\
  The most prestigious law school admissions discussion board in the world.
BackRefresh Options Favorite

6/2/25 AI thread

https://x.com/NicoleSHsing/status/1929217404619673873 &qu...
floppy persian stage
  06/02/25
why is she not using thinking models in her comparisons? i w...
marvelous home
  06/02/25
https://x.com/dionysianagent/status/1929337613292327231?s=46...
floppy persian stage
  06/02/25
...
swashbuckling boiling water
  06/02/25
It’s only just beginning.
Snowy flatulent community account
  06/02/25
...
Citrine Chad
  06/02/25
...
Amethyst multi-colored roommate
  06/02/25
Nah that guy probably has zero usable code after all that
floppy persian stage
  06/02/25
it only has a 200K token context window. models doing this s...
marvelous home
  06/02/25
Yup half their current business model is pretty much scammin...
floppy persian stage
  06/02/25
unfortunately the only open "source" (which is rea...
Citrine Chad
  06/02/25
Yeah the Chinese ones are way better and will probably alway...
floppy persian stage
  06/02/25
Gemma 3 from Google is good. The llama models are fraud b...
marvelous home
  06/02/25


Poast new message in this thread



Reply Favorite

Date: June 2nd, 2025 10:59 AM
Author: floppy persian stage

https://x.com/NicoleSHsing/status/1929217404619673873

"I am proud to present MIRROR: a cognitive AI architecture that implements inner monologue & self narrative in LLMs."

LLMs are female I can't say it enough

(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48980531)



Reply Favorite

Date: June 2nd, 2025 3:30 PM
Author: marvelous home

why is she not using thinking models in her comparisons? i would like to see the benefit for models that use thinking tokens.

(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981339)



Reply Favorite

Date: June 2nd, 2025 3:22 PM
Author: floppy persian stage

https://x.com/dionysianagent/status/1929337613292327231?s=46

Lmao

(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981327)



Reply Favorite

Date: June 2nd, 2025 3:24 PM
Author: swashbuckling boiling water



(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981332)



Reply Favorite

Date: June 2nd, 2025 3:36 PM
Author: Snowy flatulent community account

It’s only just beginning.

(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981350)



Reply Favorite

Date: June 2nd, 2025 3:58 PM
Author: Citrine Chad



(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981397)



Reply Favorite

Date: June 2nd, 2025 3:56 PM
Author: Amethyst multi-colored roommate



(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981389)



Reply Favorite

Date: June 2nd, 2025 3:59 PM
Author: floppy persian stage

Nah that guy probably has zero usable code after all that

(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981401)



Reply Favorite

Date: June 2nd, 2025 6:52 PM
Author: marvelous home

it only has a 200K token context window. models doing this sort of thing would saturate it quickly. i'm imagining it making the same mistakes and quickly forgetting them. meanwhile anthropic drains your bank account with API costs.

(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981734)



Reply Favorite

Date: June 2nd, 2025 6:55 PM
Author: floppy persian stage

Yup half their current business model is pretty much scamming retards like this and large institutions with superfluous useless api costs

These guys just need to improve the open source models a little more and we can leave them behind forever

(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981737)



Reply Favorite

Date: June 2nd, 2025 7:02 PM
Author: Citrine Chad

unfortunately the only open "source" (which is really only weights) models worth a damn are out of china and lmao at trusting that shit. meta is the only major US company open sourcing all of its models and zuckerfuck has really dropped the ball

(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981753)



Reply Favorite

Date: June 2nd, 2025 7:09 PM
Author: floppy persian stage

Yeah the Chinese ones are way better and will probably always be the best ones because I assume “meta” has figured out that there’s not actually a way to make money with their current business model for AI

(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981769)



Reply Favorite

Date: June 2nd, 2025 7:16 PM
Author: marvelous home

Gemma 3 from Google is good.

The llama models are fraud bullshit anymore though.

(http://www.autoadmit.com/thread.php?thread_id=5732488&forum_id=2...id.#48981788)