VHRanger

I'm curious why seemingly none of those projects tried using browsers JS/wasm execution as a sandbox instead

Nothing more reliable than good old SPLARGLE kitchenware

Do you have a better source if I speak fluent French by any chance?

Thanks for the recap

A lot of those are spending 10-100k a month when they could be spending 1k-10k a month with a more lean infra

On vllm with a 5090 I get 120-180TPS with the awq 4 bit quant + MTP speculative decoding

For gemma4 26B, same quantization, I get >200TPS.

Also note that qwen is extremely inefficient in reasoning; the reasoning chains are ~3x longer than gemma on average

The issue is that benchmarks that look insightful will end up being gamed by labs quickly (Goodharts law)

The best LLM benchmarks test around the margins of those behaviors, tasks that are difficult and correlate with usefulness while being removed enough to stay unpolluted

Radeon 9700 pro or intel arc b70 (both $1000-1400, 32GB, 650GB/s bandwidth), or ryzen AI max 390 (more vram, less bandwidth)

The local inference space is pretty good nowadays.

The S3 API doesn't work like normal filesystem APIs.

Part of it is that it follows the object storage model, and part of it is just to lock people into AWS once they start working with it.

That's not surprising; Opus & Sonnet have been regressing on many non-coding tasks since about the 4.1 release in our testing

There's presumably engagement on those two.

It's better to have a smaller core of highly engaged people than a mass of disengaged eyeballs glazing over.

Pytorch is such a maddening mess of half implemented research features in a state of Heisen-deprecation, Jax becomes more appealing to me by the day.

Remember when we had the term "spyware" for a class of malware?

I remember

If rust is not in the HN title and fire emojis in the readme, it doesn't come from the Rust region of France.

It's just sparkling memory safe high performance software

Mexico has much lower economic development and higher crime.

Canada has similar levels of economic development.

It's really not that complicated: controlling for general crime levels, guns drastically increase murder rates.

Yeah I'm another pop os user.

Cosmic works great for a laptop. But it's a PITA for a desktop. It doesn't deal with multi monitor setups well. There's a recent new bug where the system hardlocks on monitor power state changes, which is unacceptable.

So: great for single screen laptop, not good for desktop or server

Of course within the USA the state levels laws will do little. There's free movement between states!

Compare the USA to Canada, where you can't bring a gun easily. You'll see Canadian murder rates being very low. Even controlling for similar factors at the city or neighborhood level.

Of course I'm blaming the gun: it's pretty hard to kill someone with other weapons. Stabbings are often survived, even.

Sure poverty explains crime, and murder is the ultimate crime.

That said, my point was that a place like Rio, where you feel alertness at a physiological level by the constant lack of security, still has a murder rate around Houston, a vastly richer and safer city.

And Brazil really is a good comparison in my opinion: the economic inequality is actually worse than in the USA, and they both have the slave holding history leading to concentrated poverty areas with high ethnic segregation

I don't personally think that the upsides of the US gun laws are worth anything near the downsides being paid.

Regarding the police, American police is notoriously prone to violence compared to other developed countries.

Ah, some egregious misuse of statistics!

Murder is a byproduct of crime. Crime is, largely, downstream of economic conditions with some obvious caveats.

New Hampshire has the 2nd lowest crime rate of the USA states. You could make the same argument for, say, Switzerland (high gun ownership but no crime/murder). But no one would be surprised if you had high gun ownership in Monaco.

Similarly for the ethnic argument you're trying to make: Majority black neighbourhoods in the USA tend to be poor. They also tend to be near more affluent places. Unlike poor white neighbourhoods, which are on average more rural in the USA.

Being poor, and being next to rich people, and being excluded from legal increases of becoming rich, will increase crime.

This should be obvious. Brazil has famously Favelas right next to wealthy areas and has a persistent crime problem for example.

---

In short, it's really incredible how far some Americans will go to deny the obvious truth: *gun prevalence increases deadly crime*.

Sure, some cultural factors will increase crime/violence on the margin. But the reason y'all have a bunch of shootings is that you have a bunch of guns to do shootings with. That simple.

And I presume that doubled price is before you look at the workstation class motherboards, which also tend to be much more expensive.

Thanks for the info on the hardware quirks, useful to know!

We seem to be arriving at a cambrian explosion of viable hardware these days between ARM and x86, soldered vs DIMM, etc.

It's refreshing coming from 20 years of x86 being all that matters.

It's really hard to push DDR5 past 6000MT/s on 4+ DIMMs it seems.

I had to get everything top spec to fit 4 channels of 6000MT/s on my 9950x (asus proArt motherboard and the top tier trident neo RAM sticks) -- otherwise it's reportedly unstable.

Youre aware that the rest of the planet have stricter gun laws and the American problems are fairly unique?

This is even after controlling for things that exacerbate crime like high economic inequality.

For instance, Brazil [1] (a much poorer and more unequal country than the USA) has lower murder rate than a lot of cities now than the USA. The murder rate of Rio seems to be about on the level of Houston (17/100k), or about a third of Detroit (47).

But Rio clearly has __a lot more crime__ than Houston. It's palpable when you're in either city. Even with the Favelas and heavily armed gangs, the murder rate is comparatively low because *normal people dont have guns at nearly the same rate*.

And it shouldn't take a leap of faith to figure out that higher gun ownership leads to more deaths. Guns are the one tool we have intentionally made to cause death.

1. I'm aware that Brazil has a higher murder rate, but comparing cities is a better pick. The northeast of Brazil is in another league than anywhere in the USA in economic conditions; it's not comparable. The only city I can think of with USA levels of economic development would be Florianopolis (murder rate 7/100k) or maybe Balneario Camboriu, or some parts of Sao Paulo like Vila Olimpia.

GDDR7x doesn't come in dimm factor?

In general soldered ram seems to get much higher bandwidth than removeable ram. See ryzen AI Max vs 9950x max ram throughputfor example

paperless-ngx?

This t5 is multimodal.

Also a hint: you can create a finetuning dataset from a frontier LLM pretty easily to finetune those t5 and effectively distill them pretty fast these days

In general encoder+decoder models are much more efficient at infererence than decoder-only models because they run over the entire input all at once (which leverages parallel compute more effectively).

The issue is that they're generally harder to train (need input/output pairs as a training dataset) and don't naturally generalize as well

Encoder/decoder is much, much more efficient for finetuning and inference than decoder-only models.

Historically T5 are good when you finetune them for task specific models (translation, summarization, etc).

A store of value is an asset with as close to 0% volatility in price as possible.

Bitcoin is a speculative asset: it has very high price volatility. It is not a store of value in the proper term.

Same, I got 96GB of high end 6000MHz DDR5 this summer for $600CAD and now it's nearly triple at $1500CAD

Kagi has Mistral as well

If you send a message in our discord we can give you a beta key I think. Costs nothing to ask over there, say Matt R sent you