Hacker News Reader

Tuna-Fishon Gemini 4 Argon

The ads in google search are ~55% of their revenue and ~70% of their profit. Will the LLM drive as much traffic to the ads as the current google search? Will it feel as organic to the users as the current solution, or will LLM hawking a product seem forced by the users?

Tuna-Fishon Micron CEO Says Memory Supply Will Be Much Tighter in 2027 and 2028 Than in 2026

Well, CXMT is not gated by ASML machines because they can't buy them. They can raise their capacity because they have access to the domestically produced steppers. They are not as good as the ASML ones, and if the memory prices were not bonkers, they'd never make a profit using them, but as it is the low yields and high die sizes don't matter.

Tuna-Fishon Micron CEO Says Memory Supply Will Be Much Tighter in 2027 and 2028 Than in 2026

They are not prioritizing extremely high prices. TSMC has a similar waiting list, so does ever other tool manufacturer. Memory capacity is being increased approximately as fast as it can be.

Tuna-Fishon The state of SIMD in Rust in 2026

Why? AVX-512 is the best SIMD ISA anywhere right now. And the width is its least important feature.

Tuna-Fishon Topcoat is pushing the boundary of server applications with Rust

Auditing Rust is much easier and faster than either Go or Python. They both contain a lot of footguns that are not always immediately apparent from visible code. Rust is much more explicit.

Tuna-Fishon Topcoat is pushing the boundary of server applications with Rust

I've found that the guardrails of a strong, expressive type system are amazing for LLMs, especially if you don't use the top-of-the-line models but stick to the cheaper options. Agents need feedback to do their magic, and the type system is great for that.

But on the other hand, the compile time of Rust is really counterproductive. On large, established projects, most prompts I write now spend more time compiling on my machine than they spend outputting tokens. In a way this is great because the tokens are the expensive part, but it does mean that when faster LLMs will come, they will not meaningfully improve iteration speed for me.

I wonder how much of the compile time of Rust is inherent to the type system. There might be room for a language with slower runtime speed (GC?), but as good of a type system as Rust, so long as compile times are much faster. Switching languages has never been easier, anyone have any suggestions? IMO hard requirements are algebraic types and error handling based on them.

Tuna-Fishon Mercury 2.5 LLM hits 770 tokens per second

The Taalas architecture makes loading weights free, but they basically has to pay the same silicon for every weight, whether it's used or not. MoE models are more efficient than dense models per weight you load, but less efficient per weight you have to store.

MoE models are the path to local models with traditional system architectures, but they are antithetical to what Taalas was doing. If you spent all the money to etch 125B weights into silicon, you'd want to activate them all for each token, instead of only touching 6B. You cannot match the 125B sparse model with a 27B dense one, but you might be able to match it with a 60B or so one.

Tuna-Fishon Mercury 2.5 LLM hits 770 tokens per second

You don't want to use a sparse model for a Taalas-like design. Something like a Qwen 3.8 27B makes much more sense.

Tuna-Fishon Claude discovers a novel enzyme system with CRISPR-like repeats

https://xkcd.com/1838/

We might have built them, we sure as hell didn't design them. And no, we do not have a decent idea about how it works.

Tuna-Fishon Black Holes or Black Hole Stars? Astronomers Spar over 'Little Red Dots'

The light curve is not smooth, there is a clear spectrum caused by absorption lines. You find the pattern in the lines that you know to be hydrogen, and measure how much redder it is than it would be without redshift.

Tuna-Fishon Black Holes or Black Hole Stars? Astronomers Spar over 'Little Red Dots'

Yes, and we are already accounting for it. The objects as seen by us are actually very deep in infrared, when you correct for the redshift to match the key absorbtion lines of hydrogen, they still remain very red.

Tuna-Fishon How SpaceX streamlined the Raptor engine

In SSME, most of the LOX never goes through a preburner.

The SSME uses the same kind of fuel-rich combustion chambers for both preburners. This presents a problem for the oxidizer side because any leaking along the shaft between ox pump and turbine in either direction would be catastrophic. This is resolved by an extremely expensive and complex multi-stage helium-purged rotating seal.

One of the cool parts about most FFSC designs, and the raptor in particular, is that they can afford slight leaking from high pressure to low pressure through any seal in the system and it's broadly safe, and only marginally reduces performance. This allows them to be made much more cheaply.

Tuna-Fishon How SpaceX streamlined the Raptor engine

Eh. A lot of materials get their strength from being worked, which 3d printing doesn't do at all.

Tuna-Fishon How SpaceX streamlined the Raptor engine

I'm pretty sure that the heat exchanger is going to end up being one part they have to put back in. Pushing CO2 and H2O (both ices at prop temp) into the tanks just seems so ill-advised to me. The amounts of contamination are small, but there are just so many different ways that can go wrong.

Tuna-Fishon Backprop Alternative: Augmented Lagrangian Predictive Coding

Backprop has significant costs, if a local solution truly matches its performance, it will be dropped like a hot potato.

Tuna-Fishon Backprop Alternative: Augmented Lagrangian Predictive Coding

> You can do that with backprop too. Nothing says you can't inference on a set of weights at the same time as you produce an updated copy from them.

Nothing but economics.

Inference requires dramatically more memory, meaning it gets radically more expensive, if you are doing backprop alongside. This is why it is not done, despite the obvious advantages. If this doesn't require storing as much data for updates, it might make such systems more feasible.

Tuna-Fishon Amazon vs. Perplexity – U.S. Court of Appeals for the Ninth Circuit

I have clearly communicated the constraints of use on my site using robots.txt, which is the established standard for communicating things like these. You are expected to fetch it and figure it out. It is not necessary to build walls and lock doors when a no trespassing sign is clearly visible.

Seriously, though, making robots.txt have the weight of law would be a massive improvement for the ecosystem in general.

Tuna-Fishon Steam Frame starts at $1059

I think having a horizon is very useful for the immersion.

If flight sims were less of a niche category, I think VR would be mainstream. VR just improves them that much. Sadly, the entire flight sim market is probably insufficient to sustain the VR market.

Tuna-Fishon Houthis 'take control' of key island in global shipping route

Aren't those really easy to explain?

EU opposes the Houthis because they are a horrifying slave state that distributes children as sex slaves to their troops.

EU opposes Russia because it is a horrifying fascist dictatorship that has repeatedly invaded it's neighbors, and constantly talks about it's historical claims to EU territory.

In the NA conflicts it's harder to draw such a clear line, because the conflicts typically have about 15 sides each, but the basic algorithm you can use is generally that EU tries very hard to pick sides based on morality, not geopolitical interests.

Tuna-Fishon Rust is tier-1 language at Microsoft

Rust allows much more aggressive optimization of the reference types than C++. If you have a &T, the compiler can assume that it will not change over the entire lifetime of the reference, and reorder things even past things that a C++ compiler would never reorder across. This can easily bite you badly if you have the mental model of C/C++ pointers and step into unsafe land. There is a separate warning for:

    unsafe {
        mem::transmute::<&T, &mut T>(t) //takes a &T, returns a &mut T
    }

in the compiler because this is something that a lot of people think might be safe (I'm running single-threaded, everything would be so much simpler if I just mutated this bit while no-one's looking), but is in fact pretty much always UB of the nasal demons type. But there are more ways to step on this problem than the most apparent way, and the compiler is not able to protect you from all of them.

To be clear, just declaring a block to be unsafe does not immediately do anything in Rust, it just allows a set of primitives that are not normally available, so it is possible to use unsafe judiciously without immediately stepping into a million landmines. You just have to be careful and ideally read the docs and the nomicon page for the operations you do, especially if they are very long.

Tuna-Fishon DeepSeek v4.1 Flash

Or one Medusa Halo.

Tuna-Fishon Growing proof that autonomous cars save lives

If nxm isn't saying that, I cannot imagine what the point of their comment was? Would you care to explain what you think it was?

Why would anyone ever type that comment as a response to the comment above?

Tuna-Fishon Private German rocket makes history, reaches orbit from European soil

You seem to be confused about what organizations do what.

ESA neither launches or builds rockets. They are an end-user, and they own the launch infrastructure in Kourou. Both these roles will exist long after Ariane is gone.

Arianespace does not build rockets. It's a launch services provider, they purchase the rockets from other companies, handle all the logistics and manage the launchpad. It's in no way married to Ariane 6, and it has in the past and still does sell launches on multiple different rockets from different companies. There will be demand for these services even after Ariane the rocket is gone.

ArianeGroup is the company that builds Ariane 6. They are a complex co-owned project between Airbus and Safran, with workshare distributed between France and Germany. They make other things than the Ariane 6, notably the French SLBM. Eventually the Ariane 6 will go away, because it will be obsolete, but that will not necessarily mean that ArianeGroup will go under, because they have an another subsidiary called ArianeWorks developing a new rocket, called Themis.

Tuna-Fishon Private German rocket makes history, reaches orbit from European soil

Musk has benefited enormously from public contracts and policy, but all the questionable ones are in Tesla (and especially originally solarcity, but that got folded into Tesla).

The government contracts that made SpaceX were all open bid, where SpaceX bid an better solution than anything that any of the established companies did, at literally half the price. Falcon 9 was a massive gift to the US government, in that it drove the entire industry to halve the prices they extracted from the government. Dragon was just as good.

That they also built a profitable launch company out of it just highlights how much of an outlier SpaceX is. They sold the US government a dollar for fifty cents, made a profit doing it, and ended up driving half the rockets already flying out of the business.

Tuna-Fishon Private German rocket makes history, reaches orbit from European soil

ESA is going to remain on the mission/payload side. ArianeSpace makes sense as a launch services provider handling everything that's not the rocket. ArianeGroup (the company that actually builds the rockets) will just become the French rocket company that will end up competing with the Italian and German ones.

Tuna-Fishon .gitignore Everything by Default

That's fixed by having the CI server compile the code and run tests, and having it fail the merge and publicly shame the offender in slack when that happens.

It's often said that you can't fix behavioral problems with technology, but I've found that tooling that strictly enforces rules is really useful.

Tuna-Fishon Qwen 3.8 27B available on Cerebras at 1500 tokens/s

Yes and no. A single chip cannot have it's weights adjusted, once it's out, it is what it is.

But also, the model weights are in a single mask rom layer, high up in the metal stack. They could manufacture the die specialized for a given geometry of a model up to that layer, wait for updated weights, and then get the final product out in weeks after they got the weights, instead of many months which is what it would take to redesign the whole chip for the new weights.

Tuna-Fishon Saab has unveiled its A3 collaborative combat aircraft concept

Plenty of people wanted to do the work. Lot of European politicians were desperate to do the work.

But it turns out that when a fascist has nukes, there's not much you can do to convince them to not be fascist.

Tuna-Fishon Saab has unveiled its A3 collaborative combat aircraft concept

We unfortunately have not found a way to turn our kleptokratic dictatorship neighbor into one.

The greatest failure of the past 20 years was the belief that if we open up trade and grow rich together, rational self-interest will lead to liberalization and peace. Instead, it just gave our enemies more resources to spend on self-destructive imperialist ambitions.

Tuna-Fishon Casey Muratori – The Root of the Root of All Evil – BSC 2026 [video]

He tracks down where the statement "Premature optimization is the root of all evil" actually comes from, and whether the assumptions set out in the text where it was published still hold.

It doesn't really contain anything new or earthshattering, but if you find diving into the history of programming as a discipline interesting, it might be for you.

I would recommend it to anyone who liked the rambling style of "The big OOPs", but imo that one was better. I still liked it.