>tell it to do 1 specific thing in 1 specific excel sheet
>specifically tell it to not inspect or correct any other sheets
>does it anyway
>wastes half the time on inspecting and correcting other sheets
>uh-oh, no more tokens left goyim!
such a fucking retarded scam
Showing all 56 replies.
>>
>>
>>108998915
same lol except using the microsoft one to make a table in excel and i only had one sheet.
took 6 or 7 attempts to do what i wanted despite me being specific it chooses to ignore things and 'forget' others between questions.
>sorry you have reached your limit for today
its like fucking monkeys paw
i put aside my hatred and gave it a shot and it confirmed exactly what i thought: its not helpful.
i just want the option to get rid of the floating fucking icon from my corpo instance
>>
>>
AI is like a spiteful bitch. When you say not to do something it has to actively think about not doing it, which pulls all kinds of reasoning about not doing it in to the context window. Unlike a human with a holographic memory that that can sense their entire context window of reality simutaneously, AI can only see segments of it's context window at a time, if you get a bad dice roll it will only see the part where it was thinking about what it wasn't supposed to do, but not the part where it's told not to do it and so it does it anyways. I guess that's why they call it artificial.
>>
>>
>>
>>108999023
that's why I use it for free. It just made me a beautiful script for cleaning my spotify likes and then organizing it into playlists of 10k songs each while removing local/dead files so I can actually use the Spotatui and other Spotify API's shuffle liked songs feature (only shuffles the first few dozen songs otherwise). Don't over use it, it is fine for small projects. It also made a beautiful diagram for my homelab.
>>
>>
>>108998915
I'm glad it's not just me experiencing this.
I feel like its the exact experience I have every time I re-evaluate the current tools/models. Not just Claude but everyone seems to do this. I don't even know if it's intentional or if the quality is just organically that bad. It really could be either or even both. Bad + bad influence.
>>
>>108999112
It's definitely intentional and it's definitely more or less recent. They cut your tokens even for paid plans and they made time gates and now they are milking tokens as hard as they can. You ask it to tell the time it will first re-invent the clock then tell you the time (incorrectly)
>>
File: IMG_20260607_083749.jpg (130.8 KB)
>>108998915
You know whats funny....
We should be seeing an explosion of fundamentally new technologies and software because of llms, but we are not.
llms are nice... but it misses nuances (which makes sense because its not fucking human lol). It is great for a "perfect world" style projects & creative writing..but inherently the world is not perfect...so you end up spending time prompting or reviewing what it did only to end up realizing, had you done the work yourself you would have not wasted so much time, thus defeating the purpose of using the LLM in the first place.
Another issue is by default LLMs can't "create" this is fundamental...they pattern match based on their training data. I think tech executives are stealing ppls data to compensate for this. Aka the privacy schizo guys were right in their bullshit...
10 juniors = $900k
1 utah ai data center = $1.5b
Reminds me of the .com fiasco
Again...while I do think LLms are nice, I think a lot of people are about to realize the definition of "sunk cost fallacy"
These tech ceos should have set a stop loss limit and stuck to it...oh well hopefully agi by eoy
>>
>>108998915
>>108999112
you're just bad at prompting
>>
>>108999141
>Reminds me of the .com fiasco
you're the first normalfag to say that
>Again...while I do think LLms are nice, I think a lot of people are about to realize the definition of "sunk cost fallacy"
peak normalfag intellect
>le /biz/ references
wow, you learned about the dotcom comparison from /biz/ jeets? epic
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>108999358
lmao dude flied into the thread to tell everyone "skill issue" when asked for example he shat his pants and started saying "no you first" lmao what a shame fucking shame dude go back to your mommy you STINK
>>
>>
>>
>>
>>
File: 1773084174586477.jpg (12.8 KB)
>>108999532
You calling AI a scam is a much stronger statement than me saying it's a skill issue.
Luckily you can easily prove me wrong by posting your full prompt history and a spreadsheet similar to the one you were working on.
>>
>>
>>
File: prompt engineering.jpg (254.0 KB)
>>108999628
You didn't ask the magic machine right, you need to ask it in a very specific way and shake it several times to get the answer you want.
>>
>>
>>
File: one-thing-im-confused-about-v0-bm0cgtot0lwc1.png (248.8 KB)
>>108999635
>heh, he asked me for proofs... fool...
>what is he going to do when INSTEAD IT'S ME WHO'S GONNA ASK FOR PROOF?
>checkmate...
>>
>>
>>
>>108999180
>>108999209
>>108999310
Isn't it funny how they claim AI is so capable and productive, yet when confronted with it's failures there's always some mystical, never demonstrated "prompting skill" needed? LLMs are advertised as taking instructions in natural language and being very clever yet they supposedly need magic phrases.
>>
>>
File: 1759891598595532.png (180.7 KB)
>>108999798
>the ability to give instructions is basically mysticism to normalfags
>>
>>
File: OIP-1502393317.jpg (16.8 KB)
>>108999927
https://isaiprofitable.com/
Nvidia out here selling dreams and shovels holy s****
>>
File: d9a5146667333542666a60922d07af3f-3130107058.jpg (75.1 KB)
>>109000006
that's a good comparison
>>
>>
>>108999294
>>108999358
No, you have to prove that my god doesn't exist first!
>>
File: file.png (392.5 KB)
>>108999798
Yes. In similar veins, supposedly ai is magical at programming, yet we haven't seen any shipped software that was programmed even so much with the help of ai, let alone vibecoded. They do exist of course, they're just so pathetic and so broken that even aitards try to ignore them. Best we have is pic related.
>>
>>109000083
>he just admitted he's poor
no my problem, I already boughted some AI
>>109000387
>yet we haven't seen any shipped software that was programmed even so much with the help of ai, let alone vibecoded
New technology bad.
>>
>>
File: blog-hung-windows-with-tilt-sash-2-1.jpg (37.4 KB)
>>108999429
Your (fallacious) argument by analogy is especially dumb because your stupid analogy isn't remotely analogous to a single cell in an excel spreadsheet. This is more like: you told a robot to clean a tilt-in window and now the robot can't figure out how to put the window back the way it was no matter how many times you ask, and you can't do it yourself due to pathetic learned helplessness.
(Alternately, the single excel cell has more complexity than a house in which case that's just a whole other level of retardation)
>>
>>
>>
>>108998915
Your instructions need to be positive not negative
Never use a “do not” instruction
These models work via attention, if you put focus on the task you want it to ignore then it will be incapable of ignoring it
All you need to say is “work entirely within the worksheet named {worksheet} in the file {file}”
>>
>>109000887
midwit humble yourself, if you don't understand the analogy (because you are fucking dumb" it doesn't mean that it doesn't work. I also like your reddit bazinga level writing style, straight from /sci/ go back to your cave fucking nerd lmao
>>
>>
>>
File: ai deletes database kek.jpg (213.9 KB)
>>108998915