#aivshuman — Public Fediverse posts

@Ruby Skunkloon

For both alt text and CWs, I wish there was a more automated way to handle this. Like, having canned CW tags at the ready, or an image detection bot that actually works as intended for filling out alt text. (That's before getting into the weeds of "AI", which is contentious in itself.)

For CWs, in theory, things could be a whole lot better. A whole lot better than what Mastodon's culture demands.

Instead of the writer supplying a CW and forcing the self-same CW upon all readers all the same, CWs could be automatically generated on the readers' side. Like, those who need that particular CW get it, and those who don't need it don't get it.

This is technically possible. Mastodon has introduced it with version 4.0 in October, 2022. It's a filter setting. In general, this has been available in the Fediverse since 2010 when Friendica introduced it first. On Friendica and all its descendants, CW culture has always been built around this rather than writing CWs into the summary field and forcing them upon everyone all the same.

This has got a few downsides, though:

• Writers still have to remember to take care of adding filter-triggering keywords.
• The best way to add keywords without shoehorning them into the text is as hashtags, but many Mastodon users have a problem with more than, say, four hashtags per message.
• Whereas Friendica and its descendants simply have a keyword list for this purpose, Mastodon requires one new filter to be set up for each keyword.
• I guess there are lots of Mastodon apps that don't have creating or editing filters built in. And I know there are lots of Mastodon users who would never ever touch Mastodon's Web UI under any circumstances.
• Mastodon users are used to being coddled and pampered all over. Most of them would never take care of CWs being generated on their side by themselves. They expect CWs to be served to them by the writers.
• And then there are those Fediverse server applications that either don't have a hiding/CW-generating filter or no filters at all.

Fully automated, zero-click, fire-and-forget image description generators have been desired by Mastodon users for about as long as alt-texts have been a thing. But let me tell you: Fully automated, fire-and-forget image description generators that can describe any image perfectly accurately are utopic. Even more if they're also expected to adhere to all existing image-describing and alt-text rules and adapt the descriptions both to the context and to the audience that's to be expected.

Trust me. My original images could barely be more obscure and niche. They're renderings from extremely obscure 3-D virtual worlds. And I've actually had two of them described by an AI after writing my own highly detailed image descriptions 100% by hand with no AI assistance. It took me several hours to write one description and two days to write the other one.

The results were abysmal. The AI descriptions were no match for my hand-written descriptions at all. They omitted things that really stood out in the images, they got things wrong, they guessed things that I knew for a fact.

Granted, it was extra unfair because the AI only had the image to look at whereas I wrote the image descriptions by looking at the real deal in-world with an almost infinitely higher resolution, with the ability to look around obstacles even. Still, at least for my own images, I consider this style of writing image descriptions the benchmark. AI will never get there. It will never have the extreme niche knowledge of an expert, and it will never be able to see things that I see.

CC: @Princess Nettles Succubish

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #Mastodon #MastodonCulture #CW #CWs #CWMeta #ContentWarning #ContentWarnings #ContentWarningMeta #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta #AI #LLM #LLMs #AIVsHuman #HumanVsAI

@Ruby Skunkloon

For both alt text and CWs, I wish there was a more automated way to handle this. Like, having canned CW tags at the ready, or an image detection bot that actually works as intended for filling out alt text. (That's before getting into the weeds of "AI", which is contentious in itself.)

For CWs, in theory, things could be a whole lot better. A whole lot better than what Mastodon's culture demands.

Instead of the writer supplying a CW and forcing the self-same CW upon all readers all the same, CWs could be automatically generated on the readers' side. Like, those who need that particular CW get it, and those who don't need it don't get it.

This is technically possible. Mastodon has introduced it with version 4.0 in October, 2022. It's a filter setting. In general, this has been available in the Fediverse since 2010 when Friendica introduced it first. On Friendica and all its descendants, CW culture has always been built around this rather than writing CWs into the summary field and forcing them upon everyone all the same.

This has got a few downsides, though:

• Writers still have to remember to take care of adding filter-triggering keywords.
• The best way to add keywords without shoehorning them into the text is as hashtags, but many Mastodon users have a problem with more than, say, four hashtags per message.
• Whereas Friendica and its descendants simply have a keyword list for this purpose, Mastodon requires one new filter to be set up for each keyword.
• I guess there are lots of Mastodon apps that don't have creating or editing filters built in. And I know there are lots of Mastodon users who would never ever touch Mastodon's Web UI under any circumstances.
• Mastodon users are used to being coddled and pampered all over. Most of them would never take care of CWs being generated on their side by themselves. They expect CWs to be served to them by the writers.
• And then there are those Fediverse server applications that either don't have a hiding/CW-generating filter or no filters at all.

Fully automated, zero-click, fire-and-forget image description generators have been desired by Mastodon users for about as long as alt-texts have been a thing. But let me tell you: Fully automated, fire-and-forget image description generators that can describe any image perfectly accurately are utopic. Even more if they're also expected to adhere to all existing image-describing and alt-text rules and adapt the descriptions both to the context and to the audience that's to be expected.

Trust me. My original images could barely be more obscure and niche. They're renderings from extremely obscure 3-D virtual worlds. And I've actually had two of them described by an AI after writing my own highly detailed image descriptions 100% by hand with no AI assistance. It took me several hours to write one description and two days to write the other one.

The results were abysmal. The AI descriptions were no match for my hand-written descriptions at all. They omitted things that really stood out in the images, they got things wrong, they guessed things that I knew for a fact.

Granted, it was extra unfair because the AI only had the image to look at whereas I wrote the image descriptions by looking at the real deal in-world with an almost infinitely higher resolution, with the ability to look around obstacles even. Still, at least for my own images, I consider this style of writing image descriptions the benchmark. AI will never get there. It will never have the extreme niche knowledge of an expert, and it will never be able to see things that I see.

CC: @Princess Nettles Succubish

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #Mastodon #MastodonCulture #CW #CWs #CWMeta #ContentWarning #ContentWarnings #ContentWarningMeta #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta #AI #LLM #LLMs #AIVsHuman #HumanVsAI