Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GLM 5.2 and the coming AI margin collapse

martinalderson.com

Post reply on HN
Print view View original

451–460 of 495 posts

Re: GLM 5.2 and the coming AI margin collapse

#451

Post by andrekandre »

ChrisLTD wrote: post #445
fodkodrasz wrote: post #420

Earlier quoted context omitted.

Why? Worst se we are 10 years behind them. Second mover advantage will also be there, it this is so critical indeed. What happened to China because they were third movers in the race to the nuclear weapons? Nothing. They were pretty behind for a while technologically. The wheels have turned. Stop looking at the world as a short strategy game match, or like a Hollywood movie where everything is all-or-nothing. Many pa…

If anything the first movers would theoretically be the first to destroy their own labor markets and economies.

yep, first-mover disadvantage so-to-speak

Re: GLM 5.2 and the coming AI margin collapse

#452

Post by tough »

manquer wrote: post #79
tough wrote: post #21

Earlier quoted context omitted.

OAI has announced an upcoming 750tok/s 5.6 served through their cerebras acquisition

> cerebras acquisition Partnership you mean?, Cerebras went public and are trading at around 45B in market cap. While OAI could in theory cough up that kind of money, it would massively hamper their existing committed capital outlays.

Yes sorry, i got confused some how and mixed up the partnership announcement [1] with an acq one,

maybe i should get some cerebras stock then, ty for the pointer

1. https://openai.com/index/cerebras-partnership/

Re: GLM 5.2 and the coming AI margin collapse

#453

Post by msephton »

a_c wrote: post #406
msephton wrote: post #305

Earlier quoted context omitted.

FWIW I've seen subagents remain open for followups on latest Claude.

Will give it a try thanks for the heads up. Mind if I ask if you are referring to SendMessage? I was testing on Claude code 2.1.196. SendMessage was not available. Skimming through their change log didn’t seem to have anything related to SendMessage. There is “ Fixed SendMessage silently misrouting when a re-spawned agent reuses a previous agent’s name — the tool now detects the mismatch and asks the caller to retarg…

I have no idea what things are called but somebody showed me how they could start a bunch of subagents and steer each with additional messages; some closed after their task was completed, but one stayed open for followups and continued working after being prompted by the main agent who was acting as triage.

Re: GLM 5.2 and the coming AI margin collapse

#454

Post by jononor »

0xbadcafebee wrote: post #413
davedx wrote: post #355

Meanwhile: > China’s Ministry of Commerce has led meetings over the past month with major AI companies, including Alibaba, ByteDance, and http://z.ai/ , to discuss measures that would restrict overseas access to cutting-edge AI models, including models that have not yet been released. > The discussions reportedly include not only closed-source models but also open-weight models. > Future regulations could take the fo…

Makes sense. AI is a munition now, and nations tend to restrict them. Eventually there will be international open source communities building open weights (all you need's a few smart people and a lot of GPUs for training), making the restrictions moot. Until then this will become the norm.

You also need a shitton of data. People are saying that synthetic data is widely used, but pretty sure that is on top of the organic data. Both for pretraining and posttraining. That said, I do think we will see more open and collaborative approaches over time.

Re: GLM 5.2 and the coming AI margin collapse

#455

Post by Paradigma11 »

logicchains wrote: post #276
cezart wrote: post #253

Earlier quoted context omitted.

Not OP but, EU economy is being squeezed by China on the industrial and tech front, and by the US on the innovation/startup front. It is clear EU is no longer at the technological/economic frontier like it used to be 10-20 years ago. At the same time there are serious demographic, budgetary and political challenges all across the continent. Dragi's report covers some of these. It feels like the whole system might fal…

More likely it just slowly declines like Japan. Or if anti-migrant sentiment continues growing at the current rate, it breaks into a race war when the AfD and PFN win the majority of votes in Germany and France respectively.

I dont think these slow indicators and developments hold much water the next few decades. We have seen the first inklings what the brave new multipolar world is going to look like and it does not bode well for much of the world. As long as Russia breaks down before the shit really hits the fan, Europe should be fine(r) than most.

Re: GLM 5.2 and the coming AI margin collapse

#456

Post by jononor »

sajithdilshan wrote: post #383
myrmidon wrote: post #378

Earlier quoted context omitted.

Again, you are assuming that frontier models will stay meaningfully ahead long-term. Export controls/bans are pointless if this is not the case. There is ton of strong indicators that they will not stay ahead: Assuming that technological progress of any kind follows some form of logistic function (where "gains", in this case "intelligence" become sub-linear at some point) is (long-term) a very conservative and proven…

It is not about intellectual advantage, but about capabilities as well. Assuming that Europe would get all the knowledge they need on the cutting edge technology on how to train a frontier model (which they won't because US and China would guard this as national secrets), who is going to setup infrastructure to train the model? Who is building the data centers (multi-year project) and who is going to build the energy…

EU already has supercomputers and data centers. Some examples: https://www.eurohpc-ju.europa.eu/supercomputers/our-supercom...

The goal is to triple the number of data centers over the next years. It is not like there is nothing going on...

Re: GLM 5.2 and the coming AI margin collapse

#457

Post by diseasedyak »

_def wrote: post #423
pixlmint wrote: post #199

Last month, I cancelled my Claude Pro subscription and instead used those 20$ to purchase Openrouter Credits. Most of my knowledge-seeking questions can be answered by Gemma4, for basic code editing, Qwen3.6 27b is enough, and for really difficult tasks, GLM5.2 doesn't leave me hanging. I'm by no means a heavy AI user, so I'm even saving money going the API Credit route and relying on the smallest possible model depe…

I literally burned through 20USD in a couple of hours on openrouter with deepseek v4 pro and opencode tasks - i'm sure i did something wrong

Not saying this is what you did, but I found out the hard way that keeping a context window open for a long time burns tokens like crazy. I had a Hermes session using DeepSeek v4 Pro via Ollama, and ate half my weekly token allotment in a couple hours, just by piling on more and more work inside the same prompt, instead of exiting out and starting a new session (or using the Hermes command to start a new session without exiting).

Re: GLM 5.2 and the coming AI margin collapse

#458

Post by potsandpans »

WarmWash wrote: post #421
Lalabadie wrote: post #416

Earlier quoted context omitted.

You're applying US cultural logic to Chinese bureaucracy

China is an authoritarian state with a single leader who has unilateral uncontested control for life. China has no real bureaucracy (or any other structure for that matter) because at the end of the day, it's one guy who can do whatever he wants whenever he wants. For commoners and officials there is this faux bureaucracy, but for the elite at the top making decisions, there is zero. If Xi doesn't want models exporte…

> If Xi doesn't want models exported, he's not having a legal delegation go to the supreme court of China to fight for his ruling.

"Sorry you can only ship fable to American citizens"

Re: GLM 5.2 and the coming AI margin collapse

#459

Post by andriy_koval »

sajithdilshan wrote: post #424
nightski wrote: post #402

Earlier quoted context omitted.

Meta's FAIR has several R&D offices in the EU, yes. So you are saying their labs can conduct R&D on models in the EU, potentially even train them there, they just can't have production LLM inference serving or release the model weights? I'm just not seeing it. A not-insignificant portion of the AI/ML research community is in the EU.

What you’re failing to understand is that meta is ultimately a US company. If meta unveils a powerful frontier model tomorrow and US impose a ban of its exports, regardless of how many research, data centres, meta has in EU, they will restrict access to the model for EU citizens, same way Anthropic did. Regarding the open weights, I don’t see meta doing that for their future models, especially once they have their ow…

meta researchers from eu will create new company and build model using knowledge they have.

Re: GLM 5.2 and the coming AI margin collapse

#460

Post by senordevnyc »

devinodowd wrote: post #273

I wonder if anyone has actually measured the difference in verification time between these models. A senior dev in a high cost of living area costs the company something like $200 an hour. If a cheaper model produces code that takes an extra 20 minutes to debug or verify because it missed a subtle edge case, you have already lost any savings from the lower API bill. It feels like the real moat for labs like Anthropic…

This is exactly how I see it, and why I’m often willing to pay Fable rates. Yes, I might spend $300 on a feature instead of $20, but it’ll be done in a tenth of my time, thanks to faster iteration in both the planning and post-impl verification / iteration stage. Plus a mistake could easily cause a bug that costs me a customer worth four or five figures of LTV, and the feature might easily add tens of thousands in marginal revenue over the coming years. So really $300 is nothing, it’s barely worth worrying about.
Post reply on HN