Artificial intelligence still has some way to go

do people really call him “sama”? like they say sama out loud?

ai sama bin laden

gen z loves both samas

ai sama bin laden

― flopson, Sunday, November 19, 2023 5:28 PM (fifty-eight minutes ago) bookmarkflaglink

lol

Today I resigned from my position as CEO of Cruise. (1/5)

— Kyle Vogt (@kvogt) November 20, 2023

was this forced? they were banned from the road in SF after this accident and maybe lying to the DOJ.

Sorry DMV

cruise has made a series of quite serious mis-steps recently (if you don’t assume the entire project of autonomous driving isn’t a mis-step) so yeah i’d say it was forced and i’m imagine there was a tidy sum behind it.

and altman now with microsoft. the amount of money they must have thrown at him to do that. still, it’s money they’ve got and it made sense to do it and do it quickly. the risk of a new start up would have been huge.

I found this quite moving - and it has an idiotic quote from sam altman!

https://www.newyorker.com/humor/sketchbook/is-my-toddler-a-stochastic-parrot

ai sama bin laden

― flopson, Sunday, November 19, 2023 5:28 PM (yesterday)

*slow clap*

the nighthawks thing is a masterpiece of tone. love his update

https://x.com/soncharm/status/1726305454580756866?s=20

UPDATE: Ok got a chance to skim some replies here, many seem to thing I ‘missed the point’ of the original. But I really don’t agree remember, the AI parses & detects what’s in it (into ‘tokens’) not me. All this info is then being used as a ‘seed’ for the algo. Make sense now?

— sonch (@soncharm) November 19, 2023

It's a good joke but it's crystallised one of the things I won't miss about Twitter - you can just do the joke, you don't have to reply to all of the replies in character and draaaaaaag the joke to death.

Or: jokes good, 'bits' bad

and altman now with microsoft. the amount of money they must have thrown at him to do that. still, it’s money they’ve got and it made sense to do it and do it quickly. the risk of a new start up would have been huge.

― Fizzles, Monday, November 20, 2023 4:40 AM (five hours ago) bookmarkflaglink

when you have an opportunity to light a tiny part of your huge pile of money on fire you gotta take

now 500 of openais employees have signed a letter threatening to go with sam to microsoft unless hes instated, this is funny now but itll be even funnier after the inevitable ai crash

https://pbs.twimg.com/media/F_YbOFxWcAAkIlj?format=jpg&name=medium

the ruling class just doesnt hit like it used to

there is one way microsoft can salvage its investment, one perfect use case, something so revolutionary...

https://i.imgur.com/Qq4puCe.png

clippai

"the nighthawks thing is a masterpiece of tone. love his update"

Really funny as I have a reproduction of Nighthawks. Can't look at it in the same way now.

nighthawks no thanks, dayhawks

ah, that provides the context for this, which made me laugh the other day:

https://x.com/kendrictonn/status/1725296651944751237?s=20

lol i love the anteater nighthawks had totally forgot about that

hahaha that one is great

(its not ai if thats unclear its been floating around the net for a while)

speculation on why microsoft likes ai, makes sense imo, i have my doubts that ai will ever be more than a niche or middling contributor, and its so resource intensive that getting people (or more realistically companies) to pay enough for it to make it profitable is dicey, but microsoft can certainly afford to bet on it

Generative AI might be transformative or it might be another flash in the pan. Either way, the partnership with OpenAI still gave it an advantage over AWS and Google to drive customers to its Azure cloud platform because all those AI tools require a lot of computing power — and that’s really what Microsoft wants to sell people.

https://www.disconnect.blog/p/how-sam-altman-plays-into-microsofts

full disclosure i do pay for the ai product github copilot, its helps you write c0de, you can ask it to write you a whole thing as weve seen in demonstrations of chatgpt but i dont think most people use it that way, where it really shines is code competition ie you start typing and it finishes the line for you, its pretty good! totally not trustworthy tho, you have to check its work

that kind of more focused use of the technology or series of technologies that is for some reason called artificial intelligence seems more realistic than the super wide consumer focused let chatgpt write a term paper for you approach

at $10 a month i have no idea if copilot is profitable for github and their parent... microsoft

it is hard to decouple the marketing hype from the industries actual plans, obvs its good for them if theres a million articles being all ai will change the world even if they dont actually believe it

i'm pretty bullish on generative AI.

like a lot of machine learning technologies, the binary will it/won't it discussions in much of the media often seem detached from the fact that it's already being used successfully in all sorts of industries. speeding up and facilitating the creative ideation process to get more quickly to the point where your concept has enough substance for people to give it a thumbs up or a thumbs down is one example. even in my own area we are using it for quality control purposes, enabling us to more quickly spot serious issues quicker. it's also opened up the ability for non-coding people to describe and generate applications based on concepts, which means more people are able to participate in formerly quite technical areas, reducing the bottleneck to innovation.

a wider lens shows that self-driving vehicles are very successfully used in constrained industrial environments.

the constraints of clean data, algorithm optimisation and GPU are real though, with investment into any of those becoming more demanding for uncertain incremental revenue return - as lag∞n has said, the long-term business model seems extremely uncertain.

everything ive seen ai really generate on its own is so awful to the point of uselessness, at least from my pov but im not some marketing manager maybe they like that shit idk

and yeah profitability is where the rubber meets the road, lots of things that are nice to have eg uber when theyre being subsidized by massive investments are not as nice when you have to pay full price for them

yes, it's unsurprising in retrospect, but coding does seem like an unambiguously useful application for gen AI. And there's a lesson there, imo, in the stark difference between programming languages and regular human ones.

"a wider lens" = yeah this is where the sudden, recent conflation of "AI" with "generative AI, likely text or image generating" is unhelpful. People saying "AI is doomed" are likely not talking about, say, the use of digital twins to do safety checks in the Korean shipbuilding industry or supply chain optimization or etc. I am not at all bullish on autonomous vehicles in the real world, but Fizzles' example sounds like a continuation of the roboticization of industrial environments that began decades ago. I mean, is a Roomba an AI?

even for coding it kind of doesnt make sense for a number of reasons for anything bigger than a snippet IMHO

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.