the chatgpt people don't understand they are talking to a box that tells them exactly what they want to hear. the are not embarrassed about doing this. that seems like a problem. bring back shame I guess? - ThreadSky

acollierastro.bsky.social • 1 day ago

the chatgpt people don't understand they are talking to a box that tells them exactly what they want to hear.

the are not embarrassed about doing this.
that seems like a problem.
bring back shame I guess?

Comments

spaceghost71.bsky.social•1 day ago

The new Chatgpt is literally programmed to pump up everything you say to a ridiculous degree.

nickup.bsky.social•19 hours ago

If anyone here cares: OpenAI knows about this issue and is addressing it: https://openai.com/index/expanding-on-sycophancy/

nickup.bsky.social•10 hours ago

Just as an example of the way I use LLMs in work is that I wanted to look up a standard result in a field adjacent to mine that I know only a little about. This is much easier than any alternative.

cautiouspesimist.bsky.social•7 hours ago

So a Google search but with the chance for some extra made up stuff snuck in?

refreshingtime.bsky.social•13 hours ago

AI companies and proponents should have worked out all the issues before gathering an obscene amount of investment capital and anti environmental infrastructure

nickup.bsky.social•13 hours ago

you need the capital to figure out the research since it is expensive - and the environmental impact is no different to youtube or any other web service. Eating meat for one meal produces way more GHG than using an LLM all day.

refreshingtime.bsky.social•13 hours ago

that is a complete dismissal of the actual issue and you know it. youtube and other web services are not what we are talking about. what we are talking about is a supposed service with no functional benefit over a human being in any capacity, that costs everyone more to use

nickup.bsky.social•13 hours ago

Denial of the usefulness of these models is a denial of reality. They are obviously cheaper than humans both in terms of hourly cost and environmental impact. They are far from perfect, but are becoming better *every month*. Why do you think millions of people pay for LLMs?

refreshingtime.bsky.social•13 hours ago

because they are rubes who have been scammed into thinking a wholly unreliable hallucinating machine is more useful than a thinking human being. to even compare the two the way you did is shameful. the environmental impact people have is not going anywhere by taking away their function

refreshingtime.bsky.social•13 hours ago

and the capital invested could be going towards researching actually useful technology, doing real work, helping real people

wes.wordsbywes.ink•14 hours ago

Sycophancy isn't really the issue here; it's fabrication, which is a core feature of LLMs (overagreement with prompts is too, just not usually to the degree of that latest OpenAI update). Nothing in that post addresses that.

nickup.bsky.social•14 hours ago

While fabrication still occurs it is very rare for normal everyday use. If you have not tried a model since the start of the year I really recommend doing so. I essentially use it to search the math corpus, which is otherwise impossible to search. And then I of course can find the papers still.

wes.wordsbywes.ink•12 hours ago

And yet we're in a thread started by an example of fabrication. I see it nearly every day in summaries and in forum posts from people who copy/paste blindly. I see it in bibliographies I edit (esp. in the AI field, somewhat ironically). Fabrication is unavoidable with statistical text generation.

wes.wordsbywes.ink•12 hours ago

To be clear: You've found a way to use LLMs in your workflow, and I'm certainly not here to tell you not to do that.

bryceyoungquist.bsky.social•14 hours ago

the core problem is that their training data is all the text they stole, and all the text they stole features the answer to any question being some form of yes.

it is genuinely very hard to get any LLM to answer a question with "no."

nickup.bsky.social•14 hours ago

It's not that hard. It is always easy to dunk on the dumbest people using a tool for the dumbest thing. But there are plenty of people (including myself) who get useful stuff out of these models. They are excellent tools to use to help with brainstorming.

nickup.bsky.social•14 hours ago

I even tried pushing back a bit more and it didn't relent.

bryceyoungquist.bsky.social•14 hours ago

epicycles have gone up around most of the obvious gotchas. most models also now hard cut over to a preprogrammed set of responses when you ask them to do basic math, because answering "what's 8+7" with "4" was embarrassing.

the reason the epicycles are required, well, that's the annoying bit.

nickup.bsky.social•13 hours ago

We know by research from Anthropic that these models do actually know how to perform basic calculations (and lots of other things) and mechanistically do these procedures *in the matrix multiplications*:
https://transformer-circuits.pub/2025/attribution-graphs/methods.html

nickup.bsky.social•14 hours ago

If you don't trust me at least trust some of the scientists who say that they got some useful correspondence with these models like Terrance Tao and Adam Brown

refreshingtime.bsky.social•14 hours ago

the source is unreliable, and therefore worse than useless, because every result from it must be double checked, which defeats its stated purpose

nickup.bsky.social•13 hours ago

Brainstorming is the process of coming up with hypotheses that are "plausible but not necessarily true" and LLMs are great at that! Plus in a technical field you can usually tell quite quickly if something is a just plain wrong.

refreshingtime.bsky.social•13 hours ago

guess who else is great at coming up with plausible hypotheses, in fact even greater by every metric. i will give you a hint: it has to do with who the LLMs expend an obscene amount lf energy to steal from

bryceyoungquist.bsky.social•13 hours ago

hoo boy, that 'usually'

bryceyoungquist.bsky.social•13 hours ago

it is genuinely a very neat bit of technology. as long as you don't need text to be accurate, or creative, and just match a tone, it's quite good.

this is an incredibly niche use case, though, so you keep getting these losers insisting any second now it's gonna take us all to cyber heaven

refreshingtime.bsky.social•13 hours ago

it is definitively neat, in that it is at best a fun triviality, and not something our futures should all be predicated upon

fallkhar.bsky.social•22 hours ago

I wanna scream, agghhhh. The empty set is anything but chaos!! You can only order it in a single way 😭😭😭😭

xicad.bsky.social•5 hours ago

the empty set is the least chaotic thing i can think of

endolite.bsky.social•15 hours ago

Is this topology I’m so lost

sayumn.grush.org•13 hours ago

∅ has two holes in it

robinadams.bsky.social•9 hours ago

No, it's bollocks.

gienieczko.com•13 hours ago

This is an insult to topology by someone so ignorant they couldn't even come up with the insult themselves and asked autocomplete to do it

endolite.bsky.social•11 hours ago

I tried asking chatgpt about some munkres question bc the stackexchange confused me but it kept trying to gaslight me

Ig its only purpose is to spread misinfo to people that don’t know better who spread it further without understanding

I hope ai doesn’t replace curiosity with idiocy

kendoka.bsky.social•10 hours ago

"Oh I'm sorry, you're right! The previous answer I gave you was completely wrong, the actual answer is *another wrong answer*"

thehamofficial.bsky.social•1 day ago

I hope they enjoy their glue pizza and also eating rocks

shannonhey.bsky.social•8 hours ago

How many people are out there thinking they're John Nash because this thing is spitting math symbols out in response to, let's say, "malformed inputs?" These mathematical statements don't appear to have any internal logic. Looks like the plagiarism machine is spit-balling, again.

bluefoot.bsky.social•13 hours ago

It’s basically a form of navel gazing that would make Narcissist envious. No mirror offers such a shiny reflection.

voxcpw.bsky.social•1 day ago

It looks like someone puked mathematical symbols and expressions into a bad text document. What a bunch of hot garbage nonsense.

bradleyewing.bsky.social•1 day ago

Claiming to be verified by Copilot is hilarious, it uses ChatGPT under the hood by default

chillbrobaggins.bsky.social•1 day ago

Copliot is to AI what my AMC Jeep was to cars -- it's just made of parts of everything you HAVE heard of.

starmadeguy.bsky.social•1 day ago

This whole comment is very stupid but I want to highlight the line "|{}|=|dS|".

The number of elements in the boundry of S is the number of elements in the empty set. So dS has 0 elements. So it is the empty set.

If the symbols meant literally anything to the commenter, they wouldnt write "|{}|"

spacecurves.bsky.social•14 hours ago

We've got studies from the AI makers themselves now saying that heavy genAI use hurts critical thinking skills and has the risk of addiction. If genAI came in a vape pod instead of on a command line interface it'd've been regulated to hell and gone already.

superking1000000.bsky.social•9 hours ago

CITIZEN: SCHEDULE 1 COGNITOHAZARD DETECTED: (CLAUDE SONNET 3.7) ON YOUR PERSONAL TERMINAL. SENTENCE IS 5 YEARS TOUCHING GRASS

mousebitelabs.bsky.social•1 day ago

AI has drastically shifted the peak of the Dunning-Kruger curve even farther to the left

aandegak.bsky.social•1 day ago

In the same way Sisyphus became relevant again with industrialized labor and then again with the gig economy, I think the myth of Narcissus is due a comeback

silkythick.bsky.social•1 day ago

I'm no mathematician but that all looks like completely random gibberish? Does any of it mean anything?

blonicyouth.bsky.social•14 hours ago

it's random gibberish

kiera-the-thief.bsky.social•20 hours ago

It's akin to one of those proofs of 1=0 that buries an algebra mistake halfway through, except this is set theory and is arguably dumber, even before you get into the philosophy.

omelveny.bsky.social•20 hours ago

It's gibberish, using math terms in absolute nonsense ways (and assigning them weird, almost theological meanings?)

meetkhatri.bsky.social•19 hours ago

LLMs are so stupid. I was tinkering with it and asked it tell me chapters in textbooks for a particular topic I wanted to study.
It just made up textbooks that didn't exist.
Also if you ask it to solve a physics problem, it again just makes shit up.

tfox87.bsky.social•1 day ago

I dunno what that gibberish means, I literally have a BS in management, but Marky Mark Z thinks lonely people need to befriend chatbots.
Make it stop.

jamezb.bsky.social•1 day ago

I use ChatGPT a lot for programming. It lies like a MFer. Like you can give it all the context you like and it will ignore it and make shit up. It's still very useful, but you really have to pay attention and fact check it.

benlovesapples.bsky.social•1 day ago

Having tried it a few times, it has never been more useful than a search through stack overflow / documentation. I’m curious, what kinds of cases have you found it useful for?

kiera-the-thief.bsky.social•20 hours ago

Speaking for myself, it's a nice natural language interface for generating regexes. I've also used it to figure out SQL query syntax when doing something unfamiliar.

Can't trust it and need to validate everything it gives you, of course, but it's about as reliable as StackOverflow.

jamezb.bsky.social•23 hours ago

I use it for full on development using the Chat GPT projects. It's a very useful tool if you make good prompts and give it as much context as possible. I also use it for debugging. It's very good at parsing large log files and finding the important bits.

alientech.bsky.social•1 day ago

I use it at work for simple programming and fixing the expats’ crappy translations into English. I just used it to help me create a 200-question technical multiple-choice assessment test for new hires. All I had to do was give it a question idea, and it generated the questions and answers.

acollierastro.bsky.social•1 day ago

what an embarrassing thing to admit!

alientech.bsky.social•1 day ago

It’s a tool. It makes me more productive and helps me separate those who interviewed well but don’t know anything and need remedial training from those I can actually train right away and turn into productive team members. Are you embarrassed when you use a computer to do your calculations?

whatsejstandfor.bsky.social•22 hours ago

There is one tool in this thread

alientech.bsky.social•13 hours ago

It’s pretty obvious your egos are bruised because you thought you were irreplaceable geniuses—and it turns out you’re not. Real human genius will never be replaced by machines,but the masses of half-educated, degree-holding fools churned out by American colleges we’re seeing today definitely will be

mahmoud-khalil.bsky.social•1 day ago

I can see how AI can be intoxicating. They are looking at a mirror that talks back to them, shows what they want to see, tells them what they want to hear. I think it is a problem of incomplete social maturity. The inability to understand that negative social feedback is sometimes vital.

silkythick.bsky.social•1 day ago

People are deeply affected by their emotional responses to their parents. Both excessive negative feedback and a lack thereof would influence individuals to more readily reject any kind of negativity, and seek out perpetual validation even if it's through a digital teddy bear.

radarwithcheeze.bsky.social•1 day ago

We should bring back calling things cringe, but only for AI bros

carados.bsky.social•1 day ago

Oh no there’s more of it in the alt text

omelveny.bsky.social•20 hours ago

unlocking all new levels of crackpot-ery in the field of mathematics

ishkur.bsky.social•1 day ago

They're making the same mistakes theologians make: They start with false or unverifiable premises.

An irreducibly complex system does not imply an intelligent designer. A designer, maybe, but not a very good one. Rube Goldberg machines are deliberate exercises in complex non-intelligent design.

ishkur.bsky.social•1 day ago

We cannot make the distinction of design if we cannot know the method of design, the definition of design, the manner or construction of design, or anything about the designer (up to and including adequate and intimate knowledge of how the designer designs things).

ishkur.bsky.social•1 day ago

Moreover, the recognition of design is heavily predicated on the human standard of what design looks like. But we aren't looking for human-design.

To detect design, we must have an understanding of what non-design is. If a designer created everything, then what are we comparing its designs to?

josiahclark1234.bsky.social•1 day ago

bro is me getting gaslit by chatgpt while doin my abstract algebra homework

medievalkermit.bsky.social•5 hours ago

More people need to realize that ai chatbots don't understand their response. They dont't understand if responses are correct or false or misleading or evil or good or biased. They simply don't understand, at all. They're just Chinese rooms, despite tech bros saying otherwise.

aricake.eu•14 hours ago

I’m a tech bro, and like - this is cringe. I think there’s cool stuff you can do with LLMs, but this ain’t it

actionscripter.bsky.social•1 day ago

Oh look, it's the Play-Doh meal

sydneybaker.bsky.social•14 hours ago

Ah, very good! I see what you did there 👍

quollveth.bsky.social•1 day ago

an equation of the equivalence between order and chaos, the cardinality of the chaos set is equal to the boundary of the order set.

i don't think they know what these words mean, feels like one step away from calling out the Illuminati

matesrsen.bsky.social•1 day ago

LLM chatbots are such a gift from heaven to cranks of all stripes.

frankbruder.bsky.social•12 hours ago

Got to be specific, lest someone brings back shame about fun stuff.

We should bring back embarrassment about stupidity.

We should also bring back shame for crimes.

Also, in the political sphere, we should revive the lost practice of stepping down for shameful or embarrassing things.

david--oconnor.bsky.social•12 hours ago

Great point. I think to drive this, we need its partner: Social pressure and negative reactions/ostracizing. Yet, it's remarkably rare. A politician or corporation lies? People brush it off and/or forget quickly. I don't understand how or why; this is an observation.

darksapiens.bsky.social•1 day ago

The other day I got a comment on my blog by someone offering to send me the texts he got from ChatGPT and Copilot, supposedly demonstrating the existence of hidden variables in quantum mechanics "based on practically trivial (and therefore convincing!) calculations" 😑

gienieczko.com•15 hours ago

"Empty set (chaos)" now that's some quality math crankery.

Also now I want a "this was verified by three independent AIs" version of this meme

xanindigo.bsky.social•12 hours ago

these people are so embarrassing that they make me feel embarrassed to be part of the same species

joeauerbach.bsky.social•1 day ago

I think the bigger news is that we found Terrance Howard's secret account.

carlchansen.bsky.social•1 day ago

This looks like Randy Marsh on South Park trying to explain a math topic he completely made up.

chicken.tube•1 day ago

I put my crank theory into the crank amplification machine and it amplified my crank! I must be on to something!

Comments

Posting Rules

Reply