HN user

rustyhancock

910 karma
Posts3
Comments315
View on HN

One of the interesting things about LLM training is that high quality documentation can get your product effectively promoted by them.

I imagine the kind of advertising you're talking about is obtained by providing that documentation and a curated training dataset to the providers.

Maybe even funding some training time.

The net result would be product placement, probably without having to declare an advert.

Kind of like when drug manufacturers provide subsidized workshops to clinicians.

It is now possible to merge multiple PDFs by dragging a PDF into the PDF sidebar.

It is now possible to add images as new pages within PDFs using the Firefox PDF editor.

I have always found it odd that browsers are also PDF editors increasingly.

Android makes this a chore with all its gestures too!

But I'm not sure it could (or that I'd want) it to supress gestures like back and close application

The README mentions it and conventionally it seems right.

But back of the envelope it seems you're shaving not much off if the latency is primarily WiFi latency (tens of milliseconds) not flash reads (microseconds?).

If you use loads of transient clients perhaps a LRU cache would help since round trips might be quite slow.

Sounds like that'll be a great read.

This is (literally) none of my business, but would you consider a cut down or basic open source version (hardware and software, with open software running in the proper hardware). Yes it'll be cloned by factories, and yes it will cut into sales no matter what proponents claim. (And yes it's probably more work than I imagine it is).

But also yes, it'll be genuine open hardware and might out live your interest in the project.

I wish they called it confirmed and not discovered

It was known about by the locals, they even gave it a name, it's call was very well known even if few had sighted it.

They did important scientific work to confirm a distinct species. But they did not discover it unless you discount the local people!

What's the problem exactly? Could you not follow the example in the text?

The standard text to build understanding in physics is University Physics by Sears & Zemansky.

It's worth remembering you're quite far from the ground in physics, and it's mostly taught with "neat" cases that give insight into physics. I.e. the thought experiment to show why kinetic energy must scale quadratically with velocity is carefully designed to show that conclusion. You shouldn't expect to have come up with it off the cuff.

I think I'd personally develop a minimal patch and then publically disclose.

I'm not sure it's be reasonable to leave an actively exploited critical bug until August. Nor would I be too interested in playing middle man or paying for support from curl to get it out.

the White House intends to generally regulate Mythos-class models (whatever exactly that means)

This is not at all surprising. And I hope people don't make the mistake that it's a "this administration" problem.

It was obviously from the early days of these LLMs that the shoe was going to drop and we (as Joe public) would not retain access. I mean that once ChatGPT3 dropped it was clear there was some level of functionality at which we would be denied further access.

The only carve out will be as per older technical innovations the US is more concerned with foreign national access than US citizen access at home.

I don't remember the details with encryption but it was basically you have to ship a breakable version for the rest of the world, and you generally sometimes ship a backdoored version.

And Anthropic is more concerned by what they are asked to do to US citizens than the broader group.

Same story with encryption, CPUs, GPUs, blah blah blah.

For now all one can say is transmission is assumed to be dramatically reduced.

The bigger risk is likely that in some the suppression is temporary or transient flares of replication under some circumstances.

The other question is, does this avoid all the sequela of HBV. It seems to reduce risk of cirrhosis atleast.

For hiv, it took many decades to be able to make the clam undetectable = untransmittable using serodiscordant couple studies.

Ikeas instructions are such an oddity.

It feels like there is precisely enough information to deduce each step. But only just enough miss one clue and you have something on upside down on step 7 that you won't notice until step 37.

I feel whoever makes them could probably make a wicked NY Times Crossword puzzle.

Importantly my work involves me often being able to look at C and think about the assembly and back and I regularly work on ESP32, ch42(riscv) and atmega avr8.

I couldn't do that with mciropython on any platform.

C is a thin abstraction, python isn't.

I agree.

You don't learn or know C++ in the way you learn or know C.

You never have the total language spec in mind. Much of it you will never (and for some of it should never) come across.

The way I think of it

C is an abstraction of the machine, so thin it's nearly transparent.

C++ is an abstraction over programming paradigms, letting you pick how you think.

Everything else abstracts the machine away, replacing it with a VM, runtime, or model of its own.

The same way a good project has a clear model of the problem it should have a clear C++ pattern in use.

And it seems very few proteins appear to be significant problems.

The most famous is the prion protein which can misfold in ways to cause a variety of contagious diseases. Like mad cow disease, chronic wasting disease, scrapie and in humans CJD and vCJD, fatal familial insomnia, Kuru, GSS.

Perhaps because misfoldings of the prion protein can convert others but why is it all affecting that same protein? Always baffled me why aren't other/many proteins suspitible to becoming a prion?

There are others we call "prionoid" because they can have shades of the catetrosphic misfolding prion can.

I know this is a crazy take. But I go feel so down trodden by many many tech corps these days I find it hard not to have a smidge of satisfaction for this guy pointing out the colossal favour research developers do for them by responsible disclosure.

That said, I feel bad for the inevitable victims of exploitation and also I am certain he will end up criminalized or as per usual the law will enforce a large corps will against him.

Yes. Definitely a Friday night after a hard week take.

Yes I'd happily buy one if it's quality matched the price and I'm sure in Japan it often does.

I have done some simple leather crafts, and I think the design clearly is suitable for building with rivets and full grain leather, if they do use that today then it'll be a spectacular product.

Texas Chainsaw Massacre was a flash in the pan (although the asymmetric horror genre continues basically dominated by dead by daylight).

Yet, the sheer exhilaration I felt the first time one of the "killers" walked past me as I kneeled in a bush was quite spectacular.

It's not the same as splinter cell (it's much more chaotic, you don't get to totally dominate the enemies, it definitely doesn't have that mindful quite as you systematically work your way through a level you know we' ll).

But the key, I can stand in the right spot and human can't see me really is its own kind of feeling.

This study tracked study resource usage in 2021 and mentions a study in 2006.

In 2006 medical students spent 10.8hours per week studying with textbooks, on 2021 4.2hours.

So under 40% the textbook usage as 2006. That's a fairly precipitous decline and it's pre-LLMs being mainstream. I down chatgpt 4/5 have sent the students back to the library!

It mentions question banks have expanded as have online resources. Also learning style has changed from lecture based to problem based learning.

I can't say this is objectively bad. But that I'm sure it contributes to narrowed knowledge bases.

https://pmc.ncbi.nlm.nih.gov/articles/PMC8651945/