Artificial intelligence still has some way to go

the term artificial intelligence is so bad and causes so much confusion and doesnt even refer to anything in particular, tho on the other hand its obvs an incredibly provocative phrase that people really find compelling lol

all of that otm imv. (xpost rob, but also lag∞n's last)

specifically wrt trad AI (rather than GenAI), i think the vehicle example really indicates the importance of constrained environments for much automation. too open, and you introduce edge cases on which you can't guarantee 100% success. which for a fast-moving vehicle is a problem.

Conversation reminded me of this, from a while ago now, mind, which was quite good I thought (II = intelligent infrastructure)

Second, and more importantly, success in these domains is neither sufficient nor necessary to solve important IA and II problems. On the sufficiency side, consider self-driving cars. For such technology to be realized, a range of engineering problems will need to be solved that may have little relationship to human competencies (or human lack-of-competencies). The overall transportation system (an II system) will likely more closely resemble the current air-traffic control system than the current collection of loosely-coupled, forward-facing, inattentive human drivers. It will be vastly more complex than the current air-traffic control system, specifically in its use of massive amounts of data and adaptive statistical modeling to inform fine-grained decisions. It is those challenges that need to be in the forefront, and in such an effort a focus on human-imitative AI may be a distraction.

I've been trying to keep on top of it just because it's potentially a minor apocalypse for my line (government guidance) either in its chatbot forms or through subject matter experts braindumping into chatgpt and asking for a summary in plain English. Consultancy spivs selling shiny flakey solutions while rates drop and human intervention becomes dull monkeywork where you're editing reams of flat AI output and finding the hallucinations _somewhere_ in there. There'll still be work, but more boring and for less money.

Gov risk aversion means I've probably got 1-3 years before it's intolerable (and it sounds like the GOV.UK chat pilot is not setting the Thames on fire) but I'm working on the assumption that it's already shadow IT almost everywhere. Was chatting with someone over Teams the other day and I'm 90% I was getting chatgpt responses from her ("Hi, I'm always happy to talk about this and next week sounds great - are you able to expand on the text above?"). Maybe that's built in to Teams now and I missed an update?

So I pay for the Openai services and keep playing around. Where I'm stupid it's pretty good - like just giving me code or HTMl whenever I need a little, low-complexity job doing. And I can see its mediocrity wherever I'm competent. But I mostly enjoy getting into arguments with it where it lies and lies and lies to your face ('no, chatgpt, that is not blank verse, you have written a sonnet, again. Can you see why I am annoyed?') and telling Dall-E to make dog tarot cards.

My reading is usually stolen from Zvi Mowshowitz, who is at the saner end of the x-risk crowd imo, and gets a long round-up of AI links and news out every week or two. He's from that ea/rationalist/lesswrong/prediction markets world but seems actually human.

ai for customer service really meets the book definition of disruption its cheaper and worse, might not actually be cheaper tho, perfectly in tune with all the ways were getting squeezed with stuff like that, self checkout lines and so forth

Consultancy spivs selling shiny flakey solutions while rates drop and human intervention becomes dull monkeywork where you're editing reams of flat AI output and finding the hallucinations _somewhere_ in there. There'll still be work, but more boring and for less money.

this is a nailed on outcome, as is the 'shadow IT'.

oh and I set up one of the little guardrails-off llama 7B models from huggingface locally just to feel like I was getting my hands dirty. Spookier than chatgpt in a way - this weird sluggish mildly incoherent language engine stuttering to life in terminal.

xp
yup my gambit is basically playing with and reading about this stuff enough that my AI bullshit is 1 to 2 tiers above most of the rest of the guidance writing crowd & I can exploit The Fear that surrounds it. If that fails, dog tarot.

stuttering to life in terminal

it's weird that these folks who argue that the current gen LLMs have a spark of consciousness are totally fine with summoning and murdering it over and over. i'd like to see a better writer with a better imagination explore that idea further

such a weird bunch of people

https://www.404media.co/new-openai-ceo-emmett-shear-was-minor-character-in-hpmor-harry-potter-ai-fanfic/

https://i.imgur.com/MC7n7jK.png

xp

such a weird bunch of people

https://www.404media.co/new-openai-ceo-emmett-shear-was-minor-character-in-hpmor-harry-potter-ai-fanfic/

― 𝔠𝔞𝔢𝔨 (caek), Monday, November 20, 2023 1:07 PM (three minutes ago) bookmarkflaglink

for anyone who doesnt know about the ai cult that started as harry potter fan fic its def worth looking into

Maybe that's built in to Teams now and I missed an update?

Teams has been doing "suggested" responses for a while which I assume is AI driven since it does correctly deduce the tone of the conversation. it doesn't generate much longer than a few words though. maybe theres a setting??

xps
So weird! I have come round a little, a very little on Yudkowsky - he's an interesting weirdo at least - but every time I think '660,000-word rationalist Harry Potter fanfic', my god, I'd just sooner drink a septic tank.

harry potter and the methods of rationality has thrown me into a giggling fit.

_Maybe that's built in to Teams now and I missed an update?_


Teams has been doing "suggested" responses for a while which I assume is AI driven since it does correctly deduce the tone of the conversation. it doesn't generate much longer than a few words though. maybe theres a setting??

idk it could be traditional autosuggest but it felt very gpt and the following message seemed to have it going on too (featuring 'thanks! The extra context was very helpful'). I should have asked her in the meeting today.

maybe she put together a specific bot to deal with troublemakers like you (afaict teams is still bog standard autosuggest)

damn I think she really was dropping my responses into chatgpt and pasting the output. It was weird, I started writing clear flat helpful GPT prose in response.

xps
So weird! I have come round a little, a very little on Yudkowsky - he's an interesting weirdo at least - but every time I think '660,000-word rationalist Harry Potter fanfic', my god, I'd just sooner drink a septic tank.

― woof, Monday, November 20, 2023 6:15 PM (nine minutes ago) bookmarkflaglink

That said I've had this 4 fucking hour interview with him on in the background while I work so who am I to judge how much is too much.

Christ now that channel has two and a half hours with Dominic Cummings. I'm going down the wrong fucking rabbitholes.

it's just a load of rabbit shit.

as impressive as some AI stuff is I'm kind of astonished how bad it is most of the time, like my son will ask me a lot of random sports trivia questions that I'll have to search and despite these not being particularly complex questions the Bing AI not only seems to fundamentally misunderstand what it is it sometimes doesn't even answer the question it thought it was supposed to answer correctly

it's just a load of rabbit shit.

― woof, Monday, November 20, 2023 1:34 PM (twenty minutes ago) bookmarkflaglink

these people are so fucking stupid lol, tho tbf many of them are prob just scammers

wow otm

nerds are all "im the smartest person in the world, im fully obsessed with childrens entertainment"

— br◎ (@on3ness) June 24, 2019

uh I'd heard of that Yudkowsky guy and I'd also heard of the HP fanfic but I'd never put two and two together

the guy everyone cites seriously is the fanfic guy. hmm, going to have to reconcile this later

https://i.giphy.com/media/MeEWVUlrNVCGk/200w.gif

^^^when u spellcheck "basilisk"

so i do copywriting for side money. that was the career that chat gpt was supposed to eviscerate. i actually have tried to integrate it into my work -- like having gpt 4 write a rough draft, or reword an awkward sentence I am stuck on -- but it seems to suck even at this.

imo the secret sauce is training it on non-public data so it can do domain-specific things, and you can set it up to only give verbatim quotes/answers based on what you’ve supplied it

My company had to finally issue an AI policy and outline what AI use was acceptable and when.

Probably because agents began using them when they got stuck on phone calls only to get misinformation from the chatbot.

Or executives using it to write proposals, leading our legal team to freak out about "giving away proprietary secrets that may be later leaked by AI tools"

lmao

I just sat in a conference room with four people of roughly my age and roughly my background (professional
writers) and all of them had used chatgpt.

I have rarely felt more alone.

I have been on the top of a mountain in January and I have been in a small sailboat on the Atlantic Ocean. That conference room made me feel more isolated and weird.

https://i.imgur.com/Kq1vJha.png

lmao

"There'll still be work, but more boring and for less money."

Maybe. If this stuff keeps working in terms of quality of results but idk if it will.

I feel that its just as likely "people who actually know something" could make it at overinflated prices in a few years.

I think both those things can be true - that there'll be money at the top end of trades but the bread-and-butter middle gets slowly eaten away.

progress/quality may stall but my version of the working world looks very different if it doesn't - I'd be a fool not to have at least a saver on 'extreme change'.

xp lagoon lolol
screenwriters will adapt to be un ai-imitable

tell me how

I used to work as a commercial translator (French to English) and I'd say this is one thing Chat GPT does pretty well, much better than Google Translate. I'm guessing just about anyone working in translation (in the more common languages at least) is using Chat GPT and I'd be surprised if the word rates haven't gone right down.

apparently even google translate which produces terrible results took a big bite out of the industry cause it was good enough for some things

i was under the impression that google translate was already using llms to help it improve its translations for at least the past few years

yeah, that's the model & I think translation was the harbinger - like people will decide gen ai is good enough to turn out some functional copy and they just decide not to use a freelancer unless it really matters.

Some suggestions that it's starting -

https://archive.is/qMJUg

yup

oh god for the innocent joyful days
Neural Network, what u eat?

Xps Tracer the big jump in quality in 2016 was down to neural net/deep learning stuff but idk if that’s equivalent to llm/gen ai methods

whatever you think of OpenAI and Altman, firing him seems to have been an incredibly dumb decision by the board who are now facing the resignation of basically the entire fucking company.

(also, it doesn't sound like there was anything else behind it...not because of abuse allegation, not because he was in the midst of doing anything illegal... just a board of impetuous and junior people who didn't think through the ramifications of their actions)

Insert New Yorker cartoon with the police officer saying: "Does your car know why my car pulled it over?"

whatever you think of OpenAI and Altman, firing him seems to have been an incredibly dumb decision by the board who are now facing the resignation of basically the entire fucking company.

― I? not I! He! He! HIM! (akm), Monday, November 20, 2023 6:03 PM (thirty-seven minutes ago) bookmarkflaglink

it was dumb but also very cool

they still havent said wtf he did they accused him of lying what was it tell us

i thought it was basically that he said he supported guardrails and safer advancement of the research but then turned around and did pushed for faster development

"did pushed" sorry this post was written by chatGPT. "did the exact opposite"