Artificial Intelligence thread

I

iewgnem

Captain
Registered Member
Tried GLM5.2, not bad so far.
tphuang said:
https://twitter.com/i/web/status/2066940737011560652

https://twitter.com/i/web/status/2066938937344495629

https://twitter.com/i/web/status/2066957802741043641

https://twitter.com/i/web/status/2066947839591084212

https://twitter.com/i/web/status/2066951439793242193

GLM-5.2 looks like a really good model. I mean I've already picked Kimi for its news research stuff, but it looks like in terms of engineering, GLM-5.2 is the best Chinese model ever in terms of closeness to frontier US models.
Click to expand...
Tried GLM5.2, I'm liking it so far, doesn't spam and just works, hasn't tried it on anything hard yet, will see how long my $50 last.

Honestly I'm a little conflicted right now, Chinese models are all getting so good I'm being torn between my dopemine reward cycle conditioning for Kimi and FOMO for not using GLM or whatever the latest model is.

Kinda make me appreciate how much easier life is for an American who only has Claude: the simple life of a poor man with no desire and no temptation.
 
bsdnf

bsdnf

Senior Member
Registered Member
iewgnem said:
They're saying DS4.1 is coming soon with much better coding
https://twitter.com/i/web/status/2066940167211102662
Click to expand...
No, Xiuyu Li previously stated that Deepseek V4.1 is undergoing a complete retraining process and its release won't be soon. He meant to oppose these rumors.

A few days ago, the V4 web version rolled out a checkpoint update. While it featured a different CoT style and faster generation, the quality suffered, which clearly indicated that things weren’t quite right yet, so it has now been rolled back. Keep in mind that V4 also went through multiple checkpoint iterations over a period of time before its official release.
 
9dashline

9dashline

Major
Registered Member
iewgnem said:
Tried GLM5.2, not bad so far.

Tried GLM5.2, I'm liking it so far, doesn't spam and just works, hasn't tried it on anything hard yet, will see how long my $50 last.

Honestly I'm a little conflicted right now, Chinese models are all getting so good I'm being torn between my dopemine reward cycle conditioning for Kimi and FOMO for not using GLM or whatever the latest model is.

Kinda make me appreciate how much easier life is for an American who only has Claude: the simple life of a poor man with no desire and no temptation.
Click to expand...
just pay $200/mo for claude

right now on 20x plan you can burn about 600 million tokens a month if you use up quota... its cheaper than GLM api pricing
 
Wrought

Wrought

Captain
Registered Member
Apparently sanctions are on hold for Deepseek, CXMT, and a hundred more companies.

June 16 (Reuters) - The U.S. has held off adding China’s AI startup DeepSeek, memory chipmaker CXMT and more than 100 other companies flagged as national security risks to a trade blacklist, according to two people familiar ‌with the matter, as the Trump administration tries to avoid escalating tensions with Beijing.

At least 75 Chinese entities in advanced semiconductor production, semiconductor manufacturing equipment production and AI modeling have gone through the committee and were slated for blacklisting, one of the sources said.

Since late 2025, Jeffrey Kessler, under secretary of commerce for industry and security, has sought to avoid listing Chinese parties for fear of escalating tensions between the U.S. and China, according to the first source and other people familiar with the matter. The dearth of listings offers ⁠a window into what many see as a larger problem at the Bureau of Industry and Security under the second Trump administration — an
Please, Log in or Register to view URLs content!
Please, Log in or Register to view URLs content!
to combat threats that can be reduced by restricting exports. Early last year, for instance, the bureau said it would replace a regulation created under former President Joe Biden to govern global access to U.S.-origin AI chips. But it has still not published a replacement, and is not enforcing ⁠the earlier rule,
Please, Log in or Register to view URLs content!
that may have allowed the chips to be exported to Chinese companies outside China.
Click to expand...

Please, Log in or Register to view URLs content!
 
bsdnf

bsdnf

Senior Member
Registered Member
iewgnem said:
Actually this reminds me, SpaceX IPO only raised $75B, so acquiring Cursor for $60B in all stock transaction means Cursor just received as much stock as almost all SpaceX IPO investor combined by making a Kimi 2.5 wrapper, lol
Click to expand...
However, this is essential for XAI.

Cursor 2 trained on the Kimi k2.5 offers significantly improved programming capabilities, meaning they have ample private data and post-training capabilities. While not perfect for XAI, this is still way superior to XAI's current state, which offers little beyond computing resources.
 
Last edited:
You must log in or register to reply here.
Top