blovescoffee

I feel like in this instance “them” is pretty clearly news agencies

I guess you aren’t the target audience?

This is cool. Though I’m a little confused on who the ideal ideal target is. Is it focused on a non technical user or a technical user. Based on what stripe does, I think these are two separate product categories and combining them could result in a worse result for both user types. But I haven’t actually tried implementing this yet so I can’t speak too much on it. Looks cool!

True, although I always had the impression that the violence in large urban centers leaned towards "organized crime" and less "cartel" but that's just my impression.

I've read some weird hacks but the traditional non-centrifuge way to clarify the stock is to use some protein like ground meat or egg white to "absorb" any discolorant, protein, etc.

At least in mexico, criminal violence is definitely tied to poverty.

They're linked, but don't you think general corruption is even more damaging to Mexico? In cdmx we're quickly running out of water in major part due to corruption. Anyways, most cartel activity I've heard about from friends has been on highways, outside of major cities.

No. Maybe a large portion of the Foreign investment is owned by Europeans but "most of mexico and latam" are definitely not owned by Europeans. Have you ever spent any time in LATAM?

True!

I'm sorry this may come off as rude, not my intention: The Gary Tan interview explicitly says those things, I'm not sure that's really your "take".

To clarify, I meant neural network architectures not chip architectures.

Yes that's stretching the truth

Did they invent new DL algorithms and publish them? If I remember what I heard in the interview correctly, this targets existing architectures.

Their startup is addressing computing bottlenecks so that's what I addressed. Supervised training dat isn't a bottleneck on LLMs, Diffusion models, or any of the hot areas at the moment.

This interview makes their product seem like BS. First, they literally cannot simply explain the problem or solution. Regardless, their pitch is that they're building a more power efficient probability distribution sampler. No one in AI research thinks that's a bottleneck.

edit: btw the bottleneck in AI algos is matrix multiply and memory bandwith.

Wow that lineage includes Carl Gauss and Gilbert Strang

Maybe not 100s of UI teams but I worked at a large bank and there was dozens of full stack teams that each had to write front end code. We used an internal UI library. Otherwise it would be chaos.

I have a bachelors in math (i.e. I know what an non-abelian group is and what operators are) but don't know much physics, can you explain more?

Argentina is an economic powerhouse of the region. A very rich country, with an enormity of poverty. Poverty in Argentina has run in cycles. Current events are really quite current and you can't pin them solely on events from the 70's. An easy place to start research would be https://es.wikipedia.org/wiki/Crisis_econ%C3%B3mica_y_sanita...

You have no idea why Toyota or Adobe are funding computer vision research?

I agree with your premise that the US did substantial harm in LATAM including Argentina. I reject the notion that the issues and history in Argentina are so simple that we can pin its poverty rate on the CIA. Can you at least substantiate how the Peron government leads to the poverty rates that currently exist in Argentina?

It's non-sense.

I worked there for 1.5 years. There's lots of smart people. There's lots of people hiding in the bureaucracy of a large bank that get away with very little action. I haven't worked at Google so I can't comment on that. Overall it was a good experience but things moved very slow.

What did you previously think a vector db was?

Halfway through the article and we have

unsafe { asm!( "mov edi, 42", "mov eax, 60", "syscall", options(nostack, noreturn) ) // nostack prevents `asm!` from push/pop rax // noreturn prevents it putting a 'ret' at the end // but it does put a ud2 (undefined instruction) instead }

and

> We will need to tell the C compiler that we’re providing our own entry point, telling it not to include it’s own start files.

So it's a Rust program but it's just calling inline assembly and using a C compiler?

I've been using ST for 10+ years. I pulled a copy and installed some extensions recently. Most extensions I want have multi-year gaps since last updated. ST crushes VS Code performance wise but the DX I'm used to with Code is much better than in ST because of the community.

Very simplified There are arrays (matrices) that are length 10M inside the model.

It’s difficult to make that array longer because training time explodes.

Because people tend to write c++ for performance reasons and the perf profiles of std::vector are not always sufficient

For what it's worth, I'm jewish and there are many jews who would disagree with my answer to the question you're responding to - just keep in mind how much diversity in thought exists. Though, the sentiment of your answer I do tend to agree with.

The very beginning of the article discusses what "full self driving" means and also points out how important it is to define terms. I'm not sure your comment is a fair response to this particular article.