4 ms·

The key here is engagement. I have been increasingly hearing people complain that they don't see content from their friends as much or at all anymore on social

The key here is engagement. I have been increasingly hearing people complain that they don't see content from their friends as much or at all anymore on social media. LLMs are a way for Facebook to present information that it wants to present to keep you engaged, as if it came from someone you care about. That seems like it would keep people even more hooked than presenting information from people you don't know.

_heimdall 2y ago
That's interesting. I don't use Facebook, but I would have expected anyone annoyed with not seeing content from their friend to be even more frustrated if the content they do see is a bot posting as their friend. It'll be really interesting if this approach were to actually work at keeping people engaged as if they were interacting with friends.
- Zambyte 2y ago
  I think it's more interesting when you consider that the bot may genuinely post things that reflect the opinions and experiences of the person that it is posting on behalf of. The real person may seamlessly jump in mid conversation, and that does a lot to validate the experience of talking to the bot. I think it's an extremely unhealthy path to walk down, but I can see why they're doing it.
moffkalast 2y ago
Embedding ads deep into language models that then seamlessly and unobtrusively roll them into their responses is definitely the endgame for commercialization. Nobody wants to be the first to do it of course, since they'll get dropped by everyone like a billion degree potato.
kjkjadksj 2y ago
Do they assume people are stupid and won’t see these as ai bots?
- Carrok 2y ago
  In my experience, many people can't even identify blatantly AI generated images as such on social media. Determining some text came from an LLM will probably be even harder for that crowd.
- twelve40 2y ago
  > they assume people are stupid i think that was the fundamental assumption from day 1 (cue the "they trust me" quote)
- _heimdall 2y ago
  They could instead assume that people won't care. Based on the thread here there are people who said they wouldn't care, and I could see HN being a population that would would be more biased against obvious LLM content than the rest of the crowd.
- famouswaffles 2y ago
  Just because Open AI and co have taken the strategy of intentionally training the style of output to be stilted and robotic with RL doesn't mean it's a fundamental limitation of LLMs. LLMs can sound way way more natural and indistinguishable than GPT-4, Claude etc do by default. Even as things are, people can't actually easily tell. Has nothing to do with being stupid.
lumost 2y ago
The wave of AI friend apps also seems like a semi-persistent trend. If the goal of a social media company is to fulfill a social need. Then we should expect them to try their hand at AI friends.

_heimdall 2y ago

That's interesting. I don't use Facebook, but I would have expected anyone annoyed with not seeing content from their friend to be even more frustrated if the content they do see is a bot posting as their friend. It'll be really interesting if this approach were to actually work at keeping people engaged as if they were interacting with friends.

Zambyte 2y ago
I think it's more interesting when you consider that the bot may genuinely post things that reflect the opinions and experiences of the person that it is posting on behalf of. The real person may seamlessly jump in mid conversation, and that does a lot to validate the experience of talking to the bot. I think it's an extremely unhealthy path to walk down, but I can see why they're doing it.

moffkalast 2y ago

Embedding ads deep into language models that then seamlessly and unobtrusively roll them into their responses is definitely the endgame for commercialization. Nobody wants to be the first to do it of course, since they'll get dropped by everyone like a billion degree potato.

kjkjadksj 2y ago

Do they assume people are stupid and won’t see these as ai bots?

Carrok 2y ago
In my experience, many people can't even identify blatantly AI generated images as such on social media. Determining some text came from an LLM will probably be even harder for that crowd.
twelve40 2y ago
> they assume people are stupid i think that was the fundamental assumption from day 1 (cue the "they trust me" quote)
_heimdall 2y ago
They could instead assume that people won't care. Based on the thread here there are people who said they wouldn't care, and I could see HN being a population that would would be more biased against obvious LLM content than the rest of the crowd.
famouswaffles 2y ago
Just because Open AI and co have taken the strategy of intentionally training the style of output to be stilted and robotic with RL doesn't mean it's a fundamental limitation of LLMs. LLMs can sound way way more natural and indistinguishable than GPT-4, Claude etc do by default. Even as things are, people can't actually easily tell. Has nothing to do with being stupid.

lumost 2y ago

The wave of AI friend apps also seems like a semi-persistent trend. If the goal of a social media company is to fulfill a social need. Then we should expect them to try their hand at AI friends.