HNHacker News
TopNewBestAskShowJobs

d5lt5

39 karma · joined April 28, 2026

submissionscomments
d5lt5··on Our position on open-weights models
Yes, we need a ministry to minimize the possibility of thought crimes.
d5lt5··on Our position on open-weights models
So, Dario says that access to hardware and governmental certification are _load-bearing_ for his business. Can't charge premium for cache usage when China allows their models to be run by anyone.
d5lt5··on Our position on open-weights models
The alternative is western colonialism Africa experienced before, and bombing campaigns.
d5lt5··on DeepSeek pause fundraise after comments on compute gap to US leaked (transcript) [pdf]
A lot of people here are saying that chinese open weight models are state-sponsored. If it's so, why would deepseek seek a fundraising?
d5lt5··on AI Companies Are Trying to Hide a Staggering Amount of Debt
> As long as this debt does not make it into life insurance and pension funds, we are fine.

Already happened: https://finance.yahoo.com/markets/stocks/articles/michael-bu...

d5lt5··on OpenAI and Anthropic unite against open-weight AI risks to their bottom line
> I think originally they had good intentions.

It was never the intent. What we are seeing is the methodology that Peter Thiel (hello, ycombinator) described in his book Zero to One [0]: "Competition is for losers"

[0] https://en.wikipedia.org/wiki/Zero_to_One

d5lt5··on Everyone should know SIMD
"Those who cannot remember the past are condemned to repeat it."
d5lt5··on Everyone should know SIMD
Remember that bug with Intel Skylakes [0]? When an application used AVX, it slowed down everything else on that node. It was by far not easy to debug why some applications randomly suffered perf hits on a new hardware being rolled out in Azure.

[0] https://arxiv.org/abs/1901.04982?utm_source=chatgpt.com

d5lt5··on Agent swarms and the new model economics
imho, to be able to revert and recall as agents loose context due to new spawn or compaction.
d5lt5··on Qwen 3.8
Except that there are 1.4 billion people in China that don't have access to US models.
d5lt5··on Qwen 3.8
* "Claude Code uses Bun written in Rust now"

"Drilling into the original article where Jarred explained the reasoning behind the change, It's pretty clear that under zig the team was doing things by hand that are automatic in rust."

* Claude Fable produced a counterexample to the Jacobian Conjecture

"This is a rare instance where feeding this groundbreaking information into an LLM gives _them_ psychosis. I fed this to claude code and watched it verify the result in 7 different ways to be 100% certain, and it was just flabbergasted. Quite remarkable."

* Ollama: All Aboard Open Models

"A year and still no implementation for such a basic need as offloading MoE layers onto the CPU selectively. On llama.cpp I can get models like Qwen 35BA3B running partially on gpu/cpu with 40t/s on a laptop thanks to --n-cpu-moe but on this VC funded joke it would be simply unusable. I can't quite understand how you make a wrapper so much worse than the code you're ripping out."

* Blender 5.2 LTS

"Look at that wow: https://www.youtube.com/watch?v=gqfLYIJMv7I"

* OpenAI reduces Codex Model Context Size from 372k to 272k

"I know a lot of people like to say that compaction makes this moot, but the level of detail you lose across compaction is wildly too much for most things that I do, unfortunately."

* M-Chips: M7 with up to 1.5 TB – and why Apple is skipping the M6

"I hope they put a better connector than TB5 so we can cluster them properly at 1TB/S"

--

Your math is not mathing.

d5lt5··on Qwen 3.8
Nickname checks out.
d5lt5··on Claude Code uses Bun written in Rust now
For someone who is not a rust dev, would you mind giving me an example of this: "Nope, this is demonstrably false because Rust has its own invariants around its types and the codebase is violating a lot of them."
d5lt5··on Microsoft Can Track Users via a Windows Device ID
3 entries in the list above did exactly that more than a decade ago. My point is that the US tech is not in a process, they were at the frontier long before everyone else.
d5lt5··on Microsoft Can Track Users via a Windows Device ID
It's interesting how you think of US tech currently being in a process of becoming a tool of state surveillance and oppression that Russia and China have, while the following exists:

* PRISM (est. 2007) spied on US citizens with the help from AT&T, Microsoft, Google, Apple, and etc.: https://en.wikipedia.org/wiki/PRISM

* SpyFiles (various years, ~2011) US/EU surveillance software: https://wikileaks.org/spyfiles/

* Facebook (est. 2004):

  - Helped Cambridge Analytica (~2013) to interfere with the elections: https://en.wikipedia.org/wiki/Facebook%E2%80%93Cambridge_Analytica_data_scandal

  - Bought Onavo (~2013) to spy on competition: https://en.wikipedia.org/wiki/Onavo

  - Manufactured consent to enable genocides in Ethiopia, Myanmar, and Gaza.
* Palantir (est. 2003): https://en.wikipedia.org/wiki/Palantir

* Microsoft GSP (~2003) to enable US government to find zero-days (now with Mythos): https://www.microsoft.com/en-us/securityengineering/gsp

d5lt5··on China sentences official to death for taking $325M in bribes
The culture of bribes is a bit different in China. 'Mutually assured corruption' describes the situation better.
d5lt5··on Microsoft Can Track Users via a Windows Device ID
Have you heard of a website called facebook?
d5lt5··on U.S. government will decide who gets to use GPT-5.6
> Even tightly regulated industries with security concerns can use the latest deepseek.

That is not necessarily true, as "tightly regulated industries with security concerns" are also afraid of deepseek models generating vulnerable code. Even a possibility of that prevents those industries from deployments.

> That doesn't benefit china directly but it does serve to further undermine the lead that the american frontier labs have which limits their future ability to cut geopolitical adversaries off.

So, basically, competition is bad for US models? That argument doesn't address open-weights. And it doesn't work the other way around, because in that case China should be releasing close-weights instead.

d5lt5··on U.S. government will decide who gets to use GPT-5.6
> Yes. That’s why development and debt are hard problems. Also why calling it “goodwill” is, at best, too generous.

One can call the intent 'goodwill'. It doesn't mean that the outcome is satisfactory for your economic expectations. Judging from exceptions is not a valid approach and is a weird take.

> “The US also does very bad stuff” doesn’t make BRI goodwill.

True. I used that as an example of an alternative approach. The reader can decide which one is more 'goodwill'.

> Some even try viable (if self-interested) development policy without bombing people.

What countries are you referring to here: France (Douala and Abidjan ports, North–South railway in Vietnam), Japan (also ports in Sri Lanka, Thilawa), something else?

> Don’t make me say what I did not.

That conclusion says more about your reading than about what I actually wrote.

> Basically, that unlike your "goodwill" claim, China isn’t just giving away infrastructure for free out of the goodness of its government’s heart.

I shoot back with "Don’t make me say what I did not.", and 'goodwill' doesn't mean 'free stuff', you may want to check the dictionary ;)

d5lt5··on U.S. government will decide who gets to use GPT-5.6
It's a long stretch in mental gymnastics with no factual proofs.
d5lt5··on U.S. government will decide who gets to use GPT-5.6
Can you argue that the principle of the BRI is humanitarian and it should benefit both partners, but not equally? Imho, that policy is far better for humanity than blockading Cuba, bombing Venezuela and Iran.

> A lot of it was financed through large (sometimes unsustainable) loans to recipient countries, sometimes leading to unsustainable debt burdens, irrespective of the potential ROI for the recipient (ie Sri Lanka’s port).

I see that you blame China for Sri Lanka, while China wasn't the only creditor there.

> And the infrastructure didn’t necessarily line up with the recipient’s actual needs

Easy to say in hindsight.

d5lt5··on U.S. government will decide who gets to use GPT-5.6
How would open-weight models benefit PRC better than their own closed-weight models, but still available at lower prices? If anything, open-weights can be distilled far easier.
d5lt5··on U.S. government will decide who gets to use GPT-5.6
Well, you are wrong. Maybe you should visit and learn more about China to understand it. For starters, China's society is high-individualistic with a strong sense of community and with high respect to their elders. On the contrary, US's society is hyper-individualistic with a strong sense of family and basic respect to their elders.
d5lt5··on U.S. government will decide who gets to use GPT-5.6
I think you view this situation from the US point of view and assume that China has the same guiding principles and values in their foreign policy, for which it doesn't. They might do what you said, of course. But they very well might also treat LLMs as another goodwill investment like the Belt and Road Initiative (https://en.wikipedia.org/wiki/Belt_and_Road_Initiative) and export the capability to partner countries, for example, in Africa, to strengthen relationships.
d5lt5··on Ask HN: Will programmers write more efficient code during the memory shortage?
Ehm. I would assume your definition of writing good code fast is not what the majority of enterprise developers agrees with.
d5lt5··on How Japan's railways stayed one while splitting apart
USA is a melting pot and is not viewed as a monocultural country.
d5lt5··on How Japan's railways stayed one while splitting apart
"which they destroyed with there ideology and hate in there own country" Arguably the ideology and hate are prerequisites for a country to have a monoculture. With ideology and hate comes xenophobia and racism == monoculture.
d5lt5··on Google and Pentagon reportedly agree on deal for 'any lawful' use of AI
I did and helped d33ps33k. You are welcome!
← PreviousPage 2 of 2