HackerTrans
TopNewTrendsCommentsPastAskShowJobs

earhart

no profile record

comments

earhart
·2 years ago·discuss
I still wish Tile had caught on; einsum is really nice, but sometimes I want a dilated convolution, or a maxpool.

(OTOH, I’m not an einsum expert; please feel free to delight me by pointing out how it’s possible to do these sorts of things :-)
earhart
·2 years ago·discuss
Anyone know how to get mail to [email protected] to not bounce back as spam? :-) (I promise, I'm not a spammer! Looks like a "bulk sender bounce" -- maybe some relay?)
earhart
·3 years ago·discuss
https://www.penny-arcade.com/comic/2023/01/06/i-have-no-mout...

I honestly worry about this - I've been tinkering with ideas to try to build towards AGI, and I'd love to share them publicly to get feedback ("This is dumb and here's why" would be enormously valuable to me), but it's hard to work openly, because while I do think capitalism has been an overall good, the capitalist imperative always seeks slaves, and I'm really not excited about helping the people who'd be trying to build a new slave class.

Thoughts? Is there a ethical way to work openly on AGI?
earhart
·3 years ago·discuss
IMHO, it comes down to the software.

It turns out you need very different kernels for good performance on different GPUs, so OpenCL is a nice tool, but not sufficient; you need a hardware-specific kernel library.

From the framework side, each integration is relatively expensive to support, so you really don’t want to invest in many of them. Without some sort of kernel API standard, you’re into a proprietary solution, and NVidia did an amazing job at investing in their software, so that’s the way things go.

I think we had a pretty solid foundation for doing something smarter with PlaidML, but after we were bought by Intel, some architectural decisions and some business decisions consigned that to be a research project; I don’t know that it’s going anywhere.

These days, I’d probably look into OctoML / TVM, or maybe Modular, for a better solution in this space… or just buy NVidia.

(I worked a bit on Intel’s Meteor Lake VPU; it’s a lovely machine, but I’m not sure what the story will be for general framework integrations. I bet OpenVINO will run really well on it, though :-)
earhart
·3 years ago·discuss
Just curious - what was your motivation?
earhart
·3 years ago·discuss
FWIW, Framework also announced a new battery with the same form factor but another 10%-ish capacity. And because the system's amazingly easy to disassemble, if your battery starts to lose capacity, it's trivial to swap in a replacement.

I do the "unplug and spend a couple of hours hacking without a charger" thing all the time with mine, and mine's using the previous-generation battery. Not running Windows seems to help a surprising amount.
earhart
·3 years ago·discuss
Neat; thanks for posting this! I just started tinkering with ideas in the same line of thinking a couple weeks ago; it’s good to see that there’s prior art to learn from.

I landed in the same place they did, on using XML. I think there’re a couple of advantages over S-Expressions:

* There’s always a tag. So it’s like everything is a special form. Being able to omit the tag in sexprs is nice syntactic sugar, but I’m not expecting that anyone would be writing this code in text by hand (although it’s useful to preserve the ability to do so), and I’d rather have the clarity and explicit terminators.

* Namespaces are a killer feature - that’s what makes it an open datastructure, allowing you to incorporate multiple dialects at different semantic levels. (You can tell I’ve been using MLIR a lot the past few years… :-)

* There’s a whole ecosystem of standards and tools around it. Need a standard fragment identifier? Or to canonicalize an entity? There’s a good body of specs, carefully worked out by people more experienced than I.

I dunno about their Python approach; I was thinking of a structure editor (and trying to avoid all the historical pitfalls of structure editors, which will be tricky). We’ll see how it goes. :-)

Thanks again for the pointer!
earhart
·3 years ago·discuss
NB This is even more important in VM scenarios, where second-level address translation means something needs to walk a guest-physical-to-system-physical map for each level of the guest-virtual-to-guest-physical map. So TLB locality becomes even more important, and using huge pages cuts down on a multiplier in resolving TLB misses.
earhart
·7 years ago·discuss
Software has security bugs, and enough people are evil that you will get burned with that approach. I’m a professional in this space, and I really can’t tell whether some random app is going to be evil or not; how is a non-techie supposed to figure it out?

The real problem is that Google’s been unable to either secure Android or prevent malicious apps from showing up even given their locked-down store.