hedgehog

25 ms·

451.
by hedgehog 2y ago
Neither case is easy. Making something "simple" like a new control arm or connecting rod is way beyond a typical garage shop. The electronics are not a home shop item either but replacing parts with work-alike modules is already
452.
by hedgehog 2y ago
It seems `map` should have less restrictive semantics (specifically ordering) than `for`, does that allow more optimization? I don't know much about Rust internals.
453.
by hedgehog 2y ago
Speaking only for myself these ideas are a combination of things I've seen scanning new papers and informal discussions with other people working in the area. Feel free to shoot me an e-mail though, maybe I can point you somewhere more
454.
by hedgehog 2y ago
As other people pointed out here you can also add "verbosity sinks" as text fields in structured output, recently I've also been experimenting with tool calls to support guided self-talk in a way that doesn't necessarily
455.
by hedgehog 2y ago
I prefer to drive smaller cars but with stroller, kids' car seats, etc there are some physical constraints on getting everything to fit in a reasonable way.
456.
by hedgehog 2y ago
Looking at the current Mazda lineup the replacement for the Mazda6 (29 mpg) is the CX-50 which adds AWD and gets 28 mpg or 38 mpg for the hybrid. I was suprised the Mazda6 would have so much more cargo area but in the spec sheet it lists 22
457.
by hedgehog 2y ago
Crossovers in that zone are a pretty reasonable compromise of cost, handling, safety, and ergonomics. They have essentially replaced station wagons and are mostly built the same way, just a little taller.
458.
by hedgehog 2y ago
Let's not underestimate the complexity of the problem Gelsinger took on. I base that on the experience of selling them a company, and then having a little bit of a view from inside (under Bob Swan). It's clear to me that Gelsinger
459.
by hedgehog 2y ago
If you add AMD (design) + TSMC (fab) it's not that far off from Intel (who does both, and sells >3x AMD by units). Clearly there are struggles at Intel but the simple math doesn't explain it.
460.
by hedgehog 2y ago
I think basically only NVIDIA is reliably supported right now, it would be nice to have more hardware support to allow splitting models (like HF Accelerate or llama.cpp support).
461.
by hedgehog 2y ago
Nice. Unsloth requires a platform supported by Triton, correct?
462.
by hedgehog 2y ago
I don't think you're wrong but the overarching structure of the chip business is very different from times gone by. It's not even clear what "technical leadership" should mean. When Intel was the leading edge volume
463.
by hedgehog 2y ago
Open Web UI [1] with Ollama and models like the smaller Llama, Qwen, or Granite series can work pretty well even with CPU or a small GPU. Don't expect them to contain facts (IMO not a good approach even for the largest models) but they
464.
by hedgehog 2y ago
The special thing about ITC is you can do the splitting (to add a member) and joining (to remove one) without any external coordination. They otherwise are used in the same way as vector clocks. I find the linked post more confusing than I
465.
by hedgehog 2y ago
Multi-backend Keras was great the first time around and it might be a more widely used API today if the TF team hadn't pulled that support and folded Keras into TF. I'm sure they had their reasons but I suspect that decision direc
466.
by hedgehog 2y ago
inglor is probably referring to Strassen or Coppersmith–Winograd.
467.
by hedgehog 2y ago
There is a contingent that I think Marcus is responding to that have been claiming that all we need to get to AGI or ASI is pure transformer scaling, and that we were very close with only maybe $10B or $100B more investment to get there. If
468.
by hedgehog 2y ago
Are you aware of any efforts to do this? Even a 3B param attempt would be informative.
469.
by hedgehog 2y ago
Compare 2022 to 2020 or 2017 though.
470.
by hedgehog 2y ago
The context some commenters here seem to be missing is that Marcus is arguing that spending another $100B on pure scaling (more params, more data, more compute) is unlikely to repeat the qualitatively massive improvement we saw between say
471.
by hedgehog 2y ago
Yeah, the main issue was the lack of memory protection on Mac so the crashes could take down the whole machine. I switched to NT4 though that had drawbacks of its own.
472.
by hedgehog 2y ago
I don't remember Glider crashing at all but running anything really big and complicated like Netscape it was only a matter of time.
473.
by hedgehog 2y ago
Pot in a pot also minimizes cleanup, I'm a big fan but few people seem to use that method.
474.
by hedgehog 2y ago
Arsenic content is also highly variable depending on where the rice was grown.
475.
by hedgehog 2y ago
Rossmann is like Scotty Kilmer, the automotive guy. Lots of clickbaity breathless videos, highly variable technical accuracy.
476.
by hedgehog 2y ago
Aha, so not much need. I've always avoided SQLite for larger databases due to the extra space needed to allow compaction, maybe not a real problem in most applications though.
477.
by hedgehog 2y ago
Oh, I mean like with the vacuum command. As the databases get larger it can become unwieldy.
478.
by hedgehog 2y ago
What do you do about compaction?
479.
by hedgehog 2y ago
I'm not sure I've met anyone with significant application security experience who would claim writing correct C is anything less than very challenging. It's not just memory safety, even basic features like iterable collection
480.
by hedgehog 2y ago
I manage that by using a vacuum to clean my lint filter but folding seems to release a lot of dust so I do that next to an air filter.

More ›