hedgehog

33 ms·

721.
by hedgehog 3y ago
I hear the sentiment but the paper makes the argument that some of that time traveling can be replaced with knowledge available at compile time. Sort of like calling the stores to check inventory before leaving the house. The BasicBlocker i
722.
by hedgehog 3y ago
Correct, I was responding to parent writing "At what point does the complexity of CPU architectures become so difficult to reason about that we just accept the performance penalty of keeping it simpler?" I think we may be seeing a
723.
by hedgehog 3y ago
There's nothing conceptually hard but it's really a lot of work. In addition to the items you listed there's the actual compute kernels or compiler to generate those, and then porting frameworks over (PyTorch etc), and then d
724.
by hedgehog 3y ago
In certain cases isolation and simplicity overlap, I suspect for example that the dangers of SMT implementation complexity are part of why Apple didn't implement it for their respective CPUs. Likely we'll see this elsewhere too, f
725.
by hedgehog 3y ago
Well that combined with lack of working fuel gauges, had those been working the error would have been caught before they even pulled back.
726.
by hedgehog 3y ago
I do, often. I keep my new tab page in all browsers blank for speed and minimal distraction, regularly hide all background apps (cmd-opt-H on Mac), close windows I'm not going to use for a while (especially anything that could have uns
727.
by hedgehog 3y ago
As we are under more pressure due to all the factors you pointed out the odds that we are able to avoid spread of a highly contagious virus with a higher fatality rate than COVID (closer to say MERS) or a series of big famines (say 1% of gl
728.
by hedgehog 3y ago
Depending what all the other effects you describe do to food production and communicable disease spread "we'll all be dead" may be only a moderate exaggeration of the effect over say a century or two.
729.
by hedgehog 3y ago
I do, and about 20 other projects since. In my experience far from a solved problem. I mean, I remember IE3 too, how many times are we going to recreate the web browser?
730.
by hedgehog 3y ago
The financial sector is the clear winner in who makes worse user-facing software though. I've interacted with a whole variety of US government services from federal to city and they're on average better than the garbage I have to
731.
by hedgehog 3y ago
This looks pretty nice. I use Notion at work and mostly like it but not for personal data because I want everything available to back up on a physical disk. I've been burned too many times with data getting stuck in an inaccessible for
732.
by hedgehog 3y ago
It's funny, though I didn't really know him I did have two chance interactions 15 or so years back that are in a way core lessons for me about business. Back when he started doing consulting I ended up spending some hours on the p
733.
by hedgehog 3y ago
There's an extended discussion about the relationship and differences between the two here: https://minilanguage.medium.com/mini-the-minimal-language-3f... "The spark that led me to create Mini was realizing that
734.
by hedgehog 3y ago
Maybe this was the inspiration for your project but I just saw a paper that applied compressors as part of a text classification system getting accuracy competitive with BERT. 14 lines of Python. https://aclanthology.org/202
735.
by hedgehog 3y ago
For what alternative? My 14 Pro is way too big, and the camera software is meh, but the lower models have really nerfed cameras and I'm too entangled in the Apple ecosystem to switch to Android even if there were better options. 80% sa
736.
by hedgehog 3y ago
I think the DMCA being a massive overreach is a separate issue from whether weights should be eligible for copyright. This is a complicated legal area and I'm very much not a lawyer so let me just stick to some examples that guide my t
737.
by hedgehog 3y ago
Like the parallel scenarios of taking a book and changing a few words, slapping a new logo on someone else's app, or stylizing a photo with a filter, those are questions that will be answered in court if people can't come to an ag
738.
by hedgehog 3y ago
I understand that LLMs to date have mostly been trained on a wide variety of copyright-encumbered data but in other domains (computer vision for example) the tradeoffs are different and in practice many models are trained on private /
739.
by hedgehog 3y ago
It of course depends on workload but I know more than a little about this specific problem space and reducing the space+energy cost of the multipliers is useful. If the idea proposed worked well enough it might be a useful block in a camera
740.
by hedgehog 3y ago
15+ and there have been a few but nothing particularly notable. Gmail has had outages too, and I couldn't tell you from personal experience which is more reliable which is interesting given the big difference in the complexity of the d
741.
by hedgehog 3y ago
I checked with them on an uncommon configuration (combo hydronic heat and hot water for the taps to replace the two gas units we have now) and their estimate was a lot of money but they were also the only company that seemed to have any ide
742.
by hedgehog 3y ago
Getting anything at all done in the Seattle area is super expensive. Not sure why. The rep is wrong about the too cold bit though, even just what's in stock at Lowe's is rated to well below record lows for the region.
743.
by hedgehog 3y ago
That, and the idea that ordinary individuals have much closer reach to a corporate marketing department than before. It all seems sort of boring and obvious now but remember in the 90s online shopping was mostly the digital equivalent to ma
744.
by hedgehog 3y ago
There is a fair amount of existing work related to planning in continuous and unbounded spaces under uncertainty. It seems likely some of the existing techniques combined with modern language models could be quite effective. A couple entry
745.
by hedgehog 3y ago
Very nice engineering.
746.
by hedgehog 3y ago
> That's not a simple answer but why would it depend on whose chips you have? How do you estimate how much life is left on any given part of the physical flash? This is one of those areas where neglecting the optimizations qualitati
747.
by hedgehog 3y ago
Wear leveling: Just search Google Scholar for "wear leveling" to get an idea for how deep that rabbit hole goes. At minimum think about the difference between placing an infrequently written logical sector vs a frequently written
748.
by hedgehog 3y ago
Besides pruning (which is what you describe) you might find the "lottery ticket hypothesis" interesting. The idea is essentially that in a randomly initialized large network there is a subset of weights that are already closeish f
749.
by hedgehog 3y ago
My guess is because flash chips are quirky and not very interchangeable. The error correction, wear leveling, etc all probably needs to be configured and calibrated to the specific chips used in a way that only can really be done by the man
750.
by hedgehog 3y ago
To reinforce, this is harder to achieve than it might seem. Sparking in switches and buttons and electrostatic discharge both are kind of invisible in normal environments but can be disaster on an oil rig or in an area of a hospital where p

More ›