[ om / jp / 2d ] [ home ]

/jp/ - 2D/Random

Name
Email
Subject
Comment
File
Password (For file deletion.)

File: 1740188740595.png (2.48 MB, 1373x2121, 127172446_p0.png)

No. 643043

この日は道を歩けど歩けどおじさんに話しかけられる日でしたჱ՞ ᴗ ᴗ ՞꒱「黙れ」と言ってしまってすみません、イライラしてしまいました

No.643045

File: 1740189056542.png (92.58 KB, 836x734, 2025-02-21-204900_836x734_….png)


No.643046

File: 1740189975791.jpg (237.66 KB, 1080x1594, Screenshot_20250221_200449….jpg)

>>643045
>gook moon runes
Some of the other AIs like Gemini would of called you out for being racist

No.643047

File: 1740190192014.png (87.19 KB, 348x525, 1730407275245.png)

>>643046
I think ddg has their GPT-4o mini on the lowest possible moderation settings.

It's still highly restrictive and impossible to jailbreak, but whatever knobs do exist are turned all the way down.

No.643048

>>643047
Haha you're right so funny *starts rubbing your your back*

No.643049

File: 1740192266454.jpg (928.06 KB, 2894x4093, GI9E6EkbwAAYJzJ.jpg)

>>643046
the longer the context gets, the more likely it is to start ignoring smaller parts that have less of a pull on the conversation, I just tried it, and google's base model (flash) just responded to my japanese input in japanese and their newer reasoning model (basically the newer technique of having the llm break input down, feed back into and talk with itself)

LLMs are still pretty rudimentary to their core idea of tokenize input words, and then vectorize them and poop out where they see the conversation as going. (This is easier to see with stuff like gpt 1 and 2 where it clearly was going word by word). Even with lots of lobotimization they still really struggle with censoring themselves or admonishing users, especially with a lot to work with. A lot of the times those checks are either right at the start or at the very end as they're all still in very active development.

No.643050

>>643047
Haha you're right so funny *starts rubbing your your pussy*

No.643080

File: 1740233434801.png (684.9 KB, 578x600, 1740225477595.png)

>>643045
if you're so smart with your technology and language models tell me what it's saying then huh? huh? that's right

No.643083

my buddy loves pussy

No.643085

he loved feet

No.643141

File: 1740310846662.jpg (298.04 KB, 2560x1440, 2025-02-23-063006_2560x144….jpg)

Installing default Debian 12 stable (GNOME) in QEMU with full modern QEMU paravirt options is the most excitement I've had with technology in many years.

I haven't seen a stable Debian release since GNOME 2 days, and this feels and looks so good.

No.643142

spam

No.643143

wish devuan gnu+linux supported ventoy properly

No.643144

>>643141
i like gnome too
are you going to stop using your meme minimalist wm only setup now

No.643145

File: 1740314677042.jpg (357.05 KB, 939x852, latency frames.jpg)

>>643144
No I wouldn't use a DE on my host, nor would I use Debian on my host.

It's fine for my VM use where I install them and turn them offline forever, so I want some nice fully packaged loadout.

No.643146

Why? Why all this?

No.643147

picom is the lowest latency compositor now



Delete Post [ ]
[Return]
[ om / jp / 2d ] [ home ]