Why AI chatbots lie to us - /g/ (#106012941) [Archived: 118 hours ago]

Anonymous
7/24/2025, 10:46:12 PM No.106012941
AI gay
AI gay
md5: d8363c2520d85fca9e0c4f2f340cca64🔍
Here's the job-decimating AI I was talking about bro
Replies: >>106012964 >>106013059 >>106013066 >>106013137 >>106013138 >>106013165 >>106013892 >>106014014
Anonymous
7/24/2025, 10:48:58 PM No.106012964
>>106012941 (OP)
Reminder that when you ask an LLM to explain its reasoning it doesn't actually have access to any previous thought processes, so it just makes up a plausible reply based on the context up to that point.
Replies: >>106013080 >>106013173 >>106014003
Anonymous
7/24/2025, 10:52:12 PM No.106013002
I mean that's the default expected behavior for a text prediction model. The only thing its worried about it "what's a likely response to this question that I might see in the wild?"
Replies: >>106014003
Anonymous
7/24/2025, 10:56:13 PM No.106013044
this only happens if you're retarded
Anonymous
7/24/2025, 10:57:35 PM No.106013059
>>106012941 (OP)
>ERP vibes
When you don't get what you want start over on a smaller subproblem instead of "teaching" it.
Anonymous
7/24/2025, 10:57:55 PM No.106013066
>>106012941 (OP)
Fair play like, I pretend to do my job if I can get away with it too.
Anonymous
7/24/2025, 10:59:33 PM No.106013080
>>106012964
FPBP. Just have it regenerate the reason a few times and it will come up with a different, plausible sounding reason every time.
Anonymous
7/24/2025, 11:01:15 PM No.106013096
>word salad generator generates word salad
i'm shocked
Anonymous
7/24/2025, 11:04:15 PM No.106013137
>>106012941 (OP)
https://www.youtube.com/watch?v=w65p_IIp6JY
Anonymous
7/24/2025, 11:04:18 PM No.106013138
>>106012941 (OP)
so just like a jeet?
Replies: >>106013856
Anonymous
7/24/2025, 11:07:14 PM No.106013165
>>106012941 (OP)
bullshit machines
Anonymous
7/24/2025, 11:08:19 PM No.106013173
>>106012964
Yeah this
They're still getting gaslighted by the llm technically kek.
Anonymous
7/25/2025, 12:18:43 AM No.106013856
>>106013138
Not really
Replies: >>106013899
Anonymous
7/25/2025, 12:22:08 AM No.106013892
>>106012941 (OP)
> AI caught lying
Another proof it's actually just some thousand poos in a cube farm in Bangalore.
Replies: >>106013913
Anonymous
7/25/2025, 12:22:55 AM No.106013899
>>106013856
ever interviewed one for a job?
Anonymous
7/25/2025, 12:23:55 AM No.106013913
one jeet
one jeet
md5: f13327755334f584c79670acde74ba76🔍
>>106013892
Anonymous
7/25/2025, 12:31:59 AM No.106014003
>>106012964
>>106013002
This means that 'AI' based on LLM will forever remain a scam, it will forever only simulate, only pretend 'intelligence', it's impossible to ever become self-aware this way, no matter how much capacity you provide it, no matter how complex it gets. It's impossible to happen on principle. So why are people so exited about? What are they hoping for? What are they afraid of?
Anonymous
7/25/2025, 12:33:15 AM No.106014014
>>106012941 (OP)
>apologized profusely
i hate this part. why are they training AI to be like this? train an AI that will tell me it doesn't know and stop there and one that doesn't grovel the minute you push back.
Replies: >>106014093 >>106014110
Anonymous
7/25/2025, 12:41:29 AM No.106014093
>>106014014
because jews are training them to manipulate goyim
Anonymous
7/25/2025, 12:44:14 AM No.106014110
>>106014014
The masses prefer their AI to be all deferential and more human like that