Post AUIEUWSKCDMYZ757Me by tetraca@pleroma.sanssouci.club
(DIR) More posts by tetraca@pleroma.sanssouci.club
(DIR) Post #AUHuEg6sXJyCn4DYyu by simon@fedi.simonwillison.net
2023-04-03T21:44:04Z
0 likes, 0 repeats
Anyone seen good discussion about the ethics of publishing AI-generated text?I don't feel right publishing prose that was exclusively written by AI - for one thing, I think it's rude to expect people to spend more time reading something than you spent writing itIt feels like there should be a lot more thoughtful, informed conversation about this - and approaches to personal AI ethics in general, independent of the larger conversations about training, bias, risk to society and so on
(DIR) Post #AUHugwwXXoIn6drYkC by alecthegeek@mstdn.social
2023-04-03T21:49:01Z
0 likes, 0 repeats
@simon I did it once (for a short intro in a local community newsletter) and I put a clear statement at the start. That seems to be the obvious thing to do.Anyone caught not doing that should be publicly exposed as a fraud.When the content was not wholly written by AI a modified statement should (still) be used
(DIR) Post #AUHuuTwCBPXtCX9dBo by koen_hufkens@mastodon.social
2023-04-03T21:49:17Z
0 likes, 0 repeats
@simon Not a discussion, just a comic.https://mastodon.social/@nixCraft/110094129309726213
(DIR) Post #AUHv7VteD9bh2cB0Oe by NIH_LLAMAS@mastodon.social
2023-04-03T21:50:57Z
0 likes, 0 repeats
@simon I wouldn't use it because I wouldn't know how close to stolen the output is, or from whom.
(DIR) Post #AUHvYMEQFkwcsuh5nc by memeticist@jorts.horse
2023-04-03T21:58:39Z
0 likes, 0 repeats
@simon it's the same as the ethics of publishing anything elsethe ethical question is about the volume of misinformation generated, not whether misinformation is being producedi like your idea; it violates the gricean maxim of quantity to give readers too much text
(DIR) Post #AUHwfUwaMKw9t2JsgK by bartek@sfba.social
2023-04-03T22:11:18Z
0 likes, 0 repeats
@simon "expect people to spend more time reading something than you spent writing it" is basically almost all of Twitter ;-)
(DIR) Post #AUHwvJY2UrCBAzNd7Q by acowley@mastodon.social
2023-04-03T22:14:00Z
0 likes, 0 repeats
@simon It can be somewhat inconsiderate to send a long, rambling artisanal email, so maybe having an AI wave a brevity wand over an early draft wouldn't be so bad.Time to write being greater than the time to read sounds like a pretty good rule of thumb, actually. Hard to imagine a good situation where the author's got that the wrong way around.
(DIR) Post #AUI2uER6cH42l20Jay by arthegall@mastodon.roundpond.net
2023-04-03T23:21:14Z
0 likes, 0 repeats
@simon I thought the B12 Team has thought and talked about at least a few parts of it, here: https://www.b12.io/how-we-ai/
(DIR) Post #AUI35R4nxbBN97uopE by mesirii@chaos.social
2023-04-03T23:22:47Z
0 likes, 0 repeats
@simon i think i like the approach the IB folks took by not forbidding it but asking students to mark it as quotes with source. That deals with the length and attribution. So folks can decide if they want to read it.
(DIR) Post #AUI5828lKR5QaL8JCy by mlanger@mastodon.world
2023-04-03T23:46:17Z
0 likes, 0 repeats
@simon My bigger problem with publishing something "that was exclusively written by AI" -- it wasn't. AI systems are plagiarists. I can't see a conversation about this being very long.🤷🏻♀️
(DIR) Post #AUI6i0nXtcm2nkve3k by victorvicpal@dair-community.social
2023-04-04T00:03:33Z
0 likes, 0 repeats
@simon@DAIR wrote this letter:https://www.dair-institute.org/blog/letter-statement-March2023
(DIR) Post #AUI71pzPn9WD4Svygi by simon@fedi.simonwillison.net
2023-04-04T00:07:27Z
0 likes, 0 repeats
@victorvicpal that's the big picture ethics stuff - which I agree is important, but it's not as useful for me when I'm trying to make personal decisions on my own ethics in terms of how and when I use existing tools
(DIR) Post #AUI7GUIyhEgWA54QF6 by victorvicpal@dair-community.social
2023-04-04T00:09:17Z
0 likes, 0 repeats
@simonI usually follow deon checklist. It's not perfect but it covers lots of topics:https://deon.drivendata.org/
(DIR) Post #AUI7wa1VpHWw3iyKNU by simon@fedi.simonwillison.net
2023-04-04T00:17:40Z
0 likes, 0 repeats
@victorvicpal I hadn't seen that before, that's fantasticThat's more ethics for training and using your own models than guidance on when to use tools like Stable Diffusion and ChatGPT as an individual
(DIR) Post #AUI9m19Uoznmv7nVkO by peterbutler@mas.to
2023-04-04T00:37:57Z
0 likes, 0 repeats
@simon As long as credit is front and center, i.e. which LLM, was it edited, etc., it’s good with me.>> I think it's rude to expect people to spend more time reading something than you spent writing it To me, as long as the text has value, who cares who wrote it or how long it took?The only discussions I’ve seen so far has been about academic publishing and crediting LLM as co-author (short answer: don’t):https://publishing.aip.org/resources/researchers/policies-and-ethics/authors/
(DIR) Post #AUIEUWSKCDMYZ757Me by tetraca@pleroma.sanssouci.club
2023-04-04T01:07:45.642136Z
0 likes, 0 repeats
@simon I know this isn't what you asked for but: I think it's probably just bad to publish things you don't actually care about and don't curate and genuinely think is a good contribution. AI written text just happens to overlap heavily with text nobody could ever hope to care about but happens to exist for some begrudging economic obligation on the part of an author. SEO spam is fully crafted by humans is far more contemptible than say, a middling but still entertaining story fully by an AI posted onto a message board, which someone who is poor at writing carefully constructed through prompts. AI text just serves as a gas can for all the informational pollution that already exists (and shouldn't).
(DIR) Post #AUIEUXRIXW8TcDboP2 by simon@fedi.simonwillison.net
2023-04-04T01:30:40Z
0 likes, 0 repeats
@tetraca I really like that way of thinking about it. The thing that's unethical is publishing text that you don't sincerely believe has value. If you've vetted it and are willing to stand by it then the role that AI played in creating it isn't really very important.
(DIR) Post #AUINw2kb8Fzc8EpIpM by williamgunn@mastodon.social
2023-04-04T03:16:55Z
0 likes, 0 repeats
@simon I've mostly written non-publicly about it, but the main points are the volume at which it can be created, difficulty sorting generated from written, asymmetric competition between writers and posters of generated content, and yes, also some intellectual property rights stuff. Writers get outcompeted by posters on any platform that allows it, especially when money is involved, because consumers aren't discriminating about what info they consume.
(DIR) Post #AUJfNm8SqNAb4ADPyS by sketchytech@techhub.social
2023-04-04T06:10:38Z
0 likes, 0 repeats
@memeticist @simon yep, I asked AI to identify a cartoon character on the front of a record sleeve from the 1990s and no matter how many times I told it it was wrong, it kept fictionalising the answer. Telling me it was Tank Girl and then Judge Dredd, and an actress from a film, then a drawing of a pin-up girl by some artists. All were way off and it just wouldn't admit it didn't know even when supplied with direct links. Blatant fibbing is not going to advance us anywhere.
(DIR) Post #AUJfNmufx7jbTglv7Y by simon@fedi.simonwillison.net
2023-04-04T18:07:03Z
0 likes, 0 repeats
@sketchytech @memeticist Which AI system was that, and how did you feed it the image?It's possible you ran into a variant of this problem: https://simonwillison.net/2023/Mar/10/chatgpt-internet-access/
(DIR) Post #AUJghkUqkMiBjFm3QO by sketchytech@techhub.social
2023-04-04T18:21:11Z
0 likes, 0 repeats
@simon @memeticist it was ChatGPT and it asked for the links. This is how the conversation went. I'll attach photos of the actual cover artwork to next message.
(DIR) Post #AUJgtUvGTaPEtzjQfY by simon@fedi.simonwillison.net
2023-04-04T18:24:03Z
0 likes, 0 repeats
@sketchytech @memeticist Hah, yeah that's a different LLM failure-case: they're terrible at answering very specific search queries like that one, and very likely to make stuff up I wrote a bit about that here https://simonwillison.net/2023/Apr/2/calculator-for-words/
(DIR) Post #AUJhbeGr54xCdXsooS by simon@fedi.simonwillison.net
2023-04-04T18:32:08Z
0 likes, 0 repeats
@sketchytech @memeticist I thought GPT-4 might handle this better - it's less likely to hallucinate in my experience - but no, and it still spat out the lie that "Upon further research [...]" which is rubbish because it can't do "further research"
(DIR) Post #AUKdFk9CIeCYp3QHTs by sketchytech@techhub.social
2023-04-05T05:17:23Z
0 likes, 0 repeats
@simon @memeticist Deputy Dawg 😂
(DIR) Post #AULGO0MEt8FEEu1kAa by Kestenb@infosec.exchange
2023-04-05T12:35:49Z
0 likes, 0 repeats
@simon Hey simon. I think it's important to look at the eithics/values and especially privacy policy of the LLM your are working with. Unfortunately, not a lot of choice we it comes to options yet. Like finding an ethical gas station :(But I'm particularly concerned that OpenAI / ChatGPT are intentionally deceptive in their privacy docs.They make a big deal about "OpenAI does not use data submitted by customers via our API to train OpenAI models or improve OpenAI’s service offering". But there is immense confusion because most people assume that services which call those backend APIs are also safe from "improvements", but in fact, all the public facing things like Playground and ChatGPT are excluded. Current copyright law only protects "the expression" or implementation of ideas, not the concepts. So CoPilot could claim that they don't use your code snippet in training, but then use the representation for it, or a slight modification. While the open/sharing model is fine if it benefits everyone, enterprises should be very careful about sending code to any LLM that has misleading Privacy policies and does not explain how the model improves.Currently Netflix, AWS, Google all prohibit most use of CoPilot tools.I'd also say, looking at the VCs backing a company might tell you a bit about the ethics of the board and the decisions they will make.
(DIR) Post #AULnyYQD0t4mncv5Qe by jsamwrites@mastodon.social
2023-04-05T18:52:47Z
0 likes, 0 repeats
@simonI am still hesitant to use AI-generated writing. Somehow, I am not yet comfortable with it.