Using OpenAI's new, powerful "reasoning" model to calculate the amount of states that have R in them. Welcome to the future, everybody - ThreadSky

shaneanderson.bsky.social•85 days ago

Cc @gavinmichaelbooth.bsky.social

reply

gavinmichaelbooth.bsky.social•85 days ago

Why? Why!!

reply

amandajanestern.com•85 days ago

I thrink you mean wrhy

reply

gavinmichaelbooth.bsky.social•85 days ago

Don’t tell my AI how to think!

reply

blueskiesatnight.bsky.social•85 days ago

Rorry, Ri rased rit ron row rey’re ronounced

reply

drewmelbourne.com•85 days ago

The other 16 states can't contain an R. Rs are powerful.

reply

prezmullet.bsky.social•85 days ago

Good to know OpenAI says “Warshington.”

reply

grogofgrog.bsky.social•85 days ago

just to check if it's looking at political affiliation there too.... can you run same test with states that have D?

reply

mezzomondo.bsky.social•85 days ago

Lol

reply

mvthomster.bsky.social•85 days ago

It’s horrible at processing structured data without a lot of semantic help.

reply

spencercallaghan.ca•85 days ago

the worst thing we did as as society was start letting tech CEOs redefine the meaning of words.

They get to call this "reasoning" in the same way words like "free", "open" and "privacy" no longer mean what you think they do.

reply

gardylou-games.bsky.social•85 days ago

Not to mention how they’ve redefined AI to go from meaning a life like android or something that can think like a human to a whatever these chatbot thingys are. And I know that AI is broader than that, etc, but the most common thing people think of now are chatbots as what AI means.

reply

wigglin.bsky.social•85 days ago

Artificial intelligence has been a field since the 1950s, and it's never been defined as a "life-like Android."

reply

gardylou-games.bsky.social•85 days ago

Yes, you are technically right. I didn’t word it well but I mean popular consensus in terms of cultural use of term. I just mean how much even our collective imagination has been degraded. Like an unreliable chatbot has become our meaning of AI, a big downgrade from the interesting visions of it.

reply

spencercallaghan.ca•85 days ago

what we are seeing with LLMs is not intelligence by any definition of the word. It is a giant database stringing together probabilities at high speed.

reply

wigglin.bsky.social•85 days ago

I see what you mean. Personally, I recall using chatbots like Eliza-clones, or even Google Assistant/ Amazon Alexa, and always being disappointed by them. So when ChatGPT came out, I was truly amazed that they finally worked. Still am amazed.

reply

gardylou-games.bsky.social•85 days ago

I was really impressed at first, but over time it felt like i was more impressed by the speed than the results. The immediacy of it is wild, but I don’t find it as useful as other smart people when it comes to real issues. My fav use was messing around, but I’d rather just play video games instead.

reply

wigglin.bsky.social•85 days ago

What is this if not "reasoning"? What would be a better word for it? https://chatgpt.com/share/67525cbc-863c-8013-a5cc-f1959e58bd25

reply

rsmjr.bsky.social•85 days ago

"Why are people misunderstanding our use of the term 'full self-driving'"?

reply

thesituationroomba.bsky.social•84 days ago

Me trying to explain why AI wasn't able to solve a simple sounding math equation when it was a word problem format. "Education matters, context matters"

reply

docvivileandra.bsky.social•85 days ago

The reason LLMs consistently fail at this is because they never actually see words as collections of letters - every word (and permutation of a word) is assigned an ID, and the LLM operates on the IDs

It has no idea what letters are in ANY word, because it doesn't know the actual words

reply

docvivileandra.bsky.social•85 days ago

The fact that the LLM will give confident answers to questions about what letters a word contains despite, again, having NO IDEA WHATSOEVER what letters are in which words, showcases how they are fundamentally bullshit engines - they ALWAYS just try to give the response they think you're looking for

reply

docvivileandra.bsky.social•85 days ago

"Think", of course, being not really applicable here, as they don't actually think. They just predict the expected response, via some complicated math that amounts to an extension of lines-of-best-fit algorithms to arbitrary numbers of dimensions.

reply

csalem.bsky.social•85 days ago

I like this explanation because it highlights the flaw in this demo (the machine work on tokens, not letters) and the flaw with the machine (it appears confident while having no insight into the answer)

reply

ultimatefoot.bsky.social•85 days ago

Thanks for the explanation lol. Was wondering why they struggle so bad with this kind of problem

reply

menzer.bsky.social•85 days ago

Yup, if you ask an LLM

"How many R's are in Food?"

What the LLM actually gives you is an answer that more closely resembles

"On average, how many R's are there in words that are kind of similiar to 'Food'?"

reply

jaftt.bsky.social•85 days ago

that's not even close to how it works

reply

neilstewart.bsky.social•85 days ago

- R

reply

geographer.bsky.social•85 days ago

Minnesotar is lovely, even this time of year I hear.

reply

gardylou-games.bsky.social•85 days ago

I can’t completely shit on ChatGPT. Sure I tried using it for 6 months and even its coding help was so hit and miss it really didn’t make me more efficient, but one time it hallucinate that Tatako Mamiya was part of Tin Pan Alley, which accidentally introduced me to Hauromi Hosono—who is amazing!

reply

gardylou-games.bsky.social•85 days ago

* Haruomi Hosono

reply

uscivilwar.bsky.social•85 days ago

[I deleted me reference to South park because this is really funny to me for some reason and I didn't want to double-post:]

Connerticut, Massachuserts, Mirchigan, Minnersota, Mirsirsirpi, Montarna, Nevarda...

reply

tochsenbein.ch•85 days ago

Maybe they're counting republican states?

reply

dngrs.bsky.social•85 days ago

Pretty sure that's what the first two screenshots imply yeah

reply

tochsenbein.ch•85 days ago

Sorry, it's early morning over here

reply

dngrs.bsky.social•85 days ago

insomnia, Germany here 😅

reply

dissentinator.bsky.social•85 days ago

“Now, let’s try screening mammograms for cancer…”

reply

jaftt.bsky.social•85 days ago

AI Detects More Breast Cancers with Fewer False Positives https://www.rsna.org/news/2024/june/ai-detects-more-breast-cancers

reply

technicallyowen.bsky.social•85 days ago

That's the problem with using the term AI for everything a neural network based Computer vision system is not the same as an LLM bullshit engine.

reply

jaftt.bsky.social•81 days ago

LLMs also have vision

reply

technicallyowen.bsky.social•81 days ago

Yes vision language models exist they combine vision encoders and LLM they can't see anything they can just generate madlib bullshit based on images as well as text https://www.nvidia.com/en-us/glossary/vision-language-models/

reply

jaftt.bsky.social•81 days ago

reply

the-greetest.bsky.social•85 days ago

Not even close to the same thing.

reply

jaftt.bsky.social•81 days ago

Then why did the person I reply to bring it up

reply

the-greetest.bsky.social•81 days ago

Because it was a joke. LLMs are not machine vision programs. We've already had machine vision programs doing these things since the 90s. They're just getting better and since AI is being hyped beyond belief, we get incredulous articles pumping incremental change as something new.

reply

jaftt.bsky.social•81 days ago

This seems pretty cool

reply

eviemae.ink•85 days ago

just absolutely revolutionary technology

reply

csalem.bsky.social•85 days ago

Masherchusetts, Michergan… check out

reply

jeof.bsky.social•85 days ago

Ah yea Ryoming, where the bread comes from.

reply

dededeminion.bsky.social•85 days ago

Isn't this WORST than the preview version?

reply

mnorac.bsky.social•85 days ago

the future

reply

real-xero.bsky.social•85 days ago

Look
If you just add an R at the end, then the machines are right
😬

reply

yankeefinn.bsky.social•85 days ago

The Rs are not just silent, but also invisible

reply

eaton.fyi•85 days ago

I wonder whether that list was skewed by the number of times people have posted lists of states followed by an ‘R’ or a ‘D’

reply

edzitron.com•85 days ago

What's funniest for me about this is that this is effectively one model checking another model's work, and both got it wrong, all while superheating a GPU. Truly the future is here, we are mere steps from super intelligent computers

reply

itsnotbread.bsky.social•85 days ago

Would settle for billionaires with above room temperature IQs at this point

reply

painedumonde.bsky.social•85 days ago

Stupidly burning alive while murmuring nonsense burbles away in our ear pods.

Welcome to Costco, I love you.

reply

dangyd.bsky.social•85 days ago

Sam Altman abandons frivolous pursuit of a computer that can love him, pivots to “we’ve solved the global heat crisis”

reply

81horses.bsky.social•85 days ago

Definitely makes me feel better about Amazon's plan to build SEVEN NUCLEAR REACTORS on the banks of the Columbia River at Hanford in order to support their server farms

reply

rufusking.bsky.social•85 days ago

With how long even the pre-fab reactors they are planning on using take to get up and running, there is not an insignificant chance the whole bubble pops before the first one even comes online.

reply

81horses.bsky.social•85 days ago

Perfect! Because WA State needs more abandoned nuclear reactors. They're fun for urban ruins explorers.

reply

rackletang.bsky.social•80 days ago

Satsop represent

reply

tubbles.bsky.social•85 days ago

You can get first year programming students to write up an algorithm to check this. It's so funny that these supposed "AIs" still can't solve it despite billions and billions of dollars thrown at it.

reply

carolinequinn.bsky.social•85 days ago

What could go wrong….?….

reply

boojit.com•85 days ago

an app that should exist: when you get that AI spambot sending you an SMS, you should be able to point this app at it, and it just starts sending ChatGPT messages back to the AI spambot, and so you got these useless bot conversations carrying on en masse, melting ChatGPTs servers to the ground.

reply

carljanderson.bsky.social•85 days ago

All they need is a smidge more time and money... 😉

So much waste on a calculator that can't give a consistent answer (nevermind accurate) to 1+1.

Different outputs from the same models while using the same inputs...

These failures are trying to become too big too fail before the crash

reply

255.bsky.social•85 days ago

finally we've almost made God

reply

shawnguerra.bsky.social•85 days ago

We already did that.

reply

johnkelly.bsky.social•85 days ago

Maybe Gemini isn’t bragging about its reasoning model so it gets a pass this time, but it was interesting asking it about yesterday’s big news. So confident in its answers. And yet so very wrong.

reply

rackletang.bsky.social•80 days ago

I mean, technically the murder that hadn't happened yet was not solved a year before it happened. So for a LLM, that's not bad! (Nah it's still bad. It's all bad.)

reply

johnkelly.bsky.social•80 days ago

😂

reply

wenchoheelio.bsky.social•85 days ago

Sam A needs to get his non profit billions sorted quick and run for the hills

reply

wasteof.technology•85 days ago

it's always two dumb chiplets telling each other "exactlyyyyy"

reply

dirigiblist.bsky.social•85 days ago

Fuck lol

reply

burlaprobot.bsky.social•85 days ago

or in this case "rexactlyyyyyy"

reply

harlansd.bsky.social•85 days ago

😂😂

reply

kipenyan.bsky.social•85 days ago

It’s funny that there is a known solution to this (arbitrarily small tokens), but the computational cost to both train and run a model that granular is so far outside the realm of feasibility that it’s not even worth entertaining. Almost like this paradigm is intrinsically ill-suited to reasoning!

reply

psynautic.bsky.social•85 days ago

reminds me of a famous old joke: How many 70,000$ gpu's does it take to count the number of R's in Mississippi?

reply

defiantjazz.bsky.social•85 days ago

But, does it know how many "r"s are in "strawberry" now?

reply

hyr.mn•85 days ago

That is totally worth 200 a month! I was just planning to throw that money into a hole but now I can do that and get bad information and global warming too.

reply

waytogopaul.bsky.social•85 days ago

AI Voice: This is a classic example of the sunk cost fallacy.

reply

the-greetest.bsky.social•85 days ago

Don't you mean the surnk cost fallarcy?

reply

waytogopaul.bsky.social•85 days ago

“I apologize for my blunder. Thinking about it more carefully, I agree the answer is 42.”

reply

nuthinking.bsky.social•84 days ago

Shocking these advanced reasoning models don’t generate code as a step to test simple things like this. 🤯

reply

qf-security.co.za•85 days ago

These 'scientists' were to preoccupied with where they could or not that did not even bother to thing whether they should. I am sure all will 'just be fine'... thanks Ed!

reply

clamheaven.bsky.social•85 days ago

I notice it left out Morth Carolina. Sad.

reply

christosilvia.bsky.social•85 days ago

So "R" in Massachusetts is pronounced "a" so this is technically correct

reply

soccerlawyer.bsky.social•85 days ago

Better or worse from Meta?

reply

flootdraws.bsky.social•85 days ago

Hands up who wants to visit Ner Mexicer

reply

geographer.bsky.social•85 days ago

It's equally bad just in different ways. It's incredible really.

reply

nwp.bsky.social•85 days ago

What a load of Rs.

reply

m152.bsky.social•85 days ago

Perrsylvania

reply

cargoweasel.com•85 days ago

Correcticut

reply

sedward5.com•85 days ago

There is an r in the letter n if you cover up part of it. Check and mate.

reply

audaxdreik.net•85 days ago

Warshington belongs

reply

sneakys.bsky.social•85 days ago

If you look at literal letters, no. If you were to ask my late grandmother who grew up in DC in the 1920s and 1930s, yes.

reply

skimlines.bsky.social•85 days ago

lol I saw the "how many r are ther ein strawberry" but this is a new level

reply

oldpappythomas.bsky.social•85 days ago

YEAH I'M FROM FAHCKIN MASSERCHUSETTS

reply

m00psy.bsky.social•85 days ago

#alt4me

reply

yingtai.bsky.social•85 days ago

#alt4you

reply

difficultopinions.com•85 days ago

Using the OpenAI playground interface, models can be compared side-by-side. Here I selected gpt-4o and the Sept version of o1. Both fail with the inclusion of South Dakota. Sorry for the small screenshot, there's no page share in the playground.

reply

tawmjawn.bsky.social•85 days ago

It’s Mrs. Sippi

reply

erixsummerdown.bsky.social•85 days ago

Ah of course Wiscornson, Minnesorta, and Missrissippi have Rs in them.

But it's exactly one R and they have to share.

reply

variantz.bsky.social•85 days ago

ummm I can do that in notepad++

reply

imiuiu.bsky.social•85 days ago

time to write some deeply serious pseudointellectual pseudoecon babble... that oughta boost the stock

reply

andrew-d-evans.bsky.social•85 days ago

reply

flootdraws.bsky.social•85 days ago

Don't you mean Massarchusetts, Michirgan, Minnesotar and Missirsippi?

- AI bros, probably

reply

orphanendorphin.bsky.social•85 days ago

Is it equally bad at all letters, or uniquely bad at R and D because it's read something like 'Senator Wyoming (R)' in 28,000 news articles?

reply

flootdraws.bsky.social•85 days ago

That might explain a lot...

reply

the-greetest.bsky.social•85 days ago

It is literally just a probability machine that spits out token combinations that are most associated with the input tokens, with a little randomness sprinkled in to make it look like it's coming up with novel token patterns.

So probably the senator Wyoming scenario.

reply

orphanendorphin.bsky.social•85 days ago

yeah, I was hoping someone with access would test and possibly confirm. We owe it to ourselves to start swapping capital I and lower case L in all published writing.

lowa, ldaho, l don't Iike the current tech cIimate.

reply

the-greetest.bsky.social•85 days ago

I like how they don't really say the reason why these models are getting better and better at being able to fool humans via RLHF, which has absolutely no way to strongly re-enforce cultural biases into the probability machine (said with enormous sarcasm).

reply

kaelynski.bsky.social•85 days ago

To be fair, some people do pronounce it "Warshington"

reply

pensandlighters.bsky.social•85 days ago

To be fair, this is the best use case I’ve heard for it so far.

reply

babrahamse.bsky.social•85 days ago

it's actually spelled Marcesterssachusetts we just don't pronounce it that way

reply

gutmichaelbio.me•85 days ago

Beautiful. Amazing. Inspiring stuff. What would we do without AI?

reply

bcmerchant.bsky.social•85 days ago

That will be $200

reply

gardylou-games.bsky.social•85 days ago

Even worse—“it’s free for now while we try to build a user base, but your prompts today did do $200 worth of environmental harm to the environment. Don’t worry, we will charge you some day in the future!”

reply

ericajoy.astrel.la•85 days ago

helpful computers!

reply

cv-danes.com•85 days ago

Not just silent, but invisible!

reply

patrikharnes.bsky.social•85 days ago

Boy, I love the city of Borston, Massachursetts!

reply

birdrespecter.bsky.social•85 days ago

Lmfao

reply

hydrationchimp.bsky.social•85 days ago

Marssachursertts

reply

palecentaur.bsky.social•85 days ago

They just don't pronounce the R.

reply

hydrationchimp.bsky.social•85 days ago

This baby can fit so many silent r’s

reply

eryq.bsky.social•85 days ago

Arkansarrrrs

reply

folksong.co•85 days ago

Pahk the cah in havahd yahd, as my Massachusetts grandma used to say. The R was silent!

reply

folksong.co•85 days ago

Also my grandma taught me how to pronounce her town name Worcester (Woostuh)

reply

karengeier.com•85 days ago

I never pronounce the r in Massachusetts personally

reply

jaftt.bsky.social•85 days ago

That's 4o lmao

reply

ericajoy.astrel.la•85 days ago

helpful human!

reply

begemot.bsky.social•85 days ago

Clearly confused by the old fashioned pronunciation "Warshington"

And uhhh "Connercticut"

And "Massarchusetts"....

reply

kerbeari.bsky.social•85 days ago

Wyroming, Sorth Dakorta and Morntana…

reply

hax.io•85 days ago

Seems legit. We're fucked.

reply

tempus1.bsky.social•82 days ago

AI--fascist billionaires' wet dream, as it's even stupider than they.

reply

scottmullin.bsky.social•85 days ago

The final hurdle before they become self aware

reply

carlygoodman.bsky.social•85 days ago

R U sure

reply

berserkercoach.bsky.social•85 days ago

completely aside from this latest evidence that LLMs are useless bullshit, is it possible that it's "understanding" the question to mean what states have (R)epublicans in them, since (R) and (D) are common shorthands?

reply

rjwhite1.bsky.social•85 days ago

Maybe OpenAI is from Australia

reply

dysamoria.com•85 days ago

So intelligence. Much artificial.

reply

nmmbpodcast.bsky.social•85 days ago

I asked it for examples of certain things and it gave me some. When I looked online I couldn't find any of the examples it had given me. I asked if they were real and it said "No, they are just examples." I pressed it for real examples and it still gave me made up shit.

reply

kyleblakewilson.com•85 days ago

This is such a joke, they are trying to show improvement by just making up their own metrics of success. "It has a checkbox, it's good to go! Hallucinations, checkbox, no big deal 😎"

https://openai.com/index/openai-o1-system-card/

reply

pressxtojasonwrites.com•85 days ago

61% accuracy, it's only several points into F territory!

reply

eloweasy.bsky.social•85 days ago

Picking stawberries in Michigarn

reply

maryhumphreys.bsky.social•85 days ago

Finally, I found my way to Minnesoter

reply

dontinnovate.bsky.social•85 days ago

In fairness, Massachusetts does love hiding their R’s

reply

aredwine3.bsky.social•85 days ago

It appears actually using the reasoning model, 01, does much better. See the chat:

https://chatgpt.com/share/67527c91-0118-800a-b2e3-ce45a65ebaae

reply

kerbeari.bsky.social•85 days ago

South Dakota? A 3rd grader could still do better.

reply

edzitron.com•85 days ago

This was using o1

reply

aredwine3.bsky.social•85 days ago

Yes. 01 is their reasoning model.

reply

edzitron.com•85 days ago

Yes the one that I used in the OP

reply

aredwine3.bsky.social•85 days ago

If you can’t back up what you say, maybe you shouldn’t say it? This is @edzitron.com ‘s response to me asking him to verify what he’s saying.

I’m just going to leave this here:

reply

aredwine3.bsky.social•85 days ago

Hmm, well my output shows the complete opposite of yours when using o1. If we both used o1, I would say a fair conclusion is to always check a model’s output and not blindly trust it.

reply

jnewhouse.bsky.social•83 days ago

Can you share the link next time?

reply

edzitron.com•83 days ago

Thanks for asking! No.

reply

sudhof.me•85 days ago

They cannot do it because these names are just numeric tokens by the time they get to the LLM.

reply

thedimpsler.bsky.social•85 days ago

Did they train it exclusively on samples from Talk Like a Pirate Day?

reply

nathanielimeyer.bsky.social•85 days ago

Strongly suspect this is some wacky interpretation of R politicians representing these states.

reply

realphilgold.bsky.social•85 days ago

try it on apple intelligence

reply

briscoletsgo.bsky.social•85 days ago

Try it on o1

reply

tektology-enjoyer.bsky.social•85 days ago

LLMs are sort of like a person with Fine Motor Skill problems - they can't "pinch" anything. It can be pretty good as a "Yes-and" improv partner but it's never gonna be good at fine-grained questions like this.

reply

tektology-enjoyer.bsky.social•85 days ago

People have a fantasy of the flexibility of Natural Language AND the specificity of Formal Languages but that's fundamentally impossible. Flexibility and ambiguity are two sides of the same coin - too much Variety, too many degrees of freedom.

reply

gloriasin.bsky.social•85 days ago

Maybe it thinks you're asking which state is Republican?

reply

stcosmo.bsky.social•85 days ago

This was my thought. Which is maybe worse, if it's illustrating the tech can't even differentiate between the letter R being in a name or placed next to a name.

reply

gloriasin.bsky.social•85 days ago

I think it depends on how it's been tagged by the humans behind the LLM (probably not ID'd by letter), or the context it was used in the original data (how did that letter R show up when those states were mentioned). AI doesn't automatically understand every question you throw at it.

reply

stcosmo.bsky.social•85 days ago

For sure. Either way I think it highlights the limits of the tech, and makes me feel weirder about it being used for things like insurance claims, etc.

reply

edzitron.com•85 days ago

Incorrect. It reasons based clearly on interpreting the question correctly.

reply

scohoe.bsky.social•85 days ago

so what are you using to get that, when I just do a simple input I got this

reply

flootdraws.bsky.social•85 days ago

Texars, Warshington and Wyorming?

reply

scohoe.bsky.social•85 days ago

reply

gurtthefrog.bsky.social•85 days ago

Texas, wyoming, and washington are all on this list

reply

scohoe.bsky.social•85 days ago

you're right, but not 50, so is every response just random?

reply

oost-audio.bsky.social•85 days ago

LLMs work by predicting the next word based on everything before it. but always choosing the most likely word makes the answers too formulaic, so they added randomness by sometimes choosing the 2nd or 3rd most likely word instead. this can have a knock-on effect that changes the whole answer.

reply

geographer.bsky.social•85 days ago

And that's the very problem with the black box. Even though you had the same question worded differently it's gonna spit out different answers because it doesn't actually "understand" the question.

reply

gurtthefrog.bsky.social•85 days ago

Not an AI expert, but I believe each query is evaluated by itself, so asking the same question multiple times would elicit different answers. the AI runs uses a big statistical model to build each response. It isn’t thinking the same we do — its using math to make a statistical guess of an answer

reply

mcoorlim.com•85 days ago

yes that's how LLMs work based on what text is likely to come next without regard for content or context

reply

mcoorlim.com•85 days ago

like it doesn't get what an 'r' is or how a word might contain one

reply

mfvoss.bsky.social•85 days ago

Thanks for this. I held off reposting it cos' I wanted to double check that it's real. Not that I haven't personally seen "AI" err similarly, I just no longer blindly repost anything after getting burned a few times with misleading posts/memes etc

reply

rjy2.bsky.social•85 days ago

The original post isn't really misleading. Texas, Washington and Wyoming don't have the letter R in them.

My list contains MA, ND, PA, SD, WA, WI and WY for some reason with this prompt:

reply

rjy2.bsky.social•85 days ago

Why?

reply

mfvoss.bsky.social•85 days ago

I don't mean to say it is misleading - I just need to know more about how it was generated before reposting it.

reply

rjy2.bsky.social•85 days ago

Agreed. I'd be curious to see the original prompt/question.

reply

mfvoss.bsky.social•85 days ago

I'd also be curious what you get if you reworded your query to "the letter 'r' ". Will it understand the query better with that distinction and generate a correct answer?

reply

scohoe.bsky.social•85 days ago

I've certainly copy and pasted it when I wanted to fact check but not engage with a troll user, and not happy to see the simple facts it can't get right after 2 years of heavy investment.

reply

richburroughs.dev•85 days ago

What a lot of people don’t understand is that it doesn’t know any facts. It doesn’t do anything remotely like thinking. It’s much closer to being Autocorrect than to being an encyclopedia.

reply

mfvoss.bsky.social•85 days ago

Exactly. Which is why I hate the misuse of the term "AI" and almost always express it as "so-called AI" when posting.

reply

negdiscountrt.bsky.social•85 days ago

Oh man, they just fixed the r’s in strawberry issue.

It’s going to take a while to have all the r-related problems sorted out.

reply

nilbog3000.bsky.social•85 days ago

Some of them have the special secret R that only hallucinating computers can see

reply

kcds.bsky.social•85 days ago

The secret is that it was trained on Chris Cornell’s vocals, where every word has a few extra R’s in it.

reply

gritty7.bsky.social•85 days ago

The letter R does not exist in Massachusetts, silly AI. No, I don’t mean the name. I mean the place. 25 letters in the Massachusettsan alphabet.

reply

persianx6.bsky.social•85 days ago

How much money has been spent on this to be this wrong?

reply

bradsprigg.bsky.social•84 days ago

Are you using the right model? What was your question? Wording matters. This is what I got.

reply

celesticidal.bsky.social•85 days ago

To be fair I've heard both morntana and montanner, depending on the accent 🙃

reply

marmaladebus.bsky.social•85 days ago

This, at least, is easy to spot, but ChatGPT will just make things up. I asked ChatGPT how a classic TV episode ended and whereas a human would admit not having the answer, ChatGPT lied and continued to lie.

reply

bonforage.bsky.social•85 days ago

Because they voted Republican?

reply

Comments

Posting Rules

Comments

Posting Rules

Reply