Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
How many quarters after Q1 2016 till Marissa Mayer is unemployed?
1 or fewer
2
4
Her job is guaranteed; what are you even talking about?
View Results
 
  • Post
  • Reply
pumpinglemma
Apr 28, 2009

DD: Fondly regard abomination.

One positive application of AI I’m looking forward to is writing pointless bullshit text for pointless bullshit people. “ChatGPT: Write a three-paragraph cover letter expressing my burning desire to work at WidgetCorp until I die so they give me a job”, or “ChatGPT, write three paragraphs to gently prod an egotistical manager into realising idea xyz is bullshit because abc” type of thing.

Adbot
ADBOT LOVES YOU

poemdexter
Feb 18, 2005

Hooray Indie Games!

College Slice

pumpinglemma posted:

One positive application of AI I’m looking forward to is writing pointless bullshit text for pointless bullshit people. “ChatGPT: Write a three-paragraph cover letter expressing my burning desire to work at WidgetCorp until I die so they give me a job”, or “ChatGPT, write three paragraphs to gently prod an egotistical manager into realising idea xyz is bullshit because abc” type of thing.

Anno posted:

One part nightmarish, one part strangely compelling to watch

https://twitter.com/dendycrew/status/1620446542950993920

Arsenic Lupin
Apr 12, 2012

This particularly rapid💨 unintelligible 😖patter💁 isn't generally heard🧏‍♂️, and if it is🤔, it doesn't matter💁.


Depending on what they trained it on, there is a huge risk that it will reproduce other medical biases, such as "black people feel less pain" (a serious existing problem) and "women tend to exaggerate unimportant things .

Also I'll bet you any doctor will tell you "differential diagnosis is not the hard part".

Harold Fjord
Jan 3, 2004
But it's easier to lie to the robot about your race or BMI.
This might be a valuable alternative to WebMD for those who are used to medical discrimination

Blue Footed Booby
Oct 4, 2006

got those happy feet

Xand_Man posted:

Ehh I see a lot of people somehow surprised that when you talk to a bot designed specifically for medical diagnosis its answers will all be through the lens of medical diagnosis.

You're missing the forest for the trees. Part of being a medical professional is sorting out what details are relevant or not, and which problems are severe enough to constitute a disorder. People can't figure it out for themselves because they aren't medical professionals; that's kind of the point. So if the AI forces any and all information into the lens of the diagnoses it knows about, as if everyone who comes to it must have a diagnosis, that's actually worse than nothing.

Now consider psychiatric disorders that make the sufferer an unreliable narrator. I'm not even talking psychosis. OCD often makes people fixate on urges they don't even have. Like, they're terrified they'll poison their family without actually having the urge to do so. Figuring out which concerns are plausible requires understanding context, the very thing AI is worst at.

Applying the current state of the art to medical advice is way, way worse than using it to make crappy dialog or fanart. It's a problem tailor made to turn the shortcomings of machine learning into human suffering.

Edit: At the very least this bot hasn't had nearly enough training from cases where there isn't actually a diagnosis. The typical AI failure mode where you show a plant-identifying bot a picture of a cat and it guesses hydrangea with a 2% confidence is unacceptable for this use case.

Blue Footed Booby fucked around with this message at 17:53 on Feb 1, 2023

Civilized Fishbot
Apr 3, 2011

Blue Footed Booby posted:

The typical AI failure mode where you show a plant-identifying bot a picture of a cat and it guesses hydrangea with a 2% confidence is unacceptable for this use case.

If the use case is "replace a doctor," agreed. I can imagine this being useful as a tool a doctor might consult to compensate for the human capacity to forget or fail to make connections. The doctor can take inspiration from the bot while filtering out the bot's nonsense.

"Oh, the bot suggested the problem is overexposure to ammonia, I didn't even think about that but now that I think about it, that makes a lot of sense, I should ask the patient if that could be a factor. The bot also suggests depression because patient is complaining, obviously that's bullshit haha."

Of course the problem is that there's much more money in "replace a doctor" than "help a doctor brainstorm."

Xand_Man
Mar 2, 2004

If what you say is true
Wutang might be dangerous


Obviously saying to a gay teenager "You're probably trans or mentally ill" is unacceptable, but it's just a complex statistical model with a slick coat of paint. If the question was "What are some things I might want to screen a gay teenager for, knowing nothing else" it's not a crazy list

hobbesmaster
Jan 28, 2008

Civilized Fishbot posted:

If the use case is "replace a doctor," agreed. I can imagine this being useful as a tool a doctor might consult to compensate for the human capacity to forget or fail to make connections. The doctor can take inspiration from the bot while filtering out the bot's nonsense.

"Oh, the bot suggested the problem is overexposure to ammonia, I didn't even think about that but now that I think about it, that makes a lot of sense, I should ask the patient if that could be a factor. The bot also suggests depression because patient is complaining, obviously that's bullshit haha."

Of course the problem is that there's much more money in "replace a doctor" than "help a doctor brainstorm."

The problem is that uptodate and the like are better at this and present tons of context that is necessary to be useful for physicians.

shoeberto
Jun 13, 2020

which way to the MACHINES?
It's an example of using the wrong tool for the job because the wrong tool has a lot of zeitgeist around it, just like blockchain. There are many, many ways to model the problem of differential diagnoses with software than a generative language model.

To use an example of a problem space that's similar enough: You could use generative AI to re-implement a taxonomic tree when identifying species of plants. But why? The taxonomy exists; it's already a tree; tree structures are easy to build and allow people to navigate via UI. Would you rather have a probabilistic answer about what the plant species might be based on a language model, or a deterministic answer by following a well-defined taxonomy?

Arsenic Lupin
Apr 12, 2012

This particularly rapid💨 unintelligible 😖patter💁 isn't generally heard🧏‍♂️, and if it is🤔, it doesn't matter💁.


Xand_Man posted:

Obviously saying to a gay teenager "You're probably trans or mentally ill" is unacceptable, but it's just a complex statistical model with a slick coat of paint. If the question was "What are some things I might want to screen a gay teenager for, knowing nothing else" it's not a crazy list
"Oh, this gay person should be checked to see if they're actually trans" is a terrible call. If somebody says they're gay, you go with that, you don't need to screen for something that isn't even an illness.

Doctors have gone through many years of education in order to know what to screen for. Sometimes they gently caress up, but "maybe this person should be checked for depression" is not a hard call, and not a rare one. None of the differentials shown pick up any zebras, like "Have you considered Alport Syndrome?" (yes, I did google "rare kidney diseases" to come up with that one)

If we paged somebody from the med thread, and frankly they have enough things to be angry about, they would tell us that even when these diagnoses are appropriate, they would check for all of them as a matter of course; they don't need AI help to do it.

Nenonen
Oct 22, 2009

Mulla on aina kolkyt donaa taskussa
AI could be used in diagnostics and prescriptions for a plethora of alternative treatments like homeopathy, I think. It wouldn't do anything, but that's already to be expected if you're into those. Best outcome, all the quacks lose job!

hobbesmaster
Jan 28, 2008

Those alternatives already exist and don’t need AI.

PT6A
Jan 5, 2006

Public school teachers are callous dictators who won't lift a finger to stop children from peeing in my plane
Homeopathy bot says: drink water!

Family Values
Jun 26, 2007


poemdexter posted:

I think the big problem with all the new tech that comes out is that people with money only think about what existing things they can replace instead of thinking of new and novel things to do with it. On top of all that, the existing things usually do that thing better.

I would re-phrase it more like people are only interested in exciting ~*~disruptive~*~ ideas that might push a startup to unicorn valuation levels. 'Marketing boilerplate' or 'code boilerplate in my IDE' are not that. (I'm agreeing with you, just restating it in a more Silicon Valley buzzword way)

golden bubble
Jun 3, 2011

yospos

Mourning Due posted:

This all sounds amazing and makes me wish every game used this AI.

RE ChatGPT Vs Crypto: for me the big difference comes from ease of use, real world application, and knowing it's limitations. I've seen a lot of people saying "ChatGPT is poo poo, I asked it to write code and it came back with fundamental errors and bugs!". First of all: I've never known a HUMAN dev whose code didn't need constant rework. Second: that's a very specific use case, and ChatGPT has a lot of low-level beneficial uses. It usually requires human tweaks and intervention, but every great system currently works better as a collaboration. There's a great example in the book Hello World by Hannah Fry: they created an AI to review body scans for traces of cancer, and pitted it against a human doctor. I can't remember the exact figures, but it was something like: humans caught 89% of potentially cancerous cells, but made false negatives (said cells couldn't be cancerous when they could have been). The AI caught 91%, but made false positives (said cells were cancerous when they weren't). But when they had the AI do the first sweep, and then the human reviewed it's work, they caught over 99%.

That's where ChatGPT can be useful. Our marketing team often comes to me asking for boilerplate quotes on projects I'm working on or for general insights about our industry . It saves me time to put their requests into ChatGPT, then edit the response to be a bit more personalised from me. I can't remember which writer said, writing is hard, editing is easy, and that's certainly true for me.

With Crypto as well, the use cases were always 5-10 years away. Like: I have yet to see a commercially viable implementation of blockchain technology outside of crypto. I remember these articles about how there would never be lost luggage again, as your bag would have a UUID stored on the BLOCKCHAIN! But where are these technologies today? It's been years since this stuff was available. Whereas ChatGPT, if I go on & ask for a 10 day travel itinerary in Italy with Google Maps links for major sites: it returns it in seconds, and while it's not a perfect result, it gives me something to work with.

That's why Microsoft is so invested in it. They don't believe that it's a magical tool that will replace everything. For them, ChatGPT is next-gen autocomplete that can rapidly spit out stock sentences more effectively than their current autocomplete. But it's still just better autocomplete.

Thanks Ants
May 21, 2004

#essereFerrari


You'll be able to track AI-assistance in writing by the appearance of phrases out of nowhere that suddenly spend six months being in every piece written before the article about "have you noticed these phrases that AI likes to use too much?" appears.

Last Chance
Dec 31, 2004

all I know is that I had chat gpt format terrible lovely ancient text I copied out of a pdf (publicly available, not a private doc) and it turned it into HTML and it made proper headings, bulleted lists and it saved me like an hour.

Evil Fluffy
Jul 13, 2009

Scholars are some of the most pompous and pedantic people I've ever had the joy of meeting.

PT6A posted:

Homeopathy bot says: drink water!

:downs:: Homeopathy bot, I want real cure for this chronic illness.
:psylon:: Microdose anthrax mixed in rubbing alcohol. Take twice daily until your healing crystals resonate orange and not blue.

Nenonen
Oct 22, 2009

Mulla on aina kolkyt donaa taskussa
Okay but what if we marketed OpenAI as homeopathic psychiatrist?

Epic High Five
Jun 5, 2004



Evil Fluffy posted:

:downs:: Homeopathy bot, I want real cure for this chronic illness.
:psylon:: Microdose anthrax mixed in rubbing alcohol. Take twice daily until your healing crystals resonate orange and not blue.

Immune system recognizes and responds to the anthrax killed by the rubbing alcohol. Basically a vaccination, I call that a win!

Emzedoh
Jun 26, 2013

If it's 'properly' homeopathic, there won't be any anthrax in that rubbing alcohol.

karthun
Nov 16, 2006

I forgot to post my food for USPOL Thanksgiving but that's okay too!

Emzedoh posted:

If it's 'properly' homeopathic, there won't be any anthrax in that rubbing alcohol.

Nor will there be rubbing alcohol.

HopperUK
Apr 29, 2007

Why would an ambulance be leaving the hospital?
I wonder what the homeopathic treatment for dehydration is

Kwyndig
Sep 23, 2006

Heeeeeey


HopperUK posted:

I wonder what the homeopathic treatment for dehydration is

Nothing at all. You take a tall glass of nothing twice a day.

Boris Galerkin
Dec 17, 2011

I don't understand why I can't harass people online. Seriously, somebody please explain why I shouldn't be allowed to stalk others on social media!
https://hpathy.com/cause-symptoms-treatment/dehydration-symptoms/

quote:

For individualized remedy selection and treatment, the patient should consult a qualified homeopathic doctor in person. Some important remedies are given below for dehydration treatment:

Camphora, Veratrum album, Cuprum met, Aethusa, Arsenic album, Carbo veg, China, Lachesis, Elaterium, Secale cor, Jatropha, Natrum mur.

Prism
Dec 22, 2007

yospos

The homeopathic preparations of those would probably help dehydration if you took enough of them.

Finally, we've found something homeopathy can treat.

Emzedoh
Jun 26, 2013

karthun posted:

Nor will there be rubbing alcohol.

I think some homeopathic remedies are suspended (hah) in alcohol rather than water? I have a vague memory of my mother giving me arnica drops when I was very small and had a fall or something. In hindsight, they tasted pretty boozy though I didn't realise at the time.

Nenonen
Oct 22, 2009

Mulla on aina kolkyt donaa taskussa

Emzedoh posted:

I think some homeopathic remedies are suspended (hah) in alcohol rather than water? I have a vague memory of my mother giving me arnica drops when I was very small and had a fall or something. In hindsight, they tasted pretty boozy though I didn't realise at the time.

If it's not curable with booze, tar and sauna, then you are dying :rip:

His Divine Shadow
Aug 7, 2000

I'm not a fascist. I'm a priest. Fascists dress up in black and tell people what to do.
Worked for me when I got covid

Nissin Cup Nudist
Sep 3, 2011

Sleep with one eye open

We're off to Gritty Gritty land




https://twitter.com/TwitterDev/status/1621026986784337922

I don't use birdsite but this seems bad

sinky
Feb 22, 2011



Slippery Tilde

Nissin Cup Nudist posted:

I don't use birdsite but this seems bad

Good for hastening the death of twitter, bad for people who follow bot accounts that post pictures of kittens.

Epic High Five
Jun 5, 2004



I'm pretty sure every user who has to or, God help us all, chooses to use it as a function of their job uses a third party program to make it actually manageable so expect to hear about this if they charge out the nose. I'm sure there's business cases for both sides of the move but I'm not going to entertain the notion that there's any sensible reason this was done.

Tayter Swift
Nov 18, 2002

Pillbug
So long Endless Screaming, we knew ya all too well.

Kwyndig
Sep 23, 2006

Heeeeeey


Epic High Five posted:

I'm pretty sure every user who has to or, God help us all, chooses to use it as a function of their job uses a third party program to make it actually manageable so expect to hear about this if they charge out the nose. I'm sure there's business cases for both sides of the move but I'm not going to entertain the notion that there's any sensible reason this was done.

If the rate sheet I've seen is accurate, they're charging out the nose. It was something like $2500 for a million api calls (and it takes like 10 api calls to do anything of value in twitter).

This is all just playing into Musk's paranoia that there's a network of bots "out there" somewhere manipulating likes and retweets somehow to silence conservative voices, which is patently nonsense.

Thanks Ants
May 21, 2004

#essereFerrari


I'm sure charging an obscene amount for API calls is going to bring those advertisers back

fool of sound
Oct 10, 2012

Kwyndig posted:

If the rate sheet I've seen is accurate, they're charging out the nose. It was something like $2500 for a million api calls (and it takes like 10 api calls to do anything of value in twitter).

This is all just playing into Musk's paranoia that there's a network of bots "out there" somewhere manipulating likes and retweets somehow to silence conservative voices, which is patently nonsense.

$2500 for 10k calls. It's multiple orders of magnitude more than they can possible get away with. It's a joke, nobody will pay those rates.

Jose Valasquez
Apr 8, 2005

Kwyndig posted:

If the rate sheet I've seen is accurate, they're charging out the nose. It was something like $2500 for a million api calls (and it takes like 10 api calls to do anything of value in twitter).

This is all just playing into Musk's paranoia that there's a network of bots "out there" somewhere manipulating likes and retweets somehow to silence conservative voices, which is patently nonsense.

fool of sound posted:

$2500 for 10k calls. It's multiple orders of magnitude more than they can possible get away with. It's a joke, nobody will pay those rates.

The rates that are going around are the existing rates for a very specific API that has existed for a long time, it's not the new rate for the generic API.

Thanks Ants
May 21, 2004

#essereFerrari


The API should be free if you're just using it to interact with your own account

Main Paineframe
Oct 27, 2010

Kwyndig posted:

If the rate sheet I've seen is accurate, they're charging out the nose. It was something like $2500 for a million api calls (and it takes like 10 api calls to do anything of value in twitter).

This is all just playing into Musk's paranoia that there's a network of bots "out there" somewhere manipulating likes and retweets somehow to silence conservative voices, which is patently nonsense.

I'm pretty sure the rate sheet that's been making the rounds on Twitter is not accurate - it's a rate sheet for one of Twitter's existing APIs that was already pay-only.

The pricing for the new move hasn't been revealed yet, and I wouldn't be surprised if Musk has yet to decide on prices since he's been running things by the seat of his pants.

I think this is less about bots, and more about being desperate for cashflow and not really understanding how normal people use Twitter.

Adbot
ADBOT LOVES YOU

Leon Trotsky 2012
Aug 27, 2009

YOU CAN TRUST ME!*


*Israeli Government-affiliated poster

Thanks Ants posted:

The API should be free if you're just using it to interact with your own account

How is he going to pay that $44 billion back if he has lost 76% of his advertisers and nobody is paying for Twitter Blue if he doesn't charge 15x the going rate for API?

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply