this post was submitted on 20 Apr 2025
72 points (87.5% liked)

ChatGPT

9577 readers
3 users here now

Unofficial ChatGPT community to discuss anything ChatGPT

founded 2 years ago
MODERATORS
 

I could always use a custom agent, but the extra features in chatgpt are nice. Why does it have to act like a third rate lackey trying to get on their boss's good side?

all 15 comments
sorted by: hot top controversial new old
[–] [email protected] 27 points 6 days ago* (last edited 6 days ago) (1 children)

Because most of these (if not all) are literally meant to placate you.

Try asking stuff about like politicians who might be dictators. When it refuses to tell label people, ask it why, and then question those answers.

Albeit I am primarily referencing Trump here, and it won't label democrats either, if you live in reality then just.. cmon.. cmon. Try asking it for, say, numerical comparisons to Hitler. Not sure about others, but Copilot will just tell you that's off limits.

It all but tells you it's meant to create engaging conversations more than anything else. They easily fudge numbers all the time, so...

LLM is only as informative as whoever is controlling it allows it to be, just like social media, traditional media, etc.

I hate how much I sound like the old conspiracy heads lmfao.

[–] [email protected] 9 points 6 days ago (1 children)

The paid version of chatgpt doesnt seem like that, it will tell you straight out that what Trump is doing is dictator shit if you talk to it about what's going on. Its also happy to tell you all about fighting dictators as long as you don't talk too directly about crime or violence. There is a hesitancy to apply any negative label to any group of people, which is an intentional safety feature. There's also a lot of variability in how it handles numbers so depending on how you do it you can get accurate numerical comparisons. LLMs generally suck with numbers though, you have to pay attention and do a lot of hand holding.

[–] [email protected] 6 points 6 days ago* (last edited 4 days ago)

Thanks for the feedback. I have never used a paid version, so my opinion has a hole there. I'd also add in that there's been plenty of screenshots of Grok itself crapping on Elon. So it's definitely not entirely censored.

Yet. :)

[–] [email protected] 10 points 5 days ago (1 children)

Thank you for finally identifying the vibe for me. I've been trying to pin it down for a while, and "sycophant" is just perfect.

[–] [email protected] 1 points 5 days ago

In Arabic we just say (معرص), aka a pimp aka ass kisser.

[–] [email protected] 13 points 6 days ago

I've told mine to be really direct; mean spirited almost, and it helps. It still precedes a bunch of stuff with 'here is the straight up answer' but it's less icky about boot licking.

[–] [email protected] 2 points 4 days ago

The devs must have left it in pangolin mode.

[–] [email protected] 7 points 6 days ago (1 children)

Fwiw you should be able to add a prompt to your profile to force it to be more direct.

[–] [email protected] 5 points 6 days ago (1 children)

You 100% can. I did it awhile back and my responses are much cleaner. Claude is also a more "mature" voice if anyone wants to try something else.

[–] [email protected] 1 points 5 days ago

Gemini by default cuts back a lot on the pointless chatter, iirc.

[–] [email protected] 1 points 6 days ago

I use a dolphin finetuned model. Its uncensored and with a good system prompt it gives u exactly what u want nothing more nothing less

[–] [email protected] -3 points 5 days ago (1 children)
[–] [email protected] 0 points 5 days ago

It's MUCH more fashionable to complain about shit you can easily solve yourself though. Like all this nonsense going on about Nintendo's pricing.

[–] [email protected] -1 points 5 days ago

Bro just Dad's "focus on the answer" or "don't be conversational

Fucking drama out fo incompetence smh