Incidentally, this is why I worry about the dominant position of AI critics on this site that "AI doesn't work and is just a scam" It absolutely does work, in good ways (controlled studies on education, medicine) and in ways that create risks & problems. Pretending AI isn't real is a big mistake. - ThreadSky

emollick.bsky.social • 33 days ago

Incidentally, this is why I worry about the dominant position of AI critics on this site that "AI doesn't work and is just a scam"

It absolutely does work, in good ways (controlled studies on education, medicine) and in ways that create risks & problems. Pretending AI isn't real is a big mistake.

Reposted from Ethan Mollick

👀Today’s AIs are already hyper persuasive.

A controversial study where LLMs tried to persuade users on Reddit found that: “Notably, all our treatments surpass human performance substantially, achieving persuasive rates between three and six times higher than the human baseline.”

Comments

philphi.bsky.social•33 days ago

Ethan, try to get the twitter AI people over here. Bluesky is a woke echo chamber at times, but has the AT Protocol and some smart people.

emollick.bsky.social•33 days ago

I cannot recommend this site for AI researchers right now. It isn't about wokeness, it is purely that if you post about AI here you may get overwhelmed with disturbing attacks. So the discussions are thin as a result. I am sticking around and would be happy to see a change, but it isn't great here.

erisianrite.com•33 days ago

I post a good bit about it and even go looking for fights sometimes, and I have a great time here. But I know this isn’t universal and so I don’t recommend it either. Still better than X: The Everything App.

mackuba.eu•33 days ago

I think if you mute / block / get blocked by enough people, it might become less of a problem… there must be a finite amount of those

temujin9.t9productions.com•33 days ago

This is the problem modlists were made to solve. Unfortunately finite is relative unless you automate: turns out there's a LOT of assholes here. Not sure how to automate this one yet or I might, because they absolutely are a net negative for Bluesky as a whole.

catblanketflower.yuwakisa.com•33 days ago

With the note that it's not just anti-AI. There are other folks who are targeted pretty viciously. It really hurts the site

catblanketflower.yuwakisa.com•33 days ago

💯agree but it's like in the hundreds or thousands at a time (the curse of social media). So the scale of of clicking is a problem

Might be able to make blocklists, except I personally don't like blocklists

Twitter went through a dogpile phase. Bluesky may have to address it

catblanketflower.yuwakisa.com•33 days ago

Also just a side note that it's not just AI. Will Stancil has attracted a permanent mob 🤣 And also some lady got mobbed because she said she likes talking to her neighbor on a plane

So the anti-AI is bad, but it's part of a broader problem with the site

slime.bsky.social•33 days ago

this is mostly from ai lists and i still see these reactionary freaks swarming this site. there are so many of them and they are proud of being morons

mackuba.eu•33 days ago

Oh wow 😧

aparker.io•33 days ago

it’s not the people who use those lists that bother me, it’s the people who should and don’t :p

tedunderwood.me•33 days ago

Everyone has a different experience of a platform.

But just to illustrate variance: I post pro-AI stuff 24/7, with 18k followers, and never get enough pushback to have a cortisol spike.

What this place needs, imo, is not fewer haters, but a bunch more people interested in AI.

mackuba.eu•33 days ago

Oh, interesting! Good to have more data points. (Personally I didn't really get attacked, but I also don't talk much about it, I've just seen some dramas from 3rd person perspective)

tedunderwood.me•33 days ago

Admittedly I am ruthless about detaching hostile quote-posts; that’s part of the reason I acknowledge experience varies. Not everyone has time for that.

But in any case, more lovers is a fixable problem. Fewer haters? I don’t think there are actions anyone can take to create that.

nico-encounter.bsky.social•30 days ago

I enjoy seeing the conflict play out

mattsayar.com•32 days ago

Thank you and others like you who keep the spirit of joy and amazement about these tools alive. I'm also annoyed by the anti- crowd but I've stuffed all those thoughts in here so I don't have to dwell on it anymore

https://mattsayar.com/im-tired-of-dismissive-anti-ai-bias/

replicantw.bsky.social•33 days ago

I’m grateful that you post here in addition to X, for exactly this reason. There’s so much denial and cope about AI here, but I’m hopeful posts like yours will resonate.

holmesandgardens.bsky.social•32 days ago

I remember the first time I interacted with GPT-3 and thought “this is a miracle”. It’s amazing how fast its capability has become pedestrian to us all. In the right hands I think it could be great, but the main problem is that tech-oligarchs want this tech in use NOW whether it is ready/safe or not

ramshackle78.bsky.social•33 days ago

Are the critics dominant though? I can see a scenario where you are seeing a lot of them in your replies but that is a distortion.

emollick.bsky.social•33 days ago

The replies here are something else. https://bsky.app/profile/hankgreen.bsky.social/post/3lnjohdrwf22j

ramshackle78.bsky.social•33 days ago

Sometimes when I read your posts (and by the way very grateful that you are posting here) I think you've being too harsh on it 🙂But that's normal, since you're putting it through the ringer so it's bound to disappoint from time to time, it's still early.

emollick.bsky.social•33 days ago

Yes, I hope I am wrong or that it is just selection.

ramshackle78.bsky.social•33 days ago

Yeah but same principle (self selection)
I honestly think it might feel like critics dominant because in comparison to the AI-god-everything on X, here it's just people using it, being realistic about both its strengths and its weaknesses.
Yeah lots who don't even try it properly but malign it

emollick.bsky.social•33 days ago

Going to have to disagree with you there. Many AI researchers have been talking about how many personal attacks they get from the BlueSky community. This site has a bullying and swarming problem on AI. This past weekend another professor got a flood of death threats for a causal post on AI utility.

ramshackle78.bsky.social•33 days ago

I am sure and this is infuriating (that idiots everywhere). The only thing here is that a block is a proper block. Hiding silly/vile comments from everybody also works.

pauldevine.com•33 days ago

Saying AI doesn't work or is useless isn't so different than saying Google Search doesn't work or is useless

lolivam.bsky.social•33 days ago

@hankgreen.bsky.social you should be following this professor to get a grounded voice re:AI. He's not from silicon valley, surfaces tons of interesting information, is realistic about pros and cons, etc.

If you follow one voice on AI, it should be him.

priva.cat•33 days ago

100% on this. It's like there's two major delusional positions (AI can do everything and replace us all vs AI is useless shit), and far fewer folks going 'AI is a tool that can be useful when used correctly, and we think about outcomes'.

It's all very tribal and exhausting.

princeda.bsky.social•32 days ago

Ethan - I may prove your point, or be counter factual evidence. Either way, this doesn’t happen without that: https://open.substack.com/pub/davidaprince/p/the-january-2022-hunga-tonga-blast?r=3e655e&utm_medium=ios

pukite.com•32 days ago

Suggestion "That architectural quirk" change to "topological quirk"

catblanketflower.yuwakisa.com•33 days ago

We identified an anti-AI insult bot the other day (turns out it had that in its bio but there was some shouting first)

I have no doubt there are provocateurs who are using AI bots to polarize the discussion. Another reason it's important to be aware and not be in denial

sooup222.bsky.social•32 days ago

using AI to spam anti AI sentiment, the editor needs to get whoever is writing the script off the job man 😭 too on the nose.

tomsavage.bsky.social•33 days ago

Appreciate you surfacing good use cases and areas where LLMs do work well. I am sceptical of AI but only because the past couple of years have shown a disconnect between the first builds, the marketing and what is actually realistically achievable.

wergenga.bsky.social•33 days ago

Keep up the good work!

gallaugher.bsky.social•33 days ago

Totally agree. I am not at all an uncritical AI user, and I realize the pain caused by firms that train models on ripped-off data. But blanket dismissing upside, impact, & acceptance seems like dishonest tilt-at-windmill hoping it will all go away. Be critically engaged.

bumblefudge.com•33 days ago

I think it's tricky, the discourse as a whole needs to mature a bit. The polarization is downstream of the commercial hype-- trump-aligned technofeudalists boasting about how AI already allows them to massively downsize don't make the actual powers of LLMs popular among... people with landlords.

bumblefudge.com•33 days ago

even your example is a little scary to anyone except the most cool-headed researcher-- doesn't it kind of show what everyone's afraid of, that LLMs make propaganda cheaper to execute at scale and elections cheaper to swing? isn't that TERRIFYING to any social scientist?

bumblefudge.com•33 days ago

i think being able to tolerate more than 10 minutes/week on X is a self-selection for cool-headedness and determination... kudos to anyone who can go there every day to discuss arxiv papers and training techniques. if my dayjob required that, i would quit.

acmerocket.com•33 days ago

Please cite those studies.

move37.bsky.social•33 days ago

He regularly post them here, so you can find them on his timeline. His timeline on Twitter goes much further back with additional studies.

Comments

Posting Rules

Reply