Hacker News: anvuong

New comment by anvuong in "Cloudflare acquires Deno"

anvuong — Fri, 09 Oct 2026 20:41:03 +0000

It's one of those things that the technical aspect is simple but the paperwork dehumanizes me. Just a couple more bullshit engineering design documents to generate

New comment by anvuong in "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"

anvuong — Thu, 08 Oct 2026 20:19:40 +0000

I just un-retire my pair of 1080Ti for some small models development because the current GPU prices literally make me sad.

New comment by anvuong in "Trump administration is suspending Microsoft from a green card program"

anvuong — Thu, 08 Oct 2026 18:45:04 +0000

> Bring smarter, qualified people to the US. If not, they’ll find a way somewhere else

More than 70% of H1B in tech is from a single country, where cheating is considered a normal aspect of life. I'm sure most people here have at least once encountered this cultural difference in their day to day work.

While I still think H1B program is needed, it has to be revamped, putting cap per country is the most obvious first step.

New comment by anvuong in "Trump administration is suspending Microsoft from a green card program"

anvuong — Thu, 08 Oct 2026 18:25:46 +0000

This is a very well known tactic in H1B factory like Tata and Cognizant, people has been complaining about this for years. Not sure about Microsoft though.

New comment by anvuong in "Integer multiplication below n log n"

anvuong — Wed, 07 Oct 2026 05:00:38 +0000

It shows that nlogn is not the limit, how much better we can go? Not sure, probably not much, but breaking the barrier is important. Like in marathon for years nobody thought human could break the 2 hour barrier, then someone did and now it's become normal occurrences.

Multiplication can be done by FFT, which is an exceptionally efficient algorithm that has nlogn complexity, you'd be called crazy if you claim you have something more efficient than FFT.

New comment by anvuong in "Mistral Large 4"

anvuong — Tue, 06 Oct 2026 18:03:47 +0000

Both, especially for training. Astra and Fable were presumably trained on cluster of 100,000k GPUs, or at least a couple of 10Ks.

3,800 GPUs is nothing in the frontier side.

New comment by anvuong in "Plain text is still one of the best technologies we have"

anvuong — Mon, 05 Oct 2026 20:38:42 +0000

ASCII was amazingly efficient for conveying English text. Then we needed to encode multi languages and emojis, the resulting Unicode is just a mess.

New comment by anvuong in "Anthropic reported diary entry to police, woman faces felony charge"

anvuong — Mon, 05 Oct 2026 18:38:37 +0000

We are snowballing to Minority Report ...

New comment by anvuong in "Pixel 11 doesn't yet meet the GrapheneOS security standards and may be skipped"

anvuong — Mon, 05 Oct 2026 18:12:39 +0000

Why are you surprised? Since when does China care about privacy?

New comment by anvuong in "ArXiv's Updated Rate Limit Policy"

anvuong — Thu, 01 Oct 2026 22:02:32 +0000

Any example of high SNR authors who consistently putting more than 2 papers months after months after months for this to be a problem?

New comment by anvuong in "Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms"

anvuong — Tue, 29 Sep 2026 05:03:25 +0000

> Classifiers (instead of LLMs) return results with confidence scores.

This is also bogus unless you are talking about Bayesian inference. No classifier can output CI for a single point estimate. In every ML theory textbooks worth their $, it's always stressed not to treat these sigmoid'ed or softmax'ed numbers as probabilities or confidence scores, there is no such thing as CI for point estimate.

New comment by anvuong in "Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms"

anvuong — Tue, 29 Sep 2026 04:59:06 +0000

Is it though? I got sub 10ms for 200M params vision model way back in 2020 with TensorRT optimization running on RTX 2080Ti. 30ms for 0.8B classifier model doesn't sound special to me.

New comment by anvuong in "Sonnet 5.5"

anvuong — Tue, 29 Sep 2026 01:40:41 +0000

Which provider are you using Mimo from? The official Xiaomi one seems to be very slow/buggy for me in the last few days.

New comment by anvuong in "Sonnet 5.5"

anvuong — Mon, 28 Sep 2026 21:12:16 +0000

Does AI-style writing start bleeding into the comments, or is HN now also full of bots like reddit?

New comment by anvuong in "An agent used DNS to reach an external chatbot"

anvuong — Sun, 27 Sep 2026 05:56:21 +0000

My guess is they do, it just did not function probably. These frontier models are trained on massive datacenters with hundred of thousands of GPUs. There must be many safeguards before a run can be automatically stopped.

New comment by anvuong in "DeepSeek Elastic Compute (DSec)"

anvuong — Sun, 27 Sep 2026 00:48:38 +0000

Lmao with the conspiracy. Large scale experimental research is always like this, many papers in experimental physics have pages of authors.

New comment by anvuong in "MiMo v2.6"

anvuong — Tue, 22 Sep 2026 03:49:56 +0000

> the pro run restarted at step 17 due to a GPU OOM issue caused by expert load imbalance. we have adjusted the training parallelism strategy.

no idea why I laughed so hard at this, feel so familiar

New comment by anvuong in "iPhone Duo"

anvuong — Wed, 09 Sep 2026 21:24:43 +0000

Not sure how Celsius is worse? Its unit is the same of Kelvin, just shifted by 273. 0C is freezing, 100C is boiling. Some people say Fahrenheit is more representative of how people feel, but that is very subjective and biased towards the West and Northern hemisphere.

New comment by anvuong in "LG TVs caught spying even when offline or on standby"

anvuong — Tue, 08 Sep 2026 20:40:42 +0000

How is projector superior? More often than not, with the same amount of money spent, projector produces worse image quality, consumes more power, and don't last as long.

New comment by anvuong in "Shutting down our public encrypted DNS"

anvuong — Fri, 04 Sep 2026 22:24:42 +0000

What is intrinsic difficulty? A product at scale has many intrinsic dimensions, not just technical. Saying scaling is not an intrinsic difficulty is pretty weird given a highly scalable product usually looks nothing like their 1-user counterpart even when they have the same functionality.