Post ARmYu4H6A3lTHraqp6 by zaratustra@mastodon.social
(DIR) More posts by zaratustra@mastodon.social
(DIR) Post #ARmSODQZzoc4gOSMoi by clive@saturation.social
2023-01-18T21:52:56Z
0 likes, 0 repeats
Since ChatGPT news continues to come fast and furious, I'll re-up this essay from last month"On Bullshit and AI-Generated Prose": https://clivethompson.medium.com/on-bullshit-and-ai-generated-prose-611a0f899c5In which I consider the output of ChatGPT -- bloviatingly confident, even when studded with errors -- through the lens of Harry Frankfurt's theory of bullshit(Here's a "friend" link if you're not a Medium subscriber: https://clivethompson.medium.com/on-bullshit-and-ai-generated-prose-611a0f899c5?sk=e3ed6e4620d963cdda59282d5137e223)
(DIR) Post #ARmTZ5jx1CcnyoGdGa by darkuncle@infosec.exchange
2023-01-18T22:05:35Z
0 likes, 0 repeats
@clive there's nothing more dangerous than being right most of the time and confident all of the time.
(DIR) Post #ARmUin2lUDocqMFvDE by quanta@attractive.space
2023-01-18T22:19:05Z
0 likes, 0 repeats
@clive Your description of AI bloviating perfectly fits Boris Johnson. OMG! Our former Prime Minister is really an AI.
(DIR) Post #ARmVH2fpsdWBAJrbMG by PlasticJohnny@mastodon.ie
2023-01-18T22:25:14Z
0 likes, 0 repeats
@clive ChatGPT is trained off of white men.
(DIR) Post #ARmX97tqlY3kxulutc by clive@saturation.social
2023-01-18T22:46:14Z
0 likes, 0 repeats
@quanta Alas, it seems so
(DIR) Post #ARmYnAV8c4gWZSs1Hk by clive@saturation.social
2023-01-18T23:04:39Z
0 likes, 0 repeats
@darkuncle Lord, yes
(DIR) Post #ARmYu4H6A3lTHraqp6 by zaratustra@mastodon.social
2023-01-18T23:05:28Z
0 likes, 0 repeats
@clive also: that IQ test that someone ran on ChatGPT? it's a series of 40 true-and-false questions. that's it
(DIR) Post #ARmgNVjFrjvuVmkYLo by rami@pnw.zone
2023-01-19T00:29:22Z
0 likes, 0 repeats
Some, like Prof @emilymbender, say that this isn’t even AI. I tend to agree.
(DIR) Post #ARmjWeDLzCQfIVsPtA by clive@saturation.social
2023-01-19T01:04:33Z
0 likes, 0 repeats
@zaratustra yepHell, just ask GPT-3 a question involving *trivial* math and it can't handle itHere's one I just did
(DIR) Post #ARmsy56yzRNK7l3LkG by GregStolze@mastodon.social
2023-01-19T02:50:48Z
0 likes, 0 repeats
@clive Didja see the description of it as "mansplaining, but you pay for it"?
(DIR) Post #ARmu1l4dX30f5fGyQK by clive@saturation.social
2023-01-19T03:02:40Z
0 likes, 0 repeats
@GregStolze omg nomansplaining as an API
(DIR) Post #ARnKMJOYEYu68T362q by tommi@mastodon.social
2023-01-19T07:57:42Z
0 likes, 0 repeats
@clive You make a great point about language models. What worries me is, as you pointed out, we have segments where confident bullshit is already the norm, and it has been expanding lately. We've been really struggling when political leaders have broken new ground on how much bullshitting is possible, and mis- and disinformation have been weaponized by antidemocratic actors. As these operations often are based on quantity of text, there is a nefarious market for these tools.
(DIR) Post #ARo5vU8SzKiJQancHI by PlasticJohnny@mastodon.ie
2023-01-19T14:44:09Z
0 likes, 0 repeats
@clive Ya know, I was joking, but I wonder if they have a racial / gender breakdown of training data. My flippant remark was probably correct.
(DIR) Post #ARo5vUicotdZEjiUdc by clive@saturation.social
2023-01-19T16:50:34Z
0 likes, 0 repeats
@PlasticJohnny It's a good question! I don't know ... they crawled all the main stuff online, and in English it may be well disproportionately white dudes
(DIR) Post #ARo6cVSBxmB1y5XWnQ by clive@saturation.social
2023-01-19T16:52:33Z
0 likes, 0 repeats
@tommi Yes, this is something Gary Marcus, a friend of mine who's a prominent critic of the over-attribution of intelligence to large language models, worries about this too