nullc

55 ms·

871.
by nullc 2y ago
I think the relevance of their transness is not very significant. The lesswrong apocalypse cult has been feeding people's mental illness for years. Their transness likely made them more outsiders to the cult proper, so e.g. they didn&
872.
by nullc 2y ago
I suspect you're getting fooled by terminology. When you see "rules protecting shareholder rights" when you should be reading in this context is "rules inviting massive lawsuits from plaintiff attornies nominally represe
873.
by nullc 2y ago
requires a backwards trained llm no? I don't think anyone has pretrained a remotely-close-to-SOTA sized backwards model.
874.
by nullc 2y ago
My favorite experience with the 70b distill was to ask it why communism consistently resulted in mass murder. It gave an immediate boilerplate response saying it doesn't and glorifying the Chinese communist party, then went into think
875.
by nullc 2y ago
How does that explain the same decisions being made by other vendors that don't sell headphones?
876.
by nullc 2y ago
> Nobody has said "well thank God my new MacBook has no ports" or "I'm glad my battery isn't removable". The support column for these decisions is empty. No, but they do say "thank god my device survive
877.
by nullc 2y ago
https://en.wikipedia.org/wiki/Hotelling%27s_law
878.
by nullc 2y ago
There is a big difference between being able to train on the reasoning vs just the answers, which they can't against o1 because it's hidden. There is also a huge difference between being able to train on the probabilities (distil
879.
by nullc 2y ago
Impersonating a federal officer or employee is a crime. 18 U.S. Code § 912 (technically you have to obtain a document, money, or other thing of value to run afoul of that particular law, but getting a reply to a successful spoof might quali
880.
by nullc 2y ago
You can't distill from GPT-4 because Open AI conceals the probabilities (and has for a couple years now-- since before gpt4), presumably to prevent that. You can fine tune against output though. I might guess that they used something
881.
by nullc 2y ago
If you include multimodal data then I think it's pretty obvious that training is compute limited. Also current SOTA models are good enough that you can generate endless training data by letting the model operate stuff like a C compiler
882.
by nullc 2y ago
I don't see how that follows. Making training more effective makes every unit of compute spent on training more valuable. This should increase demand unless we've reached a point where better models are not valuable. The openness
883.
by nullc 2y ago
I've almost always found keys on the keyboards to fail before the trackpoint. The one exception I've had is a bluetooth trackpoint keyboard that someone sat on. Go check out videos of new computers users from the 90s, mice absolu
884.
by nullc 2y ago
The 'pattern matching' happens at complex layer's of abstraction, constructed out of combinations of pattern matching at prior layers in the network. These models can and do work okay with variable names that have never occur
885.
by nullc 2y ago
It's particularly funny/annoying when they're convinced that the fact they got it from the "AI" makes it more likely to be correct than other commenters who actually know what the heck they're talking about. It
886.
by nullc 2y ago
One of the major approaches is to block open source models produced by parties that aren't in the business of restricting access to AI.
887.
by nullc 2y ago
Yes, the default workflow on the product is that all prints go via their cloud service. For the first year or two of the X1C's existence this was the only way to print, but they later introduced lan mode. Leading to obvious speculation
888.
by nullc 2y ago
That article is referring to conflicting controls when using their cloud stuff. In lan mode it doesn't use anything remote and works just fine completely isolated. > you can see their MQTT server got DDOSed by some faulty 3rd party
889.
by nullc 2y ago
They just introduced firmware updates via SD card in the most recent released version, prior to that you had to put the printer online and associate it with an account to get firmware updates. But yes today there is no need to use their clo
890.
by nullc 2y ago
That hasn't been true for years, the regular X1C has an officially supported lan mode and works fine without any of the cloud stuff. (I believe the smaller ones do too, but I haven't used them so I can't speak to them).
891.
by nullc 2y ago
Yes, "lan mode" is an officially supported advertised feature, where you can happily print on an isolated network. (though as of this morning it now sounds like they're backing off after public backlash)
892.
by nullc 2y ago
I think the fact that enough users are using LAN mode and firewalling the printer off that it's less useful as a CCP data theft vector is a more credible explanation. The new interface code, running on your host computer, is much harde
893.
by nullc 2y ago
Agreed. Though 'not precise' is a common complaint from people who haven't overcome the learning curve. My partner hated using my x61, complaining that it was hard to hit anything accurately. Her laptop had a failed battery
894.
by nullc 2y ago
It's a faster, more accurate, and more effective pointing device in a very tiny fraction of the space. Primary downside is that it has a steeper learning curve-- though a couple days of forced usage is enough to overcome it. Poorly im
895.
by nullc 2y ago
They have periodically had models without the trackpoint, that isn't new. Too bad they don't have the courage to bring back some models without the touch pad... it's inches of wasted space.
896.
by nullc 2y ago
The underlying paper is the opposite of convincing. Essentially they fit a logistic regression of climate measurements the amount of area burned by wild fires each year in the forested parts of northern California to try to express how much
897.
by nullc 2y ago
What evidence do you have that these fires have anything to do with climate change? They appear to be adequately explained by the known behavior of the region, and to the extent that they're not the radical increases in habitation and
898.
by nullc 2y ago
> Now you can build concrete houses that are earthquake-resistant (eg in Japan) but it's expensive. It's expensive here, but is it expensive in Japan? Here its' expensive because it requires extensive steelwork which tak
899.
by nullc 2y ago
> the insurance rates in Pacific Palisades or on the Florida coast would be so high that no one could afford to live there I'm not so sure. The Pacific Palisades have astronomical real estate prices. (actually costly property in Fl
900.
by nullc 2y ago
It's been a decade but I have previously fuzzed multiple linux file system implementations and was able panic the kernel. I would be somewhat surprised if none had code execution vulnerabilities at some point, but I can't think o

More ›