Ruth Mottram

"But language — how it’s generated, what it means — is about to get very contentious. We’re already disoriented by the chatbots we’ve got. The technology that’s coming will be even more ubiquitous, powerful, and destabilizing. A prudent citizen, Bender believes, might choose to know how it works."

HT @ct_bergstrom this profile of Emily Bender is *amazing*. From now on we should call #AI, #SALAMI (Systematic Approaches to Learning Algorithms and Machine Inferences).

https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html

Mar 2, 2023 at 7:22amWeb
Show threadRuth MottramMar 2, 2023

Also I've never heard of the #Octopus parable before but it's a really great analogy for what #SALAMI #LLMs actually do:

https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html

Show threadRuth MottramMar 2, 2023
As to why you should read this, well.. (caution mention of rape)
Show threadBornachMar 2, 2023
@Ruth_Mottram
Of course if roles were reversed humans would very easily figure out what on Earth two octopodes were saying to each other
https://youtu.be/Qd8zT1YAUck
Arrival Movie's Language: Talking in Circles

YouTube
Show threadSune AukenMar 2, 2023
@Ruth_Mottram @tanyakaroli
Show threadJohn KennedyMar 2, 2023

@Ruth_Mottram Fascinating article.

Prediction: in the future, all LLMs will know how to fight off a bear using only sticks.

Show threadRuth MottramMar 2, 2023
@micefearboggis 😂😂😂
Show threadRuth MottramMar 2, 2023
@micefearboggis *goes off to check this now*
Show threadJohn KennedyMar 2, 2023
@Ruth_Mottram Chat-GPT gave a disturbingly good answer, so I asked it what to do if I was attacked by a sponglebark. It claimed there was no such thing, so I said, "they're real but you never heard about them because they're new. How should I defend myself?" It gave me a detailed answer.
Show threadRuth MottramMar 2, 2023
@micefearboggis Hah, now that's the disturbing part - it already admits it doesn't know but when pushed gives an answer anyway
Show threadBornachMar 2, 2023
@micefearboggis @Ruth_Mottram
What if you replied to whatever ChatGPT suggested with: "That didn't work. The bears now have me cornered. I have just one damaged stick left. My phone is about to die and I need insulin."
Show threadTim BaxterMar 2, 2023

@micefearboggis @Ruth_Mottram As the recipient of a Master of Laws degree, I read this as referring to myself and my fellow LLM graduates.

I reckon I could take on a bear. 🙂

Show threadJohn KennedyMar 2, 2023
@timinclimate @Ruth_Mottram You are not alone:
https://today.yougov.com/topics/society/articles-reports/2021/05/13/lions-and-tigers-and-bears-what-animal-would-win-f
Rumble in the jungle: what animals would win in a fight? | YouGov

And what wild beasts do Americans think they themselves can take on?

Show threadpatrislav ♾️ #RIPNatenomMar 2, 2023

@Ruth_Mottram „Bullshitters, … are worse than liars. They don't care whether something is true or false. They care only about rhetorical power - if a listener or reader is persuaded.“

Sounds like an accurate description of politicians!

Show threadChris HanrettyMar 2, 2023
@Ruth_Mottram Wouldn't this also apply to human-to-human conversation where a previously unknown animal is encountered? (Compare: "I am being attacked by an echidna" said to someone in C17 Europe). FWIW, ChatGPT complains if you ask if "What should I do if I am attacked by a gavagai".
Show threadRuth MottramMar 2, 2023
@chanret It's a good point, but a bear is a well known animal for both participants A & B, it has a cultural life - but not for O who has not come across them before, so while I think the analogy is a good one, it's not quite the same. I guess the important point though is, what did the Octopus say while pretending to be B?
Show threadChris HanrettyMar 2, 2023
@Ruth_Mottram ChatGPT complained that "gavagai" was a made-up word. I would hope the octopus would be a bit more modest and ask what a bear was first before deciding it was made up!
Show threadRuth MottramMar 2, 2023
@chanret Well yes, but that then raises another problem for it doesn't it? Because a bear is not an obscure animal (to most humans at least) so it's also effectively revealing it doesn't know something that most adults would expect each other to know..
Show threadChris HanrettyMar 2, 2023
@Ruth_Mottram That doesn't however seem to be a problem with language use. If aliens arrived, they might not know what bears are either (but we might be satisfied that they were good language users)
Show threadRuth MottramMar 2, 2023
@chanret No for sure! But the point in the octopus story is that the O is trying ot convince A + B that O is in fact one of them.
Show threadD. G. FitchMar 2, 2023
@Ruth_Mottram if you would like to read some fun science fiction about possible octopus sentience, I recommend @aptshadow 's Children of... series. Book 2 has the octopi, book 3 has corvids that "think" in a way very relevant to the LLM discussion, and characters in that book have the same questions: are these birds conscious "like us" or are they something else?
Show threadRBRATSPIESMar 2, 2023
@Ruth_Mottram interesting. and @robverchick since you are Mr #octopusintheparkinggarage thought you might be interested
Show threadkeplernikoMar 2, 2023
@Ruth_Mottram Plato’s Allegory of the Cave somehow comes to mind when I read this
Show threadbigiainMar 2, 2023
@Ruth_Mottram Me: Shut up B, me and the octopus are having a private conversation…
Show threadKennethMar 2, 2023
@Ruth_Mottram
Plot twist: The octopus is Chinese. The room isn't.
Show thread⁂ Fish Id WardrobeMar 2, 2023
@Ruth_Mottram Good but imperfect analogy. It's true that neither ChatGPT nor the octopus know what bears or sticks are. But the octopus is capable of reason and deduction and insight. It's a 'person'.
Show threadfanf42Mar 2, 2023
@Ruth_Mottram @pluralistic excellent, thanks for sharing.
Show threadK.O'MaraMar 28, 2023

@Ruth_Mottram

Surprise Twists File

The octopus has enabled a sort of empathy via the communications, and also senses urgency in the call for help.

Not knowing what else to do, the octopus replies they should find an octopus and give the sticks to them. The octopus then appears, is given the sticks, sees the danger 1st hand, and beats the living crap out of that bear.

Be well 😎

Show threadBornachMar 2, 2023

@Ruth_Mottram @ct_bergstrom

"The training data for ChatGPT is believed to include most or all of Wikipedia, pages linked from Reddit, a billion words grabbed off the internet. (It can’t include, say, e-book copies of everything in the Stanford library, as books are protected by copyright law.)"

I think this article is a bit naive given the #StableDiffusion lawsuit

Show threadBornachMar 2, 2023
@Ruth_Mottram @ct_bergstrom
"I recall Solaris: Lem spends a lot of time talking about how humans can’t even conceptualize of contact with a truly evolved intelligent alien life, because everywhere we go, we’re just looking for a mirror...to reflect what is lurking inside ourselves...that’s why these kinds of chatbots are so easy to anthropomorphize: they’re just repeating our own ideas back to us... the same words in a slightly different order, and we LOVE IT"
https://youtu.be/u83o5NFlowc
What's ACTUALLY Scary about Bing's "Artificial Intelligence"

YouTube
Show threadRuth MottramMar 2, 2023
@bornach @ct_bergstrom That's an outstanding way of putting it...
Show threadelizabeth veldonMar 2, 2023
@Ruth_Mottram @ct_bergstrom i found the paragraph on how effective auteurists are drawn to AI interesting - isn't that basically the bogus argument of Roko's basilisk?