00:15:42.560
anyone trained or fine-tuned or prompted a

00:15:44.680 --> 00:15:44.760
model with their org data yet and applied it

00:15:46.560 --> 00:15:47.040
to interesting use cases like planning,

00:15:47.920 --> 00:15:48.340
scheduling, et cetera,

00:15:49.320 --> 00:15:49.820
and maybe care to comment?

00:15:54.620 --> 00:15:55.120
[Speaker 0]: I don't know anyone who is doing that.

00:15:55.860 --> 00:15:56.360
I think it is interesting.

00:15:57.800 --> 00:15:58.300
Like this is what I kind of mentioned at the

00:16:01.060 --> 00:16:01.300
very end of the talk. There is a lot of stuff

00:16:02.440 --> 00:16:02.540
there like you could you know if you

00:16:04.760 --> 00:16:04.920
especially mean an LLM can kind of work as

00:16:07.940 --> 00:16:08.160
sort of like a secretary kind of person that

00:16:12.180 --> 00:16:12.440
could help you prioritize Still it's a

00:16:14.760 --> 00:16:14.920
slightly unclear how what the best way to use

00:16:16.480 --> 00:16:16.720
it is So I think there's more of a question

00:16:18.340 --> 00:16:18.480
for the community about like what people have

00:16:21.140 --> 00:16:21.320
been trying. I see someone has mentioned that

00:16:23.400 --> 00:16:23.900
they are using it for weekly review.

00:16:26.940 --> 00:16:27.180
And it's kind of nice to like,

00:16:29.060 --> 00:16:29.380
maybe you could read your agenda or maybe

00:16:30.480 --> 00:16:30.780
this for like weekly review.

00:16:32.040 --> 00:16:32.240
It could like read all the stuff you've done

00:16:33.340 --> 00:16:33.480
and ask you questions about it.

00:16:35.020 --> 00:16:35.280
And like, what should happen next?

00:16:36.520 --> 00:16:36.780
Or like, is this going to cause a problem?

00:16:39.060 --> 00:16:39.280
Like, I can, I can understand if that could

00:16:40.860 --> 00:16:41.180
happen? That's like, that's kind of nice.

00:16:43.660 --> 00:16:44.160
And this kind of people have had good success

00:16:48.540 --> 00:16:48.760
out of using these LLMs to bounce ideas off

00:16:49.920 --> 00:16:50.420
of are, you know, for,

00:16:52.680 --> 00:16:52.800
you know, I've seen people say that like they

00:16:55.360 --> 00:16:55.600
want, they use it for reading and they kind

00:16:58.520 --> 00:16:58.740
of dialogue with the LM to kind of like do

00:16:59.500 --> 00:17:00.000
sort of active reading.

00:17:02.500 --> 00:17:02.860
So you can imagine doing something similar

00:17:04.400 --> 00:17:04.740
with your tasks where it's sort of you're

00:17:06.560 --> 00:17:06.760
engaged in dialogue about like planning your

00:17:08.880 --> 00:17:09.000
tax with some with a alum that could kind of

00:17:10.800 --> 00:17:11.180
understand what those are and ask you some

00:17:13.780 --> 00:17:13.940
questions I think it. You know,

00:17:16.839 --> 00:17:17.040
if it'd be nice. So, the problem is like

00:17:18.480 --> 00:17:18.980
there's no great way to share all this stuff.

00:17:20.720 --> 00:17:21.220
I guess if you have something like this,

00:17:23.300 --> 00:17:23.720
put it on Reddit. If you don't have Reddit,

00:17:24.599 --> 00:17:24.880
I don't know what to do.

00:17:26.000 --> 00:17:26.500
I would say put it somewhere.

00:17:28.840 --> 00:17:29.020
At the very least, I could maybe open up like

00:17:31.320 --> 00:17:31.820
an LLM discussion session on the LLM package

00:17:34.000 --> 00:17:34.500
GitHub, But not everyone likes to use GitHub.

00:17:36.100 --> 00:17:36.180
I don't know. It'd be nice if there's a

00:17:38.940 --> 00:17:39.060
mailing list or IRC chat for this sort of

00:17:40.840 --> 00:17:41.340
thing. But there isn't at the moment.

00:17:46.560 --> 00:17:46.720
[Speaker 1]: All right. Let's see. I think that's the end

00:17:48.080 --> 00:17:48.580
of the questions on the pad so far.

00:17:51.020 --> 00:17:51.180
There was also some discussion or some

00:17:52.260 --> 00:17:52.760
chatter, I believe, on IRC.

00:17:54.560 --> 00:17:54.820
I'm not sure. Andrew, are you on IRC right

00:18:00.060 --> 00:18:00.260
[Speaker 0]: I am, but I don't think I'm on any place that

00:18:01.400 --> 00:18:01.640
has the chatter. So if there's chatter,

00:18:02.440 --> 00:18:02.940
then I'm not seeing it.

00:18:04.600 --> 00:18:05.100
[Speaker 1]: now? Okay. Yeah, it was in the emacsconf-dev

00:18:06.760 --> 00:18:07.260
channel.

00:18:09.600 --> 00:18:10.100
[Speaker 0]: Okay, let me see if I can.

00:18:25.600 --> 00:18:25.840
Oh, yes. I mean, I could see the channel,

00:18:27.520 --> 00:18:27.840
but I missed whatever came before.

00:18:29.340 --> 00:18:29.480
So if there's anything you want to kind of

00:18:30.840 --> 00:18:31.340
call out, I can try to answer it here.

00:18:35.320 --> 00:18:35.640
[Speaker 1]: OK, cool. I believe at least 2 other folks

00:18:37.500 --> 00:18:38.000
who are participating in the discussion there

00:18:40.120 --> 00:18:40.620
who have also joined here on BigBlueButton,

00:18:42.440 --> 00:18:42.940
Codin Quark and AeonTurn92.

00:18:47.000 --> 00:18:47.480
So you folks, if Andrew is still available

00:18:50.460 --> 00:18:50.640
and has time, you're welcome to chat here and

00:18:53.000 --> 00:18:53.320
ask questions or discuss here as well.

00:18:55.580 --> 00:18:55.840
[Speaker 0]: 1 Thank you. Thank you for your help,

00:18:57.740 --> 00:18:58.080
and thank you for reading all the questions.

00:18:59.700 --> 00:18:59.820
[Speaker 1]: AUDIENCE 2 Cheers, and thanks to you for a

00:19:00.540 --> 00:19:01.040
great talk and the discussion.

00:19:01.880 --> 00:19:02.380
[Speaker 0]: AUDIENCE AUDIENCE 1 Thank you.

00:19:03.140 --> 00:19:03.640
[Speaker 1]: AUDIENCE 2 Cheers.

00:19:07.900 --> 00:19:08.040
[Speaker 0]: So I'll just, I will wait here and see if

00:19:08.320 --> 00:19:08.760
there's any questions.

00:19:10.760 --> 00:19:11.260
If not, I will log off after a few minutes.

00:19:15.900 --> 00:19:16.080
[Speaker 2]: Well, I guess since we were mentioned that

00:19:18.480 --> 00:19:18.980
there was a small chat about local alarms.

00:19:22.640 --> 00:19:23.000
Because chat dpt is nice,

00:19:25.600 --> 00:19:26.100
no, but privacy concerns,

00:19:27.380 --> 00:19:27.880
and it's not free and stuff.

00:19:31.000 --> 00:19:31.500
Which, so The question is,

00:19:36.960 --> 00:19:37.460
what is the promise for local models?

00:19:39.660 --> 00:19:40.160
[Speaker 0]: Yeah, so local is definitely...

00:19:41.380 --> 00:19:41.880
[Speaker 2]: Or at least open source.

00:19:45.680 --> 00:19:46.120
[Speaker 0]: Yeah, so there is a local open source model,

00:19:47.960 --> 00:19:48.460
Misral, which you could run.

00:19:51.340 --> 00:19:51.840
The LLM package allows you to use,

00:19:56.120 --> 00:19:56.260
I think there's 3 kind of local things you

00:19:58.100 --> 00:19:58.440
could use. Like many of these things,

00:20:00.220 --> 00:20:00.480
there's like many kind of ways to do the same

00:20:03.960 --> 00:20:04.460
sort of thing. So LLM is supporting OLAMMA

00:20:10.240 --> 00:20:10.520
and LLAMMA-CPP. And let's see,

00:20:12.240 --> 00:20:12.740
1 other. Which 1 is it?

00:20:18.420 --> 00:20:18.700
And maybe that's it. Maybe the,

00:20:21.820 --> 00:20:21.940
oh, GPT for all. So each 1 of these kind of

00:20:23.100 --> 00:20:23.600
has slightly different functionality.

00:20:26.820 --> 00:20:27.180
For example, I think GPT for all doesn't

00:20:31.780 --> 00:20:32.280
support embeddings. And I hear that Olama's

00:20:33.740 --> 00:20:34.240
embeddings are kind of currently broken.

00:20:35.920 --> 00:20:36.420
But basically they should support everything.

00:20:39.100 --> 00:20:39.600
And the open source models are,

00:20:43.180 --> 00:20:43.380
so the local models are reasonably good.

00:20:44.760 --> 00:20:44.900
Like I don't think you'd use them and be

00:20:46.200 --> 00:20:46.639
like, what is this horrible nonsense?

00:20:50.200 --> 00:20:50.380
Like it's, it gives you relatively good

00:20:51.820 --> 00:20:52.120
results. Like it's not gonna be at the level

00:20:56.060 --> 00:20:56.320
of like GPT 3.5 or 4, but it's not far away

00:20:57.720 --> 00:20:58.220
from GPT 3.5, I think.

00:21:02.380 --> 00:21:02.880
[Speaker 2]: I'm just saying that Olam has like a presets

00:21:05.940 --> 00:21:06.300
for connecting the actual working servers for

00:21:06.300 --> 00:21:06.800
Olama?

00:21:08.560 --> 00:21:08.760
[Speaker 0]: So, I'll try. Yeah, so you could,

00:21:09.860 --> 00:21:10.040
what you could do is you could like for

00:21:11.940 --> 00:21:12.100
example you could download Olama which is

00:21:15.780 --> 00:21:15.940
just a way of setting up local models and

00:21:17.320 --> 00:21:17.780
running local models on your machine.

00:21:18.580 --> 00:21:18.820
So typically what it does,

00:21:19.720 --> 00:21:20.020
you like download a program,

00:21:23.720 --> 00:21:23.800
let's say Olama. Then Olama will have the

00:21:24.940 --> 00:21:25.440
ability to download models.

00:21:27.240 --> 00:21:27.360
And so you could choose from just a host of

00:21:29.280 --> 00:21:29.440
different models. Each 1 of these things has

00:21:30.200 --> 00:21:30.440
a bunch of different models.

00:21:31.920 --> 00:21:32.080
So it downloads all these things to your

00:21:36.600 --> 00:21:37.020
machine. But I would say that the key problem

00:21:40.200 --> 00:21:40.580
here is that it requires a fairly beefy

00:21:40.580 --> 00:21:41.080
machine.

00:21:42.600 --> 00:21:43.100
[Speaker 2]: So. Yeah, yeah, of course.

00:21:45.060 --> 00:21:45.300
Why I was asking, because you briefly

00:21:46.440 --> 00:21:46.880
mentioned that there are some Israeli

00:21:52.300 --> 00:21:52.440
servers. I understand that they run it like a

00:21:53.680 --> 00:21:54.000
government or stuff like that?

00:21:55.440 --> 00:21:55.940
No, no, sorry. People want everyone?

00:21:59.340 --> 00:21:59.620
[Speaker 0]: I don't, I mean, maybe you've said something

00:22:00.620 --> 00:22:01.020
that sounded like Israeli servers.

00:22:01.620 --> 00:22:02.120
[Speaker 2]: Okay, okay.

00:22:04.920 --> 00:22:05.080
[Speaker 0]: I think- There's no government LLMs as far as

00:22:06.820 --> 00:22:07.280
I know. Although, I'm sure the governments

00:22:08.200 --> 00:22:08.700
are working on their own LLMs,

00:22:10.980 --> 00:22:11.480
et cetera. But yeah, basically your choices

00:22:15.060 --> 00:22:15.220
are spend a, I mean, if you use open AI or

00:22:16.080 --> 00:22:16.580
something or anything else,

00:22:17.960 --> 00:22:18.460
you're really not spending any money.

00:22:20.560 --> 00:22:20.840
Like I've never been able to spend any money

00:22:23.680 --> 00:22:24.020
on OpenAI. Like unless you're doing something

00:22:25.840 --> 00:22:26.280
very intensive and really are using it to,

00:22:28.000 --> 00:22:28.180
you know, if you're using it for your

00:22:29.620 --> 00:22:29.780
personal use, it's just hard to spend any

00:22:31.720 --> 00:22:31.960
money. But on the other hand,

00:22:32.780 --> 00:22:32.860
it's not free. So you can,

00:22:33.040 --> 00:22:33.540
you know,

00:22:36.300 --> 00:22:36.680
[Speaker 2]: Actually, it's rather cheap.

00:22:37.680 --> 00:22:38.180
There's no question about that.

00:22:40.580 --> 00:22:40.920
The problem is that it has a bad track record

00:22:41.580 --> 00:22:42.080
on privacy.

00:22:45.540 --> 00:22:46.040
[Speaker 0]: Yes, that's, I think that is a key problem.

00:22:48.120 --> 00:22:48.280
This is probably the number 1 reason why you

00:22:51.840 --> 00:22:52.340
might want to use a local AI,

00:22:54.720 --> 00:22:55.220
a local LLM. Another 1 is like,

00:22:57.400 --> 00:22:57.900
you may not agree with the decisions.

00:23:00.360 --> 00:23:00.820
You know, there's a lot of trust and safety

00:23:05.140 --> 00:23:05.440
stuff that these companies have to do.

00:23:09.020 --> 00:23:09.240
Like they don't want like the LMs to kind of

00:23:11.400 --> 00:23:11.640
like give you, like tell you how you can make

00:23:13.180 --> 00:23:13.580
meth or how you can make a bomb,

00:23:14.960 --> 00:23:15.460
which they would do. They would totally do

00:23:19.580 --> 00:23:20.080
it. So, But each time you kind of restrict

00:23:22.540 --> 00:23:22.680
what is happening with what you can get out

00:23:23.860 --> 00:23:24.360
of the LM, it gets a little worse.

00:23:24.960 --> 00:23:25.440
So some people

00:23:27.040 --> 00:23:27.540
[Speaker 2]: want to have local. That's expected.

00:23:31.080 --> 00:23:31.400
I guess even open source language modules

00:23:33.540 --> 00:23:34.040
will soon have HR spaces because it's simply

00:23:34.720 --> 00:23:35.220
a legal issue.

00:23:40.760 --> 00:23:41.140
[Speaker 0]: I think that's true. I also think that there

00:23:42.880 --> 00:23:43.040
probably will be, although I don't know of

00:23:45.060 --> 00:23:45.300
any offhand, that will are completely

00:23:46.960 --> 00:23:47.200
uncensored. I know people are interested and

00:23:48.240 --> 00:23:48.480
are running uncensored models.

00:23:49.440 --> 00:23:49.940
I don't know how to do it.

00:23:52.280 --> 00:23:52.780
I think it's a little bit dubious,

00:23:54.960 --> 00:23:55.040
but some people do want to do it.

00:23:56.280 --> 00:23:56.780
There's another reason for using local

00:24:02.280 --> 00:24:02.780
servers. Do you have any recommendation for

00:24:05.500 --> 00:24:05.720
models to run locally and also comments on

00:24:06.780 --> 00:24:07.280
whether a GPU is required?

00:24:14.040 --> 00:24:14.160
Usually a GPU, well, you can run it without a

00:24:16.960 --> 00:24:17.460
GPU, but it does run much better.

00:24:19.480 --> 00:24:19.980
Like for example, I think when I used,

00:24:22.560 --> 00:24:23.060
Lama is sort of like a standard.

00:24:27.160 --> 00:24:27.320
This was the model for that Facebook came out

00:24:31.380 --> 00:24:31.880
with for local use. And It was,

00:24:37.260 --> 00:24:37.760
yeah, it's good. It's,

00:24:40.400 --> 00:24:40.900
but it's now it's I think,

00:24:44.620 --> 00:24:44.920
Mistral is kind of like has a better

00:24:46.480 --> 00:24:46.800
performance, But there's also different model

00:24:51.000 --> 00:24:51.500
sizes. There's 7B, like the Lama 7B is OK.

00:24:52.940 --> 00:24:53.440
The Mistral 7B, 7 billion,

00:24:54.800 --> 00:24:55.300
are like, basically it'll take like,

00:24:58.380 --> 00:24:58.880
you can run it with like 16 gigs of RAM,

00:25:02.720 --> 00:25:03.040
is pretty good. It's probably about as equal

00:25:06.900 --> 00:25:07.000
to the LLAMA13B. Those are the number of

00:25:08.360 --> 00:25:08.860
parameters, if I remember correctly.

00:25:10.680 --> 00:25:11.180
And then there's a 7B,

00:25:12.340 --> 00:25:12.840
which I've never been able to run.

00:25:16.120 --> 00:25:16.620
And even if the 7B, if you run it without a

00:25:19.640 --> 00:25:20.140
GPU, it takes quite a while to answer.

00:25:22.080 --> 00:25:22.580
I think I've had experiences where it took

00:25:23.940 --> 00:25:24.440
literally like several,

00:25:26.480 --> 00:25:26.780
like 5 minutes before it even started

00:25:28.880 --> 00:25:29.100
responding, but you do eventually get

00:25:32.220 --> 00:25:32.580
something. And it could be that like things

00:25:33.840 --> 00:25:33.960
have gotten better since the last time I

00:25:35.440 --> 00:25:35.940
tried this, because things are moving fast.

00:25:38.360 --> 00:25:38.860
But it is super recommended to have a GPU.

00:25:42.440 --> 00:25:42.620
This is the problem. It's kind of like,

00:25:43.840 --> 00:25:44.180
yes, free software is great.

00:25:46.880 --> 00:25:47.120
But if free software is requiring that you

00:25:50.460 --> 00:25:50.760
have these kind of beefy servers and have all

00:25:52.000 --> 00:25:52.500
this hardware, that's not great.

00:25:53.600 --> 00:25:54.100
I think there's a case to be made.

00:25:55.680 --> 00:25:56.180
[Speaker 1]: a hardware

00:25:59.040 --> 00:25:59.540
[Speaker 0]: with slots instead of a laptop.

00:26:01.560 --> 00:26:02.060
Yeah, yeah, that's right.

00:26:03.660 --> 00:26:03.960
[Speaker 2]: Ideally, you can have Ideally,

00:26:07.400 --> 00:26:07.660
it would be nice if FSL for all things could

00:26:12.040 --> 00:26:12.540
run something for open source model.

00:26:16.320 --> 00:26:16.640
And not free, but the key point is that it's

00:26:16.640 --> 00:26:17.140
Libre?

00:26:22.580 --> 00:26:23.080
[Speaker 0]: Yes, so actually I think Google does do that.

00:26:24.720 --> 00:26:24.800
I'll have to look it up,

00:26:27.560 --> 00:26:27.820
but I haven't explored this yet.

00:26:31.220 --> 00:26:31.720
But Google's server, which LLM does support,

00:26:33.800 --> 00:26:34.300
supports arbitrary models.

00:26:36.420 --> 00:26:36.600
So you can run LLMA or things like that.

00:26:38.940 --> 00:26:39.200
The problem is that even if you're running

00:26:40.320 --> 00:26:40.820
Mistral, which has no restrictions.

00:26:42.940 --> 00:26:43.140
So this is the kind of thing that like the

00:26:44.900 --> 00:26:45.060
Free Software Foundation cares a lot about.

00:26:47.240 --> 00:26:47.740
Like you want it to be like no restrictions,

00:26:49.740 --> 00:26:49.840
legal restrictions on you as you run the

00:26:52.080 --> 00:26:52.580
model. So even if it's running Mistral,

00:26:54.800 --> 00:26:55.300
just by using the server,

00:26:58.460 --> 00:26:58.660
the company server, it will impose some

00:26:59.440 --> 00:26:59.900
restrictions on you probably,

00:27:02.320 --> 00:27:02.480
right? There's gonna be some license that you

00:27:04.760 --> 00:27:05.260
have to, or something you have to abide by.

00:27:08.480 --> 00:27:08.600
So I think, yes, it depends on how much you

00:27:09.280 --> 00:27:09.780
care about it, I guess.

00:27:19.500 --> 00:27:19.640
I should find out more about that and make

00:27:21.580 --> 00:27:22.080
sure that it's a good point that I should,

00:27:23.980 --> 00:27:24.180
you know, people should be able to run free

00:27:25.920 --> 00:27:26.280
models over the server.

00:27:28.320 --> 00:27:28.440
So I should make sure we support that in the

00:27:40.360 --> 00:27:40.860
LLM package. So, is there any other questions

00:27:48.240 --> 00:27:48.740
Or is otherwise we can end the session.

00:28:00.800 --> 00:28:01.040
Yeah, all right. Thank you.

00:28:02.440 --> 00:28:02.940
Thank you. Thank you everyone who listened.

00:28:04.540 --> 00:28:05.040
I'm super happy like I,

00:28:06.560 --> 00:28:07.060
the interest is great.

00:28:08.900 --> 00:28:09.220
I think there's great stuff to be done here

00:28:10.960 --> 00:28:11.140
and I'm kind of super excited what we're

00:28:11.940 --> 00:28:12.160
going to do in the next year,

00:28:13.140 --> 00:28:13.440
so hopefully, like next year,

00:28:14.600 --> 00:28:14.760
and the conference we have something even

00:28:16.440 --> 00:28:16.560
more exciting to say about LLM and how they

00:28:17.320 --> 00:28:17.820
can be used with Emacs.

00:28:19.620 --> 00:28:20.120
So thank

00:28:30.060 --> 00:28:30.560
you