Post AcTmCFlWhcLf3FlUNU by MudMan@mas.to
(DIR) More posts by MudMan@mas.to
(DIR) Post #AcTJaNBhrRVKLn9wUi by simon@fedi.simonwillison.net
2023-12-04T16:46:50Z
0 likes, 1 repeats
Barnstormer of an essay by Bruce Schneier about AI and trust. Worth spending some time with - hard to extract the highlights since there are so many of themA key idea is that we are predisposed to trust AI chat interfaces because they imitate humans, which means we are highly susceptible to profit-seeking biases baked into themNeeded: "A public model is a model built by the public for the public. It requires political accountability, not just market accountability."https://www.schneier.com/blog/archives/2023/12/ai-and-trust.html
(DIR) Post #AcTKMVRqS7SDrxexvs by glyph@mastodon.social
2023-12-04T16:55:29Z
0 likes, 0 repeats
@simon stood up and started clapping at “Not regulating AI, but regulating the organizations that control and use AI”
(DIR) Post #AcTKc3kHDazQaHhIUy by darkuncle@infosec.exchange
2023-12-04T16:58:21Z
0 likes, 0 repeats
@simon this was excellent, thanks for surfacing it
(DIR) Post #AcTKulhYWs467lLVJ2 by MudMan@mas.to
2023-12-04T17:01:40Z
0 likes, 0 repeats
@simon Heh. That's a fairly static view of human behaviour, I think. I don't know about you, but I'm already deep into paranoia mode, where even fairly straightforward LLM statements immediately prompt deep fact-checking. I actively *distrust* LLM speech now, and by extension things that sound like it.
(DIR) Post #AcTLWoCFhVhYrnqJWa by jami@hachyderm.io
2023-12-04T17:08:14Z
0 likes, 0 repeats
@simon I like the idea of an open-source/shared AI. But if the government gets involved, they'll just sell it to the highest bidder. At least in the U.S.
(DIR) Post #AcTN0SgsKALGoBs07E by zenkat@sfba.social
2023-12-04T17:25:14Z
0 likes, 0 repeats
@simon Great essay, loved it."We are about to make the same category error with AI. We’re going to think of them as our friends when they’re not" is the key tag line for me.Bonus points for the @cstross Slow AI call-out!
(DIR) Post #AcTNFMLYC2JDTd01A0 by admin@mastodon.futurelab.social
2023-12-04T17:27:47Z
0 likes, 0 repeats
@simon It was definitely a great read and an important one. For me one of the key points was that AI isn't the thing that needs regulation. It's corporations, made up of people, who need to be accountable for what they *do* with AI. That's what needs regulation and transparency. Trying to regulate AI itself doesn't make sense.
(DIR) Post #AcTOwkuYq0hc4LdEsS by benjaminmetzler@social.lol
2023-12-04T17:46:55Z
0 likes, 0 repeats
@simon Here's the ChatGPT summary of the article:Trust is crucial for society, but the rise of AI poses risks as people may mistakenly treat AIs as friends. Corporations controlling AI could exploit this confusion. Government regulation is essential to ensure transparent and safe AI, with penalties for misconduct. The market alone is insufficient, and public AI models are proposed to counter corporate dominance, promoting political accountability and social trust.
(DIR) Post #AcTQf6XrreabrTjDyy by ELS@sfba.social
2023-12-04T18:06:02Z
0 likes, 0 repeats
@simon For me, this is the key point of the essay: “ Chiang’s point is that this is every corporation’s business plan. And that our fears of AI are basically fears of capitalism…. the AI single-mindedly pursues the goal of maximizing production. ”
(DIR) Post #AcTQupMvUgAJDt6knA by beach@illo.social
2023-12-04T18:08:27Z
0 likes, 0 repeats
@simon Great article
(DIR) Post #AcTW4ZgBBIdjEIa33w by simon@fedi.simonwillison.net
2023-12-04T19:06:46Z
0 likes, 0 repeats
@MudMan I use ChatGPT enough now that I'm deeply worried about how advertising on a platform like that would workI'm also into the personal AI assistant idea (provided we can build them securely against threats like prompt injection, which we can't yet) - and for that I'm going to need complete transparency and control over the model it uses
(DIR) Post #AcTWIM2MhPomyPWpCy by jami@hachyderm.io
2023-12-04T17:08:57Z
0 likes, 0 repeats
@simon surely there are projects that are actually "open" ai already? anyone know them?
(DIR) Post #AcTWIMttUOdVeQZZdw by simon@fedi.simonwillison.net
2023-12-04T19:08:22Z
0 likes, 0 repeats
@jami I believe there are a few openly licensed self-hostable models emerging now with complete transparency as to the training data, but most of the best "open" models still currently keep their training data secret
(DIR) Post #AcTWXDqBSJuVtcv7E8 by simon@fedi.simonwillison.net
2023-12-04T19:10:36Z
0 likes, 0 repeats
@22 @zenkat @cstross @natebowling yeah absolutely, I've been using romance scams as an example for a while now of the kind of harmful application of LLMs that could already work terrifyingly well with the models we have today
(DIR) Post #AcTX38jXylUNK82G8W by invisv@ioc.exchange
2023-12-04T19:17:39Z
0 likes, 0 repeats
@simon @22 @zenkat @cstross @natebowling Agreed. It's something Bruce and I were worried about when we were writing this piece: https://www.wired.com/story/large-language-model-phishing-scams/
(DIR) Post #AcTdZqfYlJWiZz9NZ2 by cstross@wandering.shop
2023-12-04T20:30:31Z
0 likes, 0 repeats
@simon @22 @zenkat @natebowling See for example: https://www.theguardian.com/uk-news/2023/jul/06/ai-chatbot-encouraged-man-who-planned-to-kill-queen-court-told
(DIR) Post #AcTgB8CwiWSnb4fjJQ by katiedigc@aus.social
2023-12-04T20:59:28Z
0 likes, 0 repeats
#Ascilite23 attendees - check out this amazing blog post on the role of trust in AI. Interesting connections to our keynote today. @simon
(DIR) Post #AcTmCFlWhcLf3FlUNU by MudMan@mas.to
2023-12-04T22:07:29Z
0 likes, 0 repeats
@simon I think those are very good conversations and they need deep analysis.The bit that gives me pause on them is that the requests for regulation based on "trust" and the assumptions that deliberate and accidental bias is inherently more dangerous when presented in natural speech seems to be at odds with open, transparent models. If I'm gonna run an AI personal assistant, as you say, I want it open sourced and ran in-house, which requires free, affordable access to FOSS AI devs.