Artificial intelligence still has some way to go

re: bananas, there were some weird banana shortages and price distortions caused by amazon basically giving away bananas in seattle

that shit is

not gonna say it

i love the amazon banana saga, just the fact that they were like lets do something nice well give people bananas, its perfectly odd

The tragedy of the banana commons

i imagine the execs were like "what is some 'low hanging fruit' kind of initiative we could do?....wait a second!!"

guy holding banana has good idea involving bananas

lol zach

*conference room all turns heads to look at surprised guy with banana in mouth*

I think it’s because their logo looks like a wang, I mean banana

Also they give free bananas away at HQ2 here in Northern Virginia

They do not given them out in New York, not even to employees. Fml.

sounds like the supply chain isn't completely flushed out. too many bananas in one place, a dearth in another. banana optimization has a ways to go

I've always wanted* to have a trading business that involved selling bananas to Australia, and buying boomerangs in exchange. The kicker is that you can use the same packaging both ways.

* = I am lying. I have not always wanted to do this.

Do they accept returns on the boomerangs? Haha get it?

This was the conversation that put me over the edge to delete my GPT plus subscription today

https://ibb.co/MGW9zwW

Trying that again
https://i.ibb.co/YyYb5nY/theplayer.jpg

You have to open it in a new tab and zoom to read it.

its cool

That's crazy, man alive. To clarify, the secretary in "The Player" is played by Dina Merrill

why do you guys keep trying to ask it about specific facts? that's the thing it obviously can't do

it's also the #1 thing people keep thinking it can do, I guess

IME, most of the time it actually did get specific facts right.

i think a lot of people's model of intelligence is that it's a synthesis of facts (which doesn't seem nuts to me), so a system that appears intelligent (and is sold as such) must know facts, and a system that does not know facts is ipso facto not intelligent.

https://lareviewofbooks.org/article/how-to-raise-your-artificial-intelligence-a-conversation-with-alison-gopnik-and-melanie-mitchell/ was pretty good on the philosophy/cogsci aspects of these systems.

Well I less assumed it "knew" facts and more that it was capable of finding them.

you could use a licensed version, specifically train it on that data (say, IMDB or some history books, whatever) and prompt it to use the information verbatim. what the version on the web is doing is responding with what the language model has trained on for the purpose of structure and language

My theory is that something threw it off about the phrase "Mill's secretary" because she's credited as "Jan." There probably aren't tons of different texts in the training materials that describe her as "Mill's secretary," so, not having the capacity to actually watch and assimilate a movie, GPT had nothing to go on and just randomly spit out actors from the cast (although it's bizarre that Angela Hall is never one of them even after dozens of tries). The role itself is also probably not mentioned that often in discussions of the film, as it's pretty minor. But I still find that kind of fascinating as a mistake, because as it assimilates more and more material and gets more "compute," it gets better and better at spitting out answers to pretty complex questions, but it's still always in a sense stumbling onto the right answer when it gets things right, even if that stumbling process becomes more and more accurate.

The other funny thing is that you can ask it leading questions with fake information in them (my idea in asking who played "The Secret Service Agent" in Nashville, since I'm pretty sure that, if you see one at all, it's just an extra and not an actual role).

*that you can do that and it will answer as though your question makes sense

The other day I asked it who said [a phrase I made up] and it credited it to Terry Eagleton. But that doesn't always work, sometimes it will respond that the phrase you have typed doesn't appear to be attributable to anyone famous.

that LARB interview is really good, thanks for posting

https://i.imgur.com/JejokME.jpeg

"written by humans, for humans"

nice to see some job creation

Yeah, you ask a leading question based on bullshit like "Why did Virago Modern Classics have such a hit with John Norman's Gor novels?" and it merrily prattles away with reasons based on total fantasy.

Taiwanese farmers were forced to grow less rice because the AI chip manufacturer needed water.

h/t @SashaMTL https://t.co/bY6um8Ok4M

— Remmelt Ellen 🛑 (@RemmeltE) July 21, 2024

yeah love how we’re all like “let’s dump
potable water in the trash to trick the silly machine”

stop being so weird and giving this bullshit attention

https://www.wired.com/story/ai-is-already-taking-jobs-in-the-video-game-industry/

My youtube feed has been filled the last couple of days with videos on how AI has peaked, AI is in a bubble, as as plateaued (There wasn't really any content on AI there before that I noticed). The plateauing angle seemed to make sense, with chatgpt 4 not obviously much of a difference from 3.5

tough out there for a tech hype cycle

https://www.sagaftra.org/sag-aftra-members-who-work-video-games-go-strike

Striking over AI protections

AI has ruined search for many topics because all the results are AI generated garbage. I tried to search for how Java handles multiple inheritance but all of the non-Oracle articles had the wrong answer -- taking the C++ answer and pretending it's the same in Java. Even giving example code with output that is blatantly false. (The correct answer I believe: diamond problem won't compile; priority is class > interface then subclass > superclass; use the syntax Interfacename.super.f() when the priority is the same.)

ChatGPT: Here is a list of all 47 European capitals ordered from west to east:

Lisbon, Portugal
Dublin, Ireland
Reykjavik, Iceland
Madrid, Spain
London, United Kingdom
Andorra la Vella, Andorra
Brussels, Belgium
Amsterdam, Netherlands
Luxembourg City, Luxembourg
Paris, France

pic.twitter.com/qpruVXpzKH

— shardcore ⧖ (@erocdrahs) July 31, 2024

Some of those in open sources
Are the same who train on corpuses

lol

lmao

I'm really amazed how quickly suno and udio have developed. Despite the stunni g results however I can't imagine anyone caring about the songs it can make

xxp lool

IDK if I'm just reading what I know into them, but the missing element seems to be a sense that there was a writer or performer or even engineer in there somewhere who actually cared about the song. It gives them a very throwaway quality.

Still pretty amazing what it can do.

the amount of raw computing power that has been organized and applied to these programs is extremely impressive and the results are also amazing compared to what was being accomplished just a decade ago. but every recent attempt to apply this technology to fill the needs of the general public has been pretty weak sauce. Google's ads for its new AI interface wants me to believe that "summarize this email" is a compelling application for this wondrous technology. nope

Gary Marcus has been ruthless about that. He’s a fun follow for generative AI skeptics like myself