If an argument only works at the extremes, it probably does not work well. Reality usually lies somewhere in between. Listen only to optimists, and you risk disappointment; listen only to pessimists, and you may never try at all. Pessimists often mistake pessimism for realism.
@pnjoyim@theo I remember my Asian parents telling me, "Don't rely too much on calculator", "Don't rely too much on internet", "Don't rely too much on computer". It's fear of missing control. "there's no X anymore" had never happened though.
5.6 SOL probably makes the best agentic model at the moment. It just hammers through any tasks given and would not give up until it complete its work. Such a reliable model when it comes to prototyping. Absolute time-saver.
Every game world you've ever played in was built by someone before you arrived.
Every room, every character, every possible outcome -- designed in advance, waiting for you to discover it.
We think that's about to change.
PixVerse Game is here.
RT + Follow + Reply = 300 Creds & Early Access in DMs (72H ONLY)
@Oddware26 My point is that GLM 5.2 is on the wrong side of the competition. It’s half‑baked in terms of both value and performance. Where Chinese models currently excel is in price efficiency not SOTA‑level performance.
As a local LLM enthusiast, I find GLM 5.2 exciting. However, I’m not sure it represents the right direction for Chinese models overall. The market is increasingly driven by compute scale, and companies have to burn an insane amount of money just to stay relevant.
@davis7 What's the point of it being next generation when they killed entire budget product line? Does being nextgen mean getting marginal improvements over getting 2x pricing? Not sure if anyone would want nextgen if that's the case.
@thsottiaux I'm slowly realizing pushing out updates like on daily basis adds almost no values. I used to think fast development was always good thing.