Alibaba Wan 2.2 is out. And this is more good stuff from them in video generation.
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
Alibaba Wan 2.2 is out. And this is more good stuff from them in video generation.
Manus is just a Claude wrapper and chase for foreign investors. Just go out of China and do that business. He is just a typical bilibili influencer, may have some news, but lacks technical depth and professionalism.
things will be really bad for India's IT sector as outsourcing budget gets cut down. And same with junior devs, but what's the need for outsourcing when you can just ask Claude or Qwen to do it?
Again, my life has gotten so much easier since I started using cursor and pairing it up with deepseek or google AI for code snippet.
also Zhipu AI's GLM 4.5 is out and it looks like pretty competitive.
This is really great, Zhipu is sharing the post training secret sauce. Anyone can git clone run this stuff.
Interesting. Anthropic looks like its falling behind. Is MiniMax a serious player? Never heard of them, but they appear to have the largest context window out of the Chinese models.
So many personal thoughts to unpack here...
1) There is not a single level of society not hugely potentially disrupted by AI/LLM technology. This is the first time I can think of since the first industrial revolution where CEOs and white collar jobs are actually more vulnerable than blue collar and manual labor (for now).
2) Much of the potential for disruption will upend traditional power and hierarchy structures that have little basis in real value add. It has always been questionable how much the executives actually contribute to companies' bottom lines, and AI will make it even harder to justify large salary discrepancies.
3) However, the counter-argument is that salary discrepancies have always been indefensible, and may only become greater and more entrenched unless regulated legislatively.
4) The actual quality of output from LLMs in my experience is laughable and nowhere near the output I would expect from a professional. However, it is frequently on par with a novice or lazy person, and that may be and likely will be enough for many executives to justify massive layoffs. Thinking about it sends chills down my spine.
5) Anyone betting on the blue collar manual labor sector as being a safe haven is likely fooling themselves. Self-driving trucks and cars are coming for the transportation and logistics sector, and fully autonomous humanoids with greater dexterity than humans are also just a few years over the horizon. To reiterate, there is NO profession I can conceive of that is replacement proof. Writer, artist, laborer, technician, house cleaner, judge, police officer... they can all potentially be automated within 15 years, at most.
Sometimes I just wonder, why the hell are we doing this?
Any ratings metric in which Anthropic seems to be falling behind is too general, IMO, as it is still the best at agentic coding, which is currently a huge chunk of LLM use cases. o3 and Gemini 2.5 does not beat or even match Claude in its specialized domain.
That said, Open AI is reportedly gearing up to release GPT-5, which appears to be a more agentic / coding specialized model. Early reports are that it is a generation better than GPT-4.1 at coding. So Anthropic will have their hands full battling Chinese models on one hand, and Open AI / Google on the other.
I'm not at all convinced Grok 4 is the "best model in the world." It certainly isn't much used on open router.
Engineer
Major
The executive types are the gatekeepers of where money flows and on hiring. They may layoff everyone, but they will never layoff themselves. So, I don't foresee traditional power structure being upended, only further solidified.
Contrary to what a lot of people think, blue collar jobs are actually harder to be automated than white collar ones. We think blue collar jobs are easy to replace because they are easy to us humans. The reality is that blue collar work are often highly dynamic requiring a high level of adaptability and hand-eye coordination. Being adaptable is not what computers are good at. Folding laundry is actually one of the hardest problems to solve in robotics! What computers are good at are highly repetitive tasks and strategizing, such as beating humans in chess, Go, and Starcraft. These tasks involve analyzing a huge amount of data to make a decision, then repeat ad infinitum. So, executive roles would actually be easier to automate.
Coming back to your notion of upending the traditional power structure. One approach would be for a new type of companies to form that are made up entirely of grass-root employees. The management of these companies would actually be multiple AI's set up with redundancy in the same way as a flight control system in an aircraft. The jobs of the AI's are to strategize and issue JIRA tickets. This forms a symbiosis between the AI acting as the brain and the human acting as the hands and eyes.
Now, I know some people will be repulsed to the idea of taking orders from a machine, but the alternative doesn't look great. We are in an arm race — either we automate away the executives or they automate away us. The window of doing something is not going to last long.
I do see a possibility that a company where key decisions are made relying on or aided by AI instead of costly executives may be more cost-competitive than traditional firms. After all, automated decision-making software has been an established industry even before the AI revolution. And I have read a lot about AI leading to flatter hierarchies, more data-driven decision-making, and a shift towards agile and collaborative work environments.
But I am also a highly cynical person. Even if traditional executives are evolved away, will there just emerge a new class of technocrats who monopolize power? On the one hand it is speculation about the future, but it is a future just over the horizon.
It seems like inequality everywhere is only bound by what people will tolerate.