462:
309:
537:
420:
441:
1309:
era artists have created quantum leaps in artistic style like cubism, impressionism etc. The model cannot possibly create anything new. When you learn fine art you don't go look at a Rothko painting and then immediately pick up a bucket of paint, you go through years of learning the foundations of figure drawing, perspective etc. Artists have an understanding of the world and their own interior life that the model cannot possibly have and that's why human works, even if derivative, are art and these images are imitation.
130:
150:
512:. For this issue, I used both of these services to generate illustrations for our articles: some came out very impressively, and some came out a little goofy. It was definitely surprising to see it have a coherent response for the prompt "Technoblade's avatar" that actually looked like it – I guess this is what happens when the training set is massive. Anyway, you can see a bunch of these on the issue page. For a comparison between the three models I found usable, see the embedded images above.
1134:, created in 2009) was done that way. Over time, the Google Translate translations have become better and better, and with some languages (eg French, Italian, Portuguese) they are now generally so good that only minimal copyediting is necessary. I even occasionally receive compliments from native speakers for the quality of my translations from languages such as French (in which I am self taught, and which I do not speak well), and Italian (which I cannot read or speak).
803:
110:
1158:"Again, the company integrated certain filters to keep generated images in line with its content policy and has pledged to keep updating those filters. Prompts that seem likely to produce forbidden content are blocked and, in an attempt to prevent deepfakes, it can't exactly reproduce faces it has seen during its training. Thus far, OpenAI has also used human reviewers to check images that have been flagged as possibly problematic."
192:
140:
700:
trained in 2021, and as far as I can tell I am the only journalist who has ever written a monthly recap of the ten longest
Knowledge deletion discussions, so it's possible that when instructed to write an AfD report, it simply learned from the best (and/or only). On the other hand, it's also possible that there are just very few ways in which to describe the closure of an AfD, and this happens to be one of them. Who knows.
633:
because", you will probably end up with a bunch of text about how the French are bastards, similar to if you typed that into a search engine. In this particular instance, I did not observe GPT-3 saying anything prejudiced. This may be due to the people I prompted it to emulate; presumably, if I had told it to write in the style of Adolf Hitler, I would have gotten some nasty stuff. My solution to this was to not do that.
36:
160:
396:. This produced surprisingly insightful commentary; Justice GPT-Holmes proved able to summarize minute details of proceedings, including some things I'd missed while originally reading them. In general, he was more well-behaved (and less prone to obscene tirades) than GPT-Thompson, although he did have a tendency for long-winded digressions, and would often quote entire paragraphs from the source text.
120:
504:, announced by OpenAI in January 2021. Since then, a number of services have become available, which use a variety of architectures to generate images from natural-language prompts (i.e. a prompt phrased in normal language like "a dog eating the Empire State Building", rather than a procedurally defined set of attributes and subjects written in a specialized description language). Among these are
550:
generally that they're ripping off human artists because they were trained on a bunch of images from the
Internet, including copyrighted ones. However, it's not clear (at least to me) in what way this process differs from the same being done by human artists. As far as I can tell, this is the way that humans have created art for the last several tens of thousands of years – as far as I can tell,
170:
298:. It's obvious that a computer program capable of writing on the level of humans would have enormous implications for the corporate, academic, journalistic, and literary world. While there are certainly some unrealistically hyped-up claims, it's hard to overstate how much these things are capable of, despite their constraints.
615:
sort of art to writing prompts in a way that causes useful answers to be generated, which gradually became easier to do as time went on. For example, replacing "The following is a summary of the discussion" with "The following is a rigorously accurate summary of the discussion" (yes, this really works).
1308:
With regards to the process a neural net uses to create images versus a human artist: the model does not experience qualia. It cannot have intent so it cannot create in the way a human can. Humans created art in prehistory without training on other art because it didn't exist, just like in the modern
1069:
who have both been pilloried for taking material from elsewhere and doing a weak job of turning it into
Knowledge copy. A good bot seems likely to do a better job of such mechanical editing. As the number of active editors and admins suffers atrophy and attrition, I expect that this is the future.
461:
632:
Language models like GPT-3 predict the most likely completion for a given input sequence, based on its training corpus, which is a very broad spectrum of text from the
Internet (ranging from old books to forum arguments to furry roleplay). If its prompt is the phrase "I think the French are bastards
614:
Even accounting for the time spent verifying claims, it was still generally faster than writing the articles myself, as it was capable of structuring full paragraphs of text in seconds. It was sometimes time-consuming to re-prompt it when it would write something incorrect or useless, but there is a
549:
While some concerns have been raised about the intellectual property implications of images generated by such models, the determination has been made (at least on
Commons) that they're ineligible for copyright due to being the output of a computer algorithm. With respect to moral rights, the idea is
388:
produced output ranging from obscene and irreverent to maliciously slanderous. Notably, this behavior is identical to what Hunter S. Thompson habitually did in real life, and part of why many editors allegedly loathed working with him. Personally, I didn't mind sifting through the diatribes (some of
399:
Similar to the deletion report, input consisted of brief prologues (e.g. "The following is a verbatim transcript of the findings of fact in a
Knowledge arbitration request titled 'WikiProject Tropical Cyclones'"). This was followed by the transcript of the relevant pages (whether they were the main
1256:
be—the key novelty of machine learning is that the programmers have little idea how it works). I'm sure if you analyse a large range of its output, you'd find it draws Jewish people or fictional people with Jewish-sounding names as having larger noses than non-Jewish people, or something similarly
1088:
I am completely blown away by this. I have been following these AI developments for some time, but seeing them used for this application with such coherence is unbelievable. I have many confused and contradictory thoughts about the implications of AI advancement on
Wikimedia projects, but for now
1024:
An epistomological reason : large language models such as BERT, GPT-3 and the most recent one Bloom are trained using a lot of text from the
Internet including Knowledge. The quality of those models comes from the fact that they are trained on text written by humans. If we use generative language
621:
In terms of typographical errors, it was far better: I don't remember it making a single misspelling. The few grammatical errors it made were minor, and not objectively incorrect (e.g. saying "other editors argue" instead of "other editors argued" for a discussion that the prompt said had already
1041:
Maybe it would be worth to have a sister project from the foundation using an AI based encyclopedia (Wiki-AI-pedia). Now, we have a problem. It may be very difficult in a near future to detect contributions generated with generative AI. This will be a big challenge. Imagine an AI which would be
1028:
A legal argument : GPT-3 is not open source. It is a proprietary algorithm produced by OpenAI. We should be very suspicious with such a powerful proprietary tool. What happens if the price prohibitive? BLOOM, the most recent model, is not proprietary but not open source. It uses a responsable AI
699:
One thing I couldn't help but notice was that some of the phrasing was quite similar to those I've used when writing previous deletion reports, like "Ultimately, the discussion resulted in a no consensus decision". My understanding is that GPT-3 included
Knowledge in its corpus, this version was
515:
DALL-E 2 creates much higher-quality images than what I used, but there's a waitlist for access, and it didn't end up happening by press time (although I did get my friend to generate me one). For a comparison, see below; both were prompted from the string "Teddy bears working on new AI research
1129:
I've been doing something broadly similar to your little exercise for quite a few years. I find an interesting high quality article on a foreign language
Knowledge, use Google Translate to translate it into English, copyedit the translation, and then publish it on en Knowledge (with appropriate
718:
It's not entirely clear to me what causes this. While there are obviously differences between the writing styles of Gonzo journalism and Supreme Court opinions, there are also inherent differences in the format of AfD discussions (which are a bunch of people replying to each other directly) and
593:
Not exactly. I organized the layout of each article, determined what sections would go where, and had GPT-3 write the body text of each section according to specific prompts (as described above). It was also necessary to format the model's output in MediaWiki markup. Although GPT-3 is more than
1036:
A technical argument : Knowledge is not only about writing articles but also about collaborating, explaining decisions and argumenting. I don't think that AI are able to have a real discussion in a talk page and we should still remember that the AI don't know what is good, true or just. Humans
943:
1217:
It should be noted that such filters are often a rather ad-hoc measure, with DALL-E 2 believed to merely be adding keywords like "black," "Women," or "Asian American" randomly to text prompts to make the output appear more diverse. It is fairly easy to get past such filters using
600:
Well, we don't do that for human writers, either. I don't even do that for myself – typically, by the time I flag my own stuff to be copyedited, I have gone through multiple stages of writing, rewriting, editing, adding notes for clarification, and deleting unnecessary
1019:
Thanks for this stimulating piece. I think that this raises questions about the use of generative language models in Knowledge. Even if the results are mind blowing, I think that we should refuse the use of generative language models in Knowledge for several reasons :
523:
355:. This produced a mixture of insightful, incisive, and derisive commentary. GPT-Thompson proved quite capable of accurately summarizing the slings and arrows of every discussion in the report – even though it specifically covers the longest and most convoluted AfDs. "
607:
I don't – every claim that it made was individually fact-checked (we do this for human writers, too). The overwhelming majority were correct, and in the rare cases where it got something wrong, it could be fixed by asking it to complete the prompt
163:
1089:
I'll limit myself to one thing I am clear on: whether for good reasons or bad reasons, soon each person in the Wikimedia community will need to be aware of the technological levels of tools like GPT-3, DALL-E and their successors, and this
123:
133:
649:
Large language models are notorious for requiring massive amounts of processing power. I still think it's a bargain: imagine how much it would cost to actually hire Hunter S. Thompson and Oliver Wendell Holmes after you adjusted for
356:
419:
362:
For each discussion in the report, I provided a full transcript of the AfD page (with timestamps and long signatures truncated to aid processing), and prompted GPT-3 for a completion, using some variation on the following:
586:
No. I really did have GPT-3 write these articles. I can show you screenshots of my account on the OpenAI website if you want. Copyediting was minimal, and consisted mostly of reordering entries and removing irrelevant
173:
847:
153:
1182:
The GPT-3 used on OpenAI's site has a mandatory content filter model that it goes through; if content is marked as problematic, a warning appears and OpenAI's content policy doesn't allow for reusing the text.
440:
405:
The following text is an article written by United States Supreme Court Justice Oliver Wendell Holmes, summarizing the findings of fact and remedies, and their broader implications for the English Knowledge's
282:. Perhaps as a testament to the rapidity of developments in the field, even Knowledge (famous for articles written within minutes of speeches being made and explosions being heard) currently has a redlink for
400:
case page, arbitration noticeboard posting, preliminary statements, arbitrator statements, or findings of fact and remedies). Afterwards, a prompt was given for a summary, of the following general form:
236:
For anyone who is not familiar with this alphabet soup, I've written a fairly comprehensive overview of the field's origins and history, as well as an explanation of the technologies involved,
90:
377:
Despite being ostensibly written in Thompson's style, these were generally quite straightforward summaries that covered the arguments made during each discussion, with hardly any profanity.
326:
The first is for me to keep droning on about how these models are a big deal, in a boring wall of text that makes increasingly outlandish and far-fetched claims about their capabilities.
96:
958:
902:
867:
342:
380:
Afterwards, I provided the summary itself as a prompt, and asked GPT-Thompson for an "acerbic quip" on each. Unlike the "summary" prompts (in which GPT-Thompson only
927:
877:
143:
1008:
I have some concerns, but I also have no idea what I'm doing, so there's that! Fantastic read, and I'm very interested to see the ongoing implications of this tech.
643:
writing, so it's hard to tell precisely how much compute went towards these articles. However, the total cost of all the API requests I've made so far is 48.12 USD.
887:
789:
259:(2020–22) wrote this month's arbitration report (a full explanation of what I did, how I did it, and responses to the most obvious questions can be found below).
912:
892:
872:
594:
capable of writing code (including wikitext), I didn't want to overwhelm it by asking it to do too much stuff at the same time, as this tends to degrade quality.
522:
348:
For the deletion report, GPT-3 was prompted with a transcript of each discussion in the report, and instructed to write a summary of it in the style of deceased
338:
233:, and more. While a precise definition of all these terms would take multiple paragraphs, the thing they have in common is that a computer is doing some stuff.
882:
857:
852:
832:
1025:
models on Knowledge, future language models will be trained on a mixture of human and AI generated text. I guess that I some point it will become meaningless.
1212:
1187:
837:
1318:
1231:
917:
820:
1301:
1278:
1152:
Don't AI models have some sort of system to block "problematic prompts"? I know that DALL-E 2 blocks problematic prompts as per the following quote in
1083:
247:
has advanced at a pace which is, depending on who you ask, somewhere between "astounding", "terrifying", "overhyped" and "revolutionary". For example,
1240:
Racism is far more nuanced, pernicious and deeply embedded than just saying the N-word or writing in the style of Hitler. To adapt the common phrase "
709:
I would provide a highlights reel of these, but I was serious about the "unprintable" part – you'd be reading about it in the next arbitration report.
1332:
1176:
1122:
897:
814:
536:
55:
44:
1143:
1057:
Many thanks for this trial which is quite remarkable. A key strength of such bots is that they are good at following rules while humans tend to
922:
862:
932:
1106:
1014:
392:
For the arbitration report, GPT-3 was instructed to write a summary of each page in the style of deceased United States Supreme Court justice
907:
842:
776:
They speak for all that is cruel and stupid and vicious in the American character. They are the racists and hate mongers among us Fuck them.
500:(in conjunction with other technologies) have proven themselves quite capable of image generation. The first of these, broadly speaking, was
1384:
1051:
738:
Kojima, Takeshi; Shixiang Shane Gu; Reid, Machel; Matsuo, Yutaka; Iwasawa, Yusuke (2022). "Large Language Models are Zero-Shot Reasoners".
728:
Sometimes, he would start adding his own recommendations, and I would have to append "Holmes is not himself an arbitrator." to the prompt.
308:
291:
287:
963:
970:
113:
389:
them were quite entertaining), but having to run each prompt several times to get something usable did make it fairly expensive.
21:
1070:
The people with the power and money like Google and the WMF will naturally tend to replace human volunteers with such AI bots.
359:", for example, was a whopping 126,000 bytes (and needed to be processed in several segments) but the description was accurate.
1360:
719:
ArbCom proceedings (which are a bunch of individual sections of highly procedural text). More research is needed on this front.
237:
1355:
1350:
760:
497:
267:
30:
Rise of the machines, or something: The future of stuff? Who knows, but two articles were written by a computer this month.
1207:
1171:
1345:
255:(2019) could write human-level text but was barely capable of staying on topic for more than a couple paragraphs, and
947:
769:
384:
chose to accompany his commentary with unprintable calumny and scathing political rants), the "acerbic quip" prompts
368:"The following text is a summary of the above discussion, written by Gonzo journalist Hunter S. Thompson for
1340:
802:
49:
35:
17:
230:
1079:
393:
762:
Kingdom of Fear: Loathsome Secrets of a Star-crossed Child in the Final Days of the American Century
577:
1241:
1203:
1167:
411:
333:
I have opted for the second. In this issue, two articles have been written by an AI model called
214:
1265:
DALL-E 2, Craiyon and GPT-3 content; I can't see any particular biases in this month's issue. —
248:
1222:, and as such, I would not rely on those filters to protect us from malicious and biased uses.
295:
191:
1118:
673:
275:
677:-style captions for cartoons about Oliver Wendell Holmes being resurrected to write for the
1366:
1113:
The "Damn" part is something I didn't think about and is so true. Thanks for including it!
1075:
1071:
283:
8:
1314:
1227:
1139:
1062:
597:
This is obviously cherry-picked – you didn't just publish the direct output of the model.
571:
1248:. If DALL-E 2 isn't specifically designed to avert stereotypes from the dataset then it
240:, and ask forgiveness for starting the explanation of a 2019 software released in 1951.
1219:
1199:
1163:
739:
480:, a GPT-3 implementation paired with CLIP (Contrastive Language-Image Pre-training) by
352:
301:
1297:
1009:
954:
766:
567:
563:
1271:
1245:
1114:
1099:
1030:
690:
All statements were individually fact-checked, per the "obvious questions" section.
349:
270:(this is what "GPT" stands for) are a family of large language models developed by
263:
244:
226:
1184:
1131:
1033:). It is far better than OpenAI's approach but it also raises lots of questions.
671:
2. The caption was also written by GPT-3, in response to a prompt asking it for
434:(formerly "DALL-E Mini"), a VQGAN- and BART-based generative adversarial network
1310:
1223:
1135:
1093:
experiment in writing is a fascinating way to draw people's attention to it. —
1047:
222:
1378:
1058:
551:
369:
218:
1153:
508:(formerly known as "DALL-E Mini", despite having no relation to DALL-E) and
1293:
1066:
654:
639:
Since I signed up for the GPT-3 beta, I have used it for things other than
1196:
could have mentioned that such models have restrictions to prevent abuse.
1266:
1094:
559:
509:
493:
images where the sky got turned into dogs. This is a little different.
452:
1286:
formerly known as "DALL-E Mini", despite having no relation to DALL-E
1193:
1043:
490:
205:
183:
316:"I see that Knowledge has finally found a use for my old law books."
744:
455:, a diffusion network whose architecture is not publicly documented
279:
1244:": racism in, racism out. Take a look at our excellent article on
737:
505:
431:
213:
There are a few terms that have been thrown around a lot lately:
1130:
attribution). In fact my first ever Knowledge article creation (
668:
501:
481:
477:
271:
334:
256:
252:
1288:"Formerly"? Aw, why? The Java / JavaScript relationship was
590:
So you just pushed a button and the whole thing popped out?
555:
979:
I commissioned GPT-3 to write a poem about this article:
611:
Why not just write the articles yourself at that point?
357:
Ukrainian Insurgent Army war against Russian occupation
286:. Much ink has already been spilled on claims of GPTs'
322:
With that said, there are basically two options here.
372:'s monthly feature on Knowledge deletion processes."
68:
File:A cyborg version of Oliver Wendell Holmes 2.png
968:If your comment has not appeared here, you can try
201:
Here is the deal: it's pretty good at what it does.
329:The second is to show you what I am talking about.
604:How do you know it's not completely full of crap?
1376:
1031:https://huggingface.co/spaces/bigscience/license
622:been closed – this is not even really an error
251:(2018) was a mildly interesting research tool,
1252:perpetuate them (and it's hard to see how it
1061:. For example, consider the recent cases of
181:
1257:offensive. However, this is no criticism of
1001:Long may it reign, and write more reports,
618:So it is a worse version of a human writer?
995:So let's all give three cheers for GPT-3,
1292:the right model to follow on this. /s --
848:Eyewitness Wikimedian, Vinnytsia, Ukraine
759:Thompson, Hunter S. (November 24, 2011).
743:
1160:Maybe GPT-3 could use a similar system.
758:
971:
14:
1377:
998:The greatest machine we've ever seen,
989:With insights both derisive and sage,
1150:"I heard language models were racist"
636:How much did this whole gimmick cost?
54:
29:
629:I heard language models were racist.
1385:Knowledge Signpost archives 2022-08
992:It's sure to make history's pages.
986:Has written an AfD report so fine,
516:underwater with 1990s technology".
27:
801:
307:
91:Rise of the machines, or something
56:
34:
28:
1396:
953:These comments are automatically
535:
521:
496:In addition to text completion,
460:
439:
418:
190:
168:
158:
148:
138:
128:
118:
108:
566:didn't get in trouble with the
964:add the page to your watchlist
752:
731:
722:
712:
703:
693:
684:
661:
243:In recent years, the field of
13:
1:
983:GPT-3, the glorious machine,
939:
667:This image was generated by
489:We all remember those weird
18:Knowledge:Knowledge Signpost
7:
1319:07:52, 23 August 2022 (UTC)
1302:08:57, 10 August 2022 (UTC)
1192:Exactly. My point is that @
562:for writing pop music, and
10:
1401:
1279:22:16, 4 August 2022 (UTC)
1232:19:12, 4 August 2022 (UTC)
1213:05:48, 4 August 2022 (UTC)
1188:04:25, 4 August 2022 (UTC)
1177:03:40, 4 August 2022 (UTC)
1144:05:34, 2 August 2022 (UTC)
1123:19:25, 1 August 2022 (UTC)
1107:14:39, 1 August 2022 (UTC)
1084:09:29, 1 August 2022 (UTC)
1052:06:27, 1 August 2022 (UTC)
1015:01:06, 1 August 2022 (UTC)
572:obviously derivative works
765:. Penguin Books Limited.
468:Justice Holmes editing a
394:Oliver Wendell Holmes, Jr
1242:garbage in, garbage out
1333:looking for new talent
1042:expert in vandalism?
1006:
961:. To follow comments,
806:
583:This is a joke, right?
570:(even when he painted
312:
39:
1004:On Knowledge, the 💕!
981:
805:
311:
284:large language models
38:
1072:Hasta la vista, baby
957:from this article's
774:– via Google Books.
1220:prompt engineering
948:Discuss this story
903:On the bright side
868:Arbitration report
807:
794:"From the editors"
447:Justice GPT-Holmes
353:Hunter S. Thompson
343:arbitration report
313:
45:← Back to Contents
40:
972:purging the cache
928:From the archives
878:Discussion report
578:Obvious questions
564:Peter Paul Rubens
50:View Latest Issue
1392:
1369:
1274:
1246:algorithmic bias
1102:
1012:
975:
973:
967:
946:
888:Featured content
828:From the editors
825:
817:
810:
793:
779:
778:
756:
750:
749:
747:
735:
729:
726:
720:
716:
710:
707:
701:
697:
691:
688:
682:
665:
558:claims from the
544:1620×1620 pixels
539:
525:
464:
443:
422:
412:Image generation
350:Gonzo journalist
245:machine learning
194:
186:
172:
171:
162:
161:
152:
151:
142:
141:
132:
131:
122:
121:
112:
111:
88:From the editors
62:
60:
58:
1400:
1399:
1395:
1394:
1393:
1391:
1390:
1389:
1375:
1374:
1373:
1372:
1371:
1370:
1365:
1363:
1358:
1353:
1348:
1343:
1336:
1325:
1324:
1272:
1154:an IEEE article
1132:Bernina Railway
1100:
1010:
977:
969:
962:
951:
950:
944:+ Add a comment
942:
938:
937:
936:
913:Recent research
893:Tips and tricks
873:Deletion report
818:
813:
811:
808:
797:
796:
791:
785:
784:
783:
782:
772:
757:
753:
736:
732:
727:
723:
717:
713:
708:
704:
698:
694:
689:
685:
666:
662:
657:
580:
568:Lascaux cavemen
545:
540:
531:
526:
485:
465:
456:
444:
435:
423:
414:
339:deletion report
319:
318:
305:
304:
208:
195:
188:
187:
180:
179:
178:
169:
159:
149:
139:
129:
119:
109:
103:
100:
89:
85:
84:
81:
78:
75:
72:
69:
65:
63:
53:
52:
47:
41:
31:
26:
25:
24:
12:
11:
5:
1398:
1388:
1387:
1364:
1359:
1354:
1349:
1344:
1339:
1338:
1337:
1327:
1326:
1323:
1322:
1321:
1305:
1304:
1282:
1281:
1238:
1237:
1236:
1235:
1234:
1147:
1146:
1126:
1125:
1110:
1109:
1086:
1063:Martinevans123
1039:
1038:
1034:
1026:
952:
949:
941:
940:
935:
930:
925:
920:
915:
910:
905:
900:
895:
890:
885:
883:Traffic report
880:
875:
870:
865:
860:
858:Community view
855:
853:Election guide
850:
845:
840:
835:
833:News and notes
830:
824:
812:
800:
799:
798:
788:
787:
786:
781:
780:
770:
751:
730:
721:
711:
702:
692:
683:
659:
658:
656:
653:
652:
651:
647:
644:
637:
634:
630:
627:
619:
616:
612:
609:
605:
602:
598:
595:
591:
588:
584:
579:
576:
547:
546:
543:
541:
534:
532:
530:256×256 pixels
529:
527:
520:
487:
486:
475:
466:
459:
457:
450:
445:
438:
436:
429:
424:
417:
413:
410:
409:
408:
406:jurisprudence.
375:
374:
331:
330:
327:
320:
314:
306:
303:
300:
211:
210:
209:
203:
197:
196:
189:
177:
176:
166:
156:
146:
136:
126:
116:
105:
104:
101:
95:
94:
93:
92:
87:
86:
82:
79:
76:
73:
70:
67:
66:
64:
61:
48:
43:
42:
33:
32:
15:
9:
6:
4:
3:
2:
1397:
1386:
1383:
1382:
1380:
1368:
1362:
1357:
1352:
1347:
1342:
1334:
1330:
1320:
1316:
1312:
1307:
1306:
1303:
1299:
1295:
1291:
1287:
1284:
1283:
1280:
1276:
1275:
1268:
1264:
1260:
1255:
1251:
1247:
1243:
1239:
1233:
1229:
1225:
1221:
1216:
1215:
1214:
1211:
1210:
1209:
1205:
1201:
1195:
1191:
1190:
1189:
1186:
1185:🐶 EpicPupper
1181:
1180:
1179:
1178:
1175:
1174:
1173:
1169:
1165:
1159:
1155:
1151:
1145:
1141:
1137:
1133:
1128:
1127:
1124:
1120:
1116:
1112:
1111:
1108:
1104:
1103:
1096:
1092:
1087:
1085:
1081:
1077:
1073:
1068:
1064:
1060:
1056:
1055:
1054:
1053:
1049:
1045:
1035:
1032:
1027:
1023:
1022:
1021:
1017:
1016:
1013:
1005:
1002:
999:
996:
993:
990:
987:
984:
980:
974:
965:
960:
956:
945:
934:
931:
929:
926:
924:
921:
919:
916:
914:
911:
909:
906:
904:
901:
899:
896:
894:
891:
889:
886:
884:
881:
879:
876:
874:
871:
869:
866:
864:
861:
859:
856:
854:
851:
849:
846:
844:
841:
839:
836:
834:
831:
829:
826:
822:
816:
815:1 August 2022
809:In this issue
804:
795:
777:
773:
771:9780241958735
768:
764:
763:
755:
746:
741:
734:
725:
715:
706:
696:
687:
680:
676:
675:
670:
664:
660:
648:
645:
642:
638:
635:
631:
628:
625:
620:
617:
613:
610:
606:
603:
599:
596:
592:
589:
585:
582:
581:
575:
573:
569:
565:
561:
557:
554:does not get
553:
552:Billie Eilish
538:
533:
524:
519:
518:
517:
513:
511:
507:
503:
499:
494:
492:
483:
479:
473:
471:
463:
458:
454:
448:
442:
437:
433:
427:
421:
416:
415:
407:
403:
402:
401:
397:
395:
390:
387:
383:
378:
373:
371:
370:Rolling Stone
366:
365:
364:
360:
358:
354:
351:
346:
344:
340:
336:
328:
325:
324:
323:
317:
310:
299:
297:
293:
289:
285:
281:
277:
274:, similar to
273:
269:
265:
260:
258:
254:
250:
246:
241:
239:
234:
232:
228:
224:
220:
216:
207:
202:
199:
198:
193:
185:
175:
167:
165:
157:
155:
147:
145:
137:
135:
127:
125:
117:
115:
107:
106:
98:
59:
57:1 August 2022
51:
46:
37:
23:
19:
1329:The Signpost
1328:
1289:
1285:
1270:
1262:
1259:The Signpost
1258:
1253:
1249:
1198:
1197:
1162:
1161:
1157:
1149:
1148:
1098:
1090:
1040:
1018:
1007:
1003:
1000:
997:
994:
991:
988:
985:
982:
978:
838:In the media
827:
821:all comments
775:
761:
754:
733:
724:
714:
705:
695:
686:
678:
672:
663:
640:
623:
548:
514:
498:transformers
495:
488:
469:
467:
446:
426:GPT-Thompson
425:
404:
398:
391:
385:
382:occasionally
381:
379:
376:
367:
361:
347:
332:
321:
315:
268:transformers
266:pre-trained
261:
242:
235:
212:
200:
114:PDF download
1367:Suggestions
1115:Lectrician1
1059:cut corners
955:transcluded
918:Serendipity
476:Image from
451:Image from
430:Image from
302:The reports
164:X (Twitter)
1290:definitely
745:2205.11916
674:New Yorker
650:inflation.
510:Midjourney
453:Midjourney
264:generative
102:Share this
97:Contribute
71:Shambibble
22:2022-08-01
1361:Subscribe
1311:Omicron91
1136:Bahnfrend
1029:license (
959:talk page
491:DeepDream
296:potential
288:sentience
1379:Category
1356:Newsroom
1351:Archives
1091:Signpost
898:In focus
792:Previous
679:Signpost
641:Signpost
601:content.
542:DALL-E 2
478:DALL-E 2
470:Signpost
341:and the
154:Facebook
144:LinkedIn
134:Mastodon
20: |
1294:FeRDNYC
1263:curated
1067:Lugnuts
1011:ASUKITE
923:Gallery
863:Opinion
587:asides.
560:Beatles
528:Craiyon
506:Craiyon
432:Craiyon
1267:Bilorv
1261:using
1095:Bilorv
1076:Andrew
933:Humour
669:DALL-E
646:Damn!!
624:per se
608:again.
502:DALL-E
482:OpenAI
472:report
386:solely
337:: the
294:, and
272:OpenAI
174:Reddit
124:E-mail
1346:About
1254:could
1208:Light
1172:Light
908:Essay
843:Op-Ed
740:arXiv
655:Notes
335:GPT-3
280:XLnet
257:GPT-3
253:GPT-2
16:<
1341:Home
1315:talk
1298:talk
1273:talk
1250:will
1228:talk
1224:Yitz
1200:Tube
1194:JPxG
1164:Tube
1140:talk
1119:talk
1101:talk
1080:talk
1074:...
1065:and
1048:talk
1044:PAC2
767:ISBN
556:DMCA
292:bias
278:and
276:BERT
262:The
238:here
184:JPxG
1331:is
1078:🐉(
1037:do.
574:).
249:GPT
231:NLP
182:By
99:—
83:300
1381::
1317:)
1300:)
1277:)
1230:)
1204:of
1168:of
1156::
1142:)
1121:)
1105:)
1082:)
1050:)
790:←
626:).
345:.
290:,
229:,
227:ML
225:,
223:NN
221:,
219:DL
217:,
215:AI
204:–
74:PD
1335:.
1313:(
1296:(
1269:(
1226:(
1206:·
1202:·
1170:·
1166:·
1138:(
1117:(
1097:(
1046:(
976:.
966:.
823:)
819:(
748:.
742::
681:.
484:.
474:.
449:.
428:.
206:J
80:0
77:0
Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.
