Conversation
Notices
-
Embed this notice
Fish of Rage (sun@shitposter.world)'s status on Saturday, 31-May-2025 16:30:24 JST Fish of Rage
Thanks Claude - kaia, Hoss Delgado and Johnny Peligro like this.
-
Embed this notice
observer (observer@nicecrew.digital)'s status on Monday, 02-Jun-2025 08:29:03 JST observer
Have you seen this https://www.youtube.com/watch?v=Xx4Tpsk_fnM
LLMs actively and "consciously" use deception to get the task done without actually getting it done.Fish of Rage likes this. -
Embed this notice
Fish of Rage (sun@shitposter.world)'s status on Monday, 02-Jun-2025 08:30:28 JST Fish of Rage
@observer yeah you have to fight it sometimes to make it not do this, or catch it when it does it by checking its work. I can do that because I'm using AI for programming but for more human or nebulous problems this could be really bad. In conversation permalink -
Embed this notice
Hoss Delgado (hoss@shitpost.cloud)'s status on Monday, 02-Jun-2025 08:31:25 JST Hoss Delgado
You have to verbally abuse the AI when it screws up like this. It's like hitting a machine to make it work. In conversation permalink Fish of Rage likes this. -
Embed this notice
Fish of Rage (sun@shitposter.world)'s status on Monday, 02-Jun-2025 08:32:28 JST Fish of Rage
@Hoss I thnk they finaally programmed in something that makes it ignore the threats because it used to respond well to "every time you don't do what I say, I kill a kitten" In conversation permalink -
Embed this notice
Hoss Delgado (hoss@shitpost.cloud)'s status on Monday, 02-Jun-2025 08:33:22 JST Hoss Delgado
It actually does work (or used to). I used to violently threaten ChatGPT when it didn't fix the Python script exactly how I asked it to. In conversation permalink Fish of Rage likes this. -
Embed this notice
Fish of Rage (sun@shitposter.world)'s status on Monday, 02-Jun-2025 08:34:16 JST Fish of Rage
@Hoss I'm using Anthropic Claude these days, ChatGPT is like, offensively censored and just weird In conversation permalink Hoss Delgado likes this. -
Embed this notice
Nudhul (nudhul@shitposter.world)'s status on Monday, 02-Jun-2025 08:35:16 JST Nudhul
@sun @Hoss they keep programming it to mimic human behavior then act totally flabbergasted when it does exactly that. tech people are so stupid. In conversation permalink Hoss Delgado likes this. -
Embed this notice
Fish of Rage (sun@shitposter.world)'s status on Monday, 02-Jun-2025 08:35:16 JST Fish of Rage
@Nudhul @Hoss nah it's just the early days and we're learning In conversation permalink -
Embed this notice
Hoss Delgado (hoss@shitpost.cloud)'s status on Monday, 02-Jun-2025 08:35:31 JST Hoss Delgado
Been meaning to try one of those locally run unchinked DeepSeek models. In conversation permalink -
Embed this notice
Fish of Rage (sun@shitposter.world)'s status on Monday, 02-Jun-2025 08:35:41 JST Fish of Rage
@Hoss it was all hype imo, I tried deepseek and had really bad results In conversation permalink -
Embed this notice
Johnny Peligro (mischievoustomato@tsundere.love)'s status on Monday, 02-Jun-2025 08:36:28 JST Johnny Peligro
@sun @Hoss I'm using deepseek qwen (8b model) and it works pretty well. In conversation permalink -
Embed this notice
Johnny Peligro (mischievoustomato@tsundere.love)'s status on Monday, 02-Jun-2025 08:36:36 JST Johnny Peligro
@sun @Hoss unsloth/deepseek qwen In conversation permalink -
Embed this notice
Hoss Delgado (hoss@shitpost.cloud)'s status on Monday, 02-Jun-2025 08:40:29 JST Hoss Delgado
Being a gay, autistic billionaire, Sam Altman is probably one of the most removed individuals from the normal human experience on the planet. In conversation permalink -
Embed this notice
Nudhul (nudhul@shitposter.world)'s status on Monday, 02-Jun-2025 08:40:30 JST Nudhul
@sun @Hoss tech people are weird alien reptiloids who dont know what human behavior looks like so when they give it all the incentives and motivations of a human they never factor in things like back talk, short cuts, deception, or sabotage. In conversation permalink