Hacker News: willy_k

New comment by willy_k in "Gemini 4 Hydrogen"

willy_k — Thu, 12 Nov 2026 19:49:34 +0000

Carbon can be more performant while being a smaller model.

New comment by willy_k in "Claude Fable 5.2 and Mythos 5.2"

willy_k — Tue, 10 Nov 2026 06:00:10 +0000

Google isn’t interested in the same way OpenAI/Anthropic are, Google has other sources of like, actual profit, to factor in, eg to resource considerations.

New comment by willy_k in "Kagi Search for Everyone"

willy_k — Thu, 05 Nov 2026 21:21:53 +0000

I’d imagine it searches google maps.

New comment by willy_k in "The Election Day Revolt That Threatens America's AI Boom"

willy_k — Thu, 05 Nov 2026 17:54:32 +0000

There’s a lot of money moving into wells, not necessarily much moving in ways that’s everyman-liquid. I mean stock go up, so some people can maybe capture some of that value, but generally once these places are built (temporary construction business), they retain a dozen or less employees and tend to sit around being a burden on local resources.

New comment by willy_k in "Progress on provable inference for large language models"

willy_k — Wed, 04 Nov 2026 21:47:13 +0000

In modern sv-wallstreet corporateness you need to be bashful over any attempt to do good for good’s sake.

New comment by willy_k in "GPT-Rosalind-6 System Card Addendum"

willy_k — Mon, 02 Nov 2026 20:50:31 +0000

Is it using comparable tokens per task?

New comment by willy_k in "Anthropic S-1"

willy_k — Mon, 26 Oct 2026 15:00:59 +0000

“Beleive” in the sense that Tom Cruise believes in the cosmic DC8?

Anthropic [employees] huffing their own koolaid fumes aside, it’s extremely difficult to believe the claims that these mega valuations are substantiated on (*one* company somehow assuming a meaningful fraction of global research and production)

New comment by willy_k in "Kimi K3.1 Pro and K3.1 Pro Highspeed"

willy_k — Sat, 24 Oct 2026 03:33:21 +0000

Granted we know that 6 Astra was built to do more computation between each token.

New comment by willy_k in "Kimi K3.1 Pro and K3.1 Pro Highspeed"

willy_k — Sat, 24 Oct 2026 00:24:01 +0000

The difference is that it’s not subtracting an otherwise present part of training, it’s choosing not to continue to push it with RLVR. Models become capable like this when they are specifically tuned to be good at cyber tasks, so without the RLVR (Reinforcement Learning w/ Verifiable Rewards) they never pick up this level of aptitude.

I agree that it’s not obviously sustainable, but it’s better than specifically training these tools to be good at wreaking havoc, and then putting on a shocked pikachu face when they wreak havoc (ahem, Sam Altman).

New comment by willy_k in "Meta ordered to pay $25B penalty in New Mexico Cambridge Analytica case"

willy_k — Thu, 22 Oct 2026 00:58:47 +0000

In case it hasn’t become abundantly clear to everyone by now: this is how Facebook operates. The people who do this stuff get promoted, and vice versa. This will keep happening until those people are scared straight, and the incentive structure that turns people into this are abolished.

New comment by willy_k in "Patch the Planet: global response coordination as agents uncover vulnerabilities"

willy_k — Wed, 21 Oct 2026 19:51:20 +0000

Mind elaborating on that Deepseek point? I am aware of the USG decision, what exactly do you mean?

New comment by willy_k in "DeepSeek banned from U.S. federal civilian agencies under new DHS order"

willy_k — Wed, 21 Oct 2026 04:56:28 +0000

Mostly referencing the hugging face incident. It’s a felony.

New comment by willy_k in "DeepSeek banned from U.S. federal civilian agencies under new DHS order"

willy_k — Tue, 20 Oct 2026 19:46:56 +0000

Yeah, but it’s different if they’re our backdoors!

New comment by willy_k in "DeepSeek banned from U.S. federal civilian agencies under new DHS order"

willy_k — Tue, 20 Oct 2026 19:22:35 +0000

So much accountability!!!

New comment by willy_k in "Lula supporters protest against Meta to denounce its 'electoral interference'"

willy_k — Tue, 20 Oct 2026 03:58:13 +0000

“Unexpected” is a little much - his approval rating is under the gutter.

New comment by willy_k in "MiniMax M3.1 Flash"

willy_k — Mon, 19 Oct 2026 23:26:12 +0000

Fair. I wasn’t advocating for “holding them right”, I feel it’s an important nuance to keep in mind when conceptualizing them, i.e. it’s procedural, determistic, yet both inherently and in practice extremely chaotic and therefore not practically deterministic software. So, in essence - software.

New comment by willy_k in "MiniMax M3.1 Flash"

willy_k — Mon, 19 Oct 2026 23:23:43 +0000

Ah, that’s interesting. That is what I was attempting to cover with my last parenthesis but I was not aware of that!

New comment by willy_k in "MiniMax M3.1 Flash"

willy_k — Mon, 19 Oct 2026 19:49:05 +0000

> LLMs are nondeterministic

Being pedantic - LLMs are deterministic software, that is chaotic and typically deployed with a PRNG seed. An LLM ran using the same seed will produce the same output consistently (theoretically - in practice this isn’t always the case because of changing factors wrt the execution environment given how sensitive/chaotic models are).

New comment by willy_k in "Mercury 3: Diffusion at scale"

willy_k — Mon, 19 Oct 2026 18:21:20 +0000

Google more or less ordinated the category. They have now seemingly dropped it, ostensibly because it doesn’t scale economically like transformers do (via batching requests).

I think though, that if Inception can keep up these speeds and prices, assuming diffusion continues to improve (likely), that Flash and Luna could lose out for many use cases.

New comment by willy_k in "Data Centers Are Turning Trump Voters into Democrats"

willy_k — Mon, 19 Oct 2026 18:01:00 +0000

Maybe not, they definitely do want right of ways for power lines on people’s property though. And they use the classic eminent domain playbook to get at it. Specifically referencing power from/through West Virginia to NoVa.

They offer a few security and telemaintenance jobs after construction.