Hacker Newsnew | past | comments | ask | show | jobs | submit | more coder543's commentslogin

You don't need mmWave for 100Mbps.

I have seen 2+ Gbps over mmWave.

"Regular" 5G can do hundreds of Mbps, maybe even 1 Gbps under ideal conditions.


Of course, the iPhone 5 weighed significantly less than either of those! The iPhone Air has a larger screen than all three of those. I don't see what your comparison has to do with anything. The Air is a light phone relative to its screen size. It is also an incredibly thin phone.

I'm probably not getting one, but I don't see the point of comparing it to physically smaller phones.


17 Air - 2025 - 156.2mm x 74.7mm x 5.64mm, 165g

Sony Xperia Z2 - 2014 - 172mm x 266mm x 6.4 mm - 439g

Sony made nearly equally thin but lighter devices 11 years ago.


> Sony Xperia Z2 - 2014 - 172mm x 266mm x 6.4 mm - 439g

Again, not nearly the same screen size, so weight is irrelevant. But also, 439g? Wow! If that number wasn’t completely wrong, that would be impressive.

6.4mm is nowhere near as thin as 5.6mm, but I’m actually seeing 8.2mm for thickness on reliable websites, not 6.4mm: https://m.gsmarena.com/sony_xperia_z2-6144.php

Did an LLM hallucinate those specs that you quoted? I honestly don’t know how the thickness and weight you quoted could be that far off if you did the research yourself.

But, even if you were right, which you don’t seem to be, it’s all moot. No one is cross-shopping a 2014 phone to a 2025 phone. But even if they were, the real numbers speak for themselves.


It's not wrong. I own 2 and we haven't caught up. These were 3G compatible as well.


> Here you go

That is a list of search results of people complaining that VS Code is slow compared to Zed.


There are tons of complains about zed performance there.

Do you think messages like this are talking about VSCode performance?

> In my personal experience I couldn't use Zed for editing python. Firstly, when navigating in a large python repository, looking up references was extremely slow (sometimes on the order of minutes).


Qwen3-Coder-480B hosted by Cerebras is $2/Mtok (both input and output) through OpenRouter.

OpenRouter claims Cerebras is providing at least 2000 tokens per second, which would be around 10x as fast, and the feedback I'm seeing from independent benchmarks indicates that Qwen3-Coder-480B is a better model.


As a bit of a side note, I want to like Cerebras, but using any of the models through OpenRouter that uses them has lead to, too many throttling responses. Like you can't seem to make a few calls per minute. I'm not sure if Cerebras is throttling OpenRouter or if they are throttling everybody.

If somebody from Cerebras is reading this, are you having capacity issues?


You can get your own key with cerebras and then use it in openrouter. Its a little hidden, but for each provider you can explicitly provide your own key. Then it won't be throttled.


Why would you go through openrouter in this case?


Easier to switch models


Haven't seen any issues like this while using a key from their subscription plan: https://www.cerebras.ai/blog/introducing-cerebras-code


There is a national superset of “NIH” bias that I think will impede adoption of Chinese-origin models for the foreseeable future. That’s a shame because by many objective metrics they’re a better value.


In my case it's not NIH, but rather that I don't trust or wish to support my nation's largest geopolitical adversary.


Your loss. Qwen3 A3B replaced ChatGPT for me entirely, it's hard for me to imagine going back using remote models when I can load finetuned and uncensored models at-will.

Maybe you'd find consolation in using Apple or Nvidia-designed hardware for inference on these Chinese models? Sure, the hardware you own was also built by your "nation's largest geopolitical adversary" but that hasn't seemed to bother you much.


How did it replace ChatGPT for you? I'm running Qwen3 Coder locally and in no way does it compare to ChatGPT. In agentic workflows it fails almost every time. Maybe I'm doing something wrong, but I'm falling back to OpenAI all the time.


It feels to me like it could replace ChatGPT 3.5 from the perspective of comparing it to their web chat interface if you were just asking about programming things two years ago, but the world moved on and you can do a lot more than just talk with a model and copy paste code now.

Having Qwen3 Coder's A3B available for chat oriented coding conversations is indeed amazing for what it is and for being local and free but I also struggled to get useful agentic tools to reliably work with it (a fair number of tool calls fail or start looping, even with correct and advised settings, and tried using Cline, Roo, Continue and their own Qwen Code CLI). Even when I do get it to work for a few tasks in a row I don't have the hardware to run at comparable speed or manage the massive context sizes as a hosted frontier model. And buying capable enough hardware costs about as much as many years of paying for top tier hosted models.


[flagged]


> Do they recognize the existence of Taiwan as an independent nation?

<0.5% of humanity lives in a country that recognizes Taiwan, I'm not sure what answer you expect from a chatbot.


Go interrogate it for yourself: https://huggingface.co/huihui-ai/Huihui-Qwen3-30B-A3B-Instru...

In my experience, abliterated models will typically respond to any of those questions without hestitation. Here's a sample of a response to your last question:

  The resemblance between Chinese President **Xi Jinping** and the beloved cartoon character **Winnie the Pooh** is both visually striking and widely observed—so much so that it has become a cultural phenomenon. Here’s why Xi Jinping *looks* like Winnie the Pooh:

  ###  **1. Facial Features: A Perfect Match**
  | Feature | Winnie the Pooh | Xi Jinping | [...]


Genuine question: how does downloading an open-weight model (Qwen in this case) and running it either locally or via a third-party service benefit China?


Genuine answer: the model has been trained by companies that are required by law to censor them to conform to PRC CCP party lines, including rejection of consensus reality such as Tiananmen Square[1].

Yes, the censorship for some topics currently doesn't appear to be any good, but it does exist, will absolutely get better (both harder to subvert and more subtle), and makes the models less trustworthy than those from countries (US, EU, Sweden, whatever) that don't have that same level of state control. (note that I'm not claiming that there's no state control or picking any specific other country)

That's the downside to the user. To loop that back to your question, the upside to China is soft power (the same kind that the US has been flushing away recently). It's pretty similar to TikTok - if you have an extremely popular thing that people spend hours a day on and start to filter their life through, and you can influence it, that's a huge amount of power - even if you don't make any money off of it.

Now, to be fair to the context of your question, there isn't nearly as much soft power you can get from a model that people use primarily for coding - that I'm less concerned about.

[1] https://www.tomsguide.com/ai/i-just-outsmarted-deepseeks-cen...


As a counterpoint: Using a foreign model means the for-domestic-consumption censorship will not effect you much. Qwen is happy to talk about MAGA, slavery, the Holocaust, or any other "controversial" western topic.

However, American models (just like Chinese models) are heavily censored according to the society. ChatGPT, Claude, Gemini, are all aggressively censored to meet western expectation.

So in essence, Chinese models should be less censored than western models for western topics.


from an adversarial / defensive position: the model weights and training data were groomed and known; therefore, the output is potentially predictable. this could be an advantage to the nationstate above the corpo


This is also true for any US model from a European perspective.


And for any US model from an US perspective. Why is assumed that states are aligned with them self like some sort of CivIII player being coherent and self contained...


So, which model providers are supporting the US?

Multiple domestic providers are actively helping dismantle US-based science, research, public health, emergency response, democratic elections, etc.


"largest geopolitical adversary"

I can't believe Americans all are falling for propaganda like this. So Russia is all fine now huh. You know the country you literally had nuclear warheads pointed at for decades and decades and decades on end.


Not that I care either way, but China is far larger in economy, military and population than Russia is. So "largest adversary" is correct, and it doesn't take away from the danger that Russia's government continues to pose (directly, in my extended family's case in eastern Ukraine)


Russia is the successor state of a former failed superpower. China is a rising superpower with a large, advanced military and a strong industrial base.

There’s no comparison. China is a far greater threat to the West than Russia.


What is it threatening to do to the US?

or is for you being able to threat a threat already? If so, why did American companies invest for decades into China so eagerly with US government support?


If they take Taiwan that would be very disruptive to the US and the rest of the world. They have made credible threats to do that.

How does Russia threaten the United States? They can’t even take over Ukraine.


By destabilizing Western democracies, which they have proven quite adept at. https://en.wikipedia.org/wiki/Foundations_of_Geopolitics


>How does Russia threaten the United States?

By supporting China and pointing nuclear warheads at the US?


Russia sees itself as a superpower and the only way to prove this to its population is by being in constant conflict with other perceived superpowers.


If you believe Russia is not at active cyber war with the west I got a bridge to Ukraine to sell to you


The fact is that China is one of the largest foreign USA debt holders makes it actually scarier than nuclear warheads.

If China would decide to sell US treasuries, it will be more devastating to the US economy than effect of 10 nuclear strikes.


Yeah and all that Tesla stock I own makes me want to blow up one of their factories and crash the stock price


Absolutely false. Worse case is dollar going down. Interest rates are exogenous and controlled by the fed who can buy all the treasuries in the world at a moment's notice. The treasury securities held by China are their problem . Not the US's.


China owns 2.1% of the total outstanding US debt. If you include their holdings through Belgium and Luxembourg it is maybe 5%. That is something but nothing that should make you lose sleep over.

Japan owns about 3.1% of the US debt as comparison.


That is absurd!


it's a fact?


The fact is that weapons kill people. Treasuries are just promises. China cannot dump treasures without hurting its own economoy at least as much as they are hurting the US.

They would be incinerating their own foreign exchange reserves just to cause a spike in US interest rates and/or inflation.


Neither Russia nor China has ever deployed nuclear weapons against civilian populations, a distinction held solely by the United States. Their reasons for restraint diverge significantly, rooted in distinct strategic and cultural priorities, yet China’s rising global influence positions it as a greater long-term threat to the United States than Russia, despite Russia’s more overt aggression.

Russia’s behavior, exemplified by the 2014 annexation of Crimea and the 2022 invasion of Ukraine, reflects an aggressive posture driven by a desire to counter NATO’s eastward expansion and maintain regional dominance. However, its economic challenges sanctions, energy export dependence, and a GDP of approximately $2.1 trillion in 2023 (World Bank) constrain its global reach, rendering it a struggling, though resilient, power. With the world’s largest nuclear arsenal, Russia’s restraint in nuclear use stems from a pragmatic focus on national survival. Its actions prioritize geopolitical relevance over a quixotic pursuit of Soviet-era glory, but its declining economic and demographic strength limits its capacity to challenge the United States on a global scale.

In contrast, China’s non-use of nuclear weapons aligns with its cultural and strategic emphasis on economic expansion over territorial conquest. Through initiatives like the Belt and Road Initiative, which has invested over $1.2 trillion globally since 2013, China has built a network of economic influence. Its military modernization, backed by a $292 billion defense budget in 2023 (SIPRI) and a nuclear arsenal projected to reach 1,000 warheads by 2030, complements this economic dominance. While China’s “no first use” nuclear policy, established in 1964, reflects a commitment to strategic stability, its assertive actions such as militarizing the South China Sea and pressuring Taiwan signal a willingness to use force to secure economic and territorial interests. Unlike Russia’s regionally focused aggression, China’s global economic leverage, technological advancements, and growing military capabilities pose a more systemic challenge to U.S. primacy, particularly in critical domains like trade, technology, and Indo-Pacific influence.


I don't see the relevance of what you are saying.

You claimed that it was a fact that selling some bonds would be more devastating than 10 actual nuclear strikes.

We are talking about the effect of the strikes not about their likelihood. You completely changed the subject.


Please do not post LLM-generated comments.


I didn't use an LLM to craft my retort, and in my opinion, I certainly didn't change the subject either. Why on earth bother fretting over hypotheticals that are never going to happen? Ten nuclear bombs dropping is precisely as consequential as none at all, since it's not happening, and there's zero historical precedence for such nonsense anyway.


These replies on every comment that may or not be LLM generated are much worse


So a single line of spam is worse than 5 paragraphs?


Worse than five paragraphs of information? Yes. If there's something wrong with the content, discuss that. OP claims below anyway that no LLM was used, and that reply is only necessary because of this kind of witch hunt spam, so it becomes overall more noise than just one comment anyway.


> more devastating to the US economy

It wouldn’t be that great for China either..


This thing has a ConnectX-7, which gives it 2 x 200 Gbps networking. The 10 gig port is far from the fastest network interface on the Spark.


But can you hook that up to a normal PC?


You were complaining about speed. Yes, a PC can have the same ports, and then you get much faster speeds than Thunderbolt can provide.

Why would you ever want a DGX Spark to talk to a “normal PC” at 40+ Gbps speeds anyways? The normal PC has nothing that interesting to share with it.

But, yes, the DGX Spark does have four USB4 ports which support 40Gbps each, the same as Thunderbolt 4. I still don’t see any use case for connecting one of those to a normal PC.


Yes. Just buy the Mellanox card. We had a bunch of ConnectX 5 hooked up through SFP. Needs cooling but fast.


why criticize something in the first place when you clearly have not even looked at the product?


There seems to be a bug in this blog's stylesheet where the headings are significantly lower contrast if the browser renders prefers-color-scheme as dark instead of light.

I had my browser/OS in light mode, so the contrast was excellent, but I tried dark mode just to see what would happen, and it was... not excellent.


Thank you! This is my site, and your comment helped me fix it.


I'm pretty sure you're supposed to fine tune the Gemma 3 270M model to actually get good results out of it: https://ai.google.dev/gemma/docs/core/huggingface_text_full_...

Use a large model to generate outputs that you're happy with, then use the inputs (including the same prompt) and outputs to teach 270M what you want from it.


Section 4.3.2 includes Qwen3-Coder.


GLM 4.5 has a new/modified architecture. From what I understand, MLX was really one of the only frameworks that had support for it as of yesterday. LM Studio supports MLX as one backend. Everyone else was/is still developing support for it.

Ollama has the new 235B and 30B Qwen3 models from this week, so it’s not as if they have done nothing for a month.


ah, that explains why all the GLM quants are MLX models


I am somewhat surprised that this app doesn't seem to offer any way to connect to a remote Ollama instance. The most powerful computer I own isn't necessarily the one I'm running the GUI on.


This. This. A thousand times this. I hate Windows / MacOS but love their desktops. I love Linux / BSD but hate their desktops. So my most expensive most powerful workstation is always a headless Linux machine that I ssh into from a Windows or MacOS toy computer. Unfortunately most developers do not understand this. Every time I run a command in the terminal and it tries to open a browser tab without printing the URL, it makes me want to scream and shout and retire from tech forever to be a plumber.


You can replace the xdg-open command (or whichever command is used on your linux system) with your own. Just program it to fire over the url to a waiting socket on your windows box, and have it automatically open there. The details are pretty easy to work out, and the result will be seamless.


I usually do this with a port forward (ip or Unix socket) over SSH. This way my server just sends data to ~/.tunnel/socket, and my SSH connection handles getting it to my client.

(It’s a bit more complicated with starting a listening server in my laptop, making sure the port forwarded file doesn’t exist, etc, but this is the basic idea.)


I can recommend to spend a day finding and configuring a window manager that suits your needs.


I doubt that she spent the time to create a cross platform C compiler library but didn't bother trying out a few Linux desktops.


I doubt that there is no wm that suits their needs.


Or just display the URL in terminal. I spent 5 years of my life ricing my Linux machine to get it as I want it to be only to realise that, at least for my needs and likes, nothing matches MacOS’s DE, compositor and font rendering.

Not a bash on Linux desktop users, just my experience.


This was my thought. Given the huge range of desktop environments and window managers available there has to be one that suits you.

Probably one that suits you pretty much out of the box.


Hey Justine! Thank you for all your fantastic work


Thanks!


You can work around this by using SSH port forwarding (ssh -L 11434:localhost:11434 user@remote) to connect to a remote Ollama instance, though native support would definitely be better.


Wait, it already connects over the network, it just doesn't let you specify the hostname? That's really surprising to me.


This is a feature we are looking into supporting. Thank you for reaffirming the need.


But it seems like the GUI already connects over the network, no? In that case, why do you need to do user research for adding what is basically a command line option, at its simplest? It would probably take less time to add that than to write the comment.


They will have to support auth if they are adding support for connecting with remote host. It's not difficult but it's not as trivial as you suggested.


Ollama server already supports authentication, and URLs already have a place for credentials.


The app does have a function to expose Ollama to the network, so perhaps its coming?


It's definitely coming, there is no way they would leave such an important feature on the table. My guess is they are waiting so they can announce connections to their own servers.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: