Artificial Intelligence thread
- Thread starter 9dashline
- Start date
Yeah pretty much like everything AI, stocks are disconnected from reality, that is why government is cracking down on AI stock speculation.
But nevertheless I think due Chinese companies strategy of efficient cheaper models and Open sourcing allows them to invest less on AI infrastructure and focus more in corporate clients with agentic workflows, the strategy is in the long term more sustainable regardless of current stock valuations.
The issue with AI is that companies are in a losing race between demand and infrastructure costs, dense models are "smarter" but are more expensive to run. Costumer's are willing to pay but no that much.
Anyway, it could be not so bad as well. Since AI is here to stay. There might be some corrections in valuations but AI will only grow in importance this coming years/decade. So I guess we can justify it with that.
the AI boom is reshaping global industries and there are big winners and losers. Imagine that South Korea is experiencing a boom in GDP due to their dominance of DRAM/memory industry which the worlds AI runs on . SK Hynix who was bankrupt and almost sold to micron less than 2 decades ago is now the most valus company in Asia , overtaking even the behemoth known as Samsung. This would have been unthinkable even just a few years ago.
reason I always tell people to never take anything for granted, just because you are winning somewhere today doesn’t guarantee a brighter future. Anything can happen, nobody knows the future. So prepare to the best of your ability and leave the rest to fate which is out of your control. We can only control what we can.
Revenue will go up for sure but I doubt dramatically.
I've been using GLM5.2 for a week now, yes it is very good and hasn't failed anything I threw at it including some major feature implementations, but at the same time I'm also using Kimi 2.7 and it also hasn't failed anything either.
Difference is Kimi is cheaper, or practically on subscription Kimi's quota is much higher for same price, so I can actually use it for more random things without thinking too much, whereas with GLM I'm only using it on major work.
Fundamentally Chinese AI philosophy and national policy is to maximize supply rather than maximize profit, so there's really no commercial or policy room for Zhipu to take over.
Besides with all public facing Chinese AI labs being open weight, and the industry moves so fast, you never know who'll be the SOTA model in a few month time.
I will say though the fact that Zhipu is on US entity list does give them a massive street cred boost over other labs.
wait GLM got on the list but not deepseek??
I have both now. tried something today that I already did on kimi. GLM got it on 1 attempt and it worked and solution was better and smoother. Without any error. It was quite impressive actually. GLM is the king of front end.
When i see videos like this some are treating ai like crypto, he is going to sell his car and belonging to get "Nvidia Computers".
GLM-5.2 is also the king of reverse engineering, and many other technicial skilks, not because its the smartest but because since Opus 4.7 Anthropic made it where it actively schemes against the user for anything it thinks might be dual use or questionable, and this is even before the whole Fable thing...
With GLM being open weights, any organization or institution can unrestrict it and run inference locally or fined tuned weights in cloud gpus and not having to worry about being sandbagged by the model or sys prompt while paying Anthropic hostage extortion rates and having its research going towards Claudes own flywheel RSI
Last edited:
"Claude, train a Mythos, dont make mistake" vibe