PeterisP

68 ms·

811.
by PeterisP 3y ago
The way I see it, things like that are critical for long term capability maintenance; having or not having fabs for mass production of simpler chips is what determines whether your country will have a workforce which would make it possible
812.
by PeterisP 3y ago
It's not like there exists any small player who can even imagine building a fab like that. All the small players in semiconductors exist by doing some particular niche like design or making some tool, but all of them outsource the actu
813.
by PeterisP 3y ago
If right now we found a weird island with a population identical to that "common ancestor that goes way further back", the general population would label that ancestor as an ape - while it's not the same as any of the modern
814.
by PeterisP 3y ago
It's trickier than that because often those "wants" aren't discrete and undividable, and ranking A over B often doesn't imply that you'd accept getting zero of B until A is fully fulfilled, rather you are likel
815.
by PeterisP 3y ago
How does it work for larger meetings? At the start of Covid I had to check many options, and while for many use-cases Google Meet was most convenient, it started to work poorly if there were a bunch of people connected, so I used Google Mee
816.
by PeterisP 3y ago
For various reasons I have a bunch of different groups where I use different videocall software for regular meetings - Zoom, Jitsi, Teams, Skype, Google Meet and Webex. Out of all those, Jitsi is the only one where I can't rely on the
817.
by PeterisP 3y ago
It's not a problem that tourists arrive and depart by air as it still is very helpful if they can move within Italy by train.
818.
by PeterisP 3y ago
For a few decades already growth of consumption/GDP has been decoupled from growth of energy and resource usage. Roughly speaking, we consume more by consuming "better" things rather than "larger" things. A $1000 iP
819.
by PeterisP 3y ago
For floating point math, there's no distinction, as "parralel computation with unknown path to the solution" inherently implies "results will vary", as (a+b)+c != a+(b+c).
820.
by PeterisP 3y ago
It's not a code quality issue, there are ways to ensure determinism (sometimes you just need to set a flag), however, they are intentionally explicitly not used in order to gain performance.
821.
by PeterisP 3y ago
US academia isn't really funded by the foreign grad student tuitions. It's not even really funded by tuition, as tuition+fees is <30% of total US academia funding. Of course every organization wants to have more money and so th
822.
by PeterisP 3y ago
It would make all sense to fork and maintain the version before the license change, if it has all the key features you (and others) need.
823.
by PeterisP 3y ago
There are all kinds of scenarios (e.g. fraud) where a CC refund is warranted even if you shipped the product, so without further details it's quite plausible that Stripe should have forcefully refunded the payment even taking your proo
824.
by PeterisP 3y ago
You can diff indices but it doesn't make economic sense to do so. The whole point of using Bing as the base is so that you don't have to make and keep an index of everything yourself, so DDG wouldn't have another exhaustive i
825.
by PeterisP 3y ago
That's the whole point of the changes - we've switched from "spend current money to grow company size and earn more money tomorrow" mode to "that tomorrow has arrived, extract profits now by reducing expenses and ke
826.
by PeterisP 3y ago
They can do better, however, it takes time and effort to do that - they can and do augment Bing results with other data sources when needed, but up until recently there was no need to add anything extra with respect to Techdirt; it certai
827.
by PeterisP 3y ago
That paper kind of does the same thing that my comment above proposed, starting with as large dataset as they can get and then filtering it to extract a much smaller dataset focused on a specific task that still is larger than all of Engl
828.
by PeterisP 3y ago
The idea of decentralized hierarchical LLMs is interesting but your chosen example is not a good illustration as all three of these data sources are small and insufficient, any model trained solely on any of them will not be a good model f
829.
by PeterisP 3y ago
It's not as simple. For example, look at the distance sensors used for parking and the data coming from them over the CAN bus. That bus needs to be able to (at least indirectly) reach the brakes so that automatic emergency braking can
830.
by PeterisP 3y ago
With SAAS, if the software is barely usable but lacks competition, vendor gets paid even if they don't fix bugs or broken functionality. When paying up-front, there always is competition - your own old version; so the vendor has stro
831.
by PeterisP 3y ago
You can also find out that you love it despite the novels (or software or paintings or poems or whatever) not being interesting for almost anyone else or even being available to anyone, but as you don't need the money you can keep doin
832.
by PeterisP 3y ago
In current ML environment, there's no single thing as "GPU support" as nVidia and AMD are two very different (and unequal) worlds - if you want things to work on AMD, then you should look for things which are explicitly label
833.
by PeterisP 3y ago
For me, I quite often use sudo to run scripts that install or reconfigure something, and they obviously do need to execute other commands.
834.
by PeterisP 3y ago
A non-profit service should be expected to have better compliance than a service which has a strong monetary motivation to exploiting users data as much as possible and thus comply with privacy requirements as little as possible.
835.
by PeterisP 3y ago
Not only, it's also because of the square/cube law making it impossible to have a 50 ton flyer that flies in the same manner as a 5 kg flyer, you can't simply scale things up.
836.
by PeterisP 3y ago
The Irish DPC had all the ability to declare this as unacceptable (and start imposing penalties if it's not changed) much, much earlier, without waiting for German DPC to proceed through courts.
837.
by PeterisP 3y ago
What exactly do you mean by "embeddings" there? The way a transformer-based LLM works is that right after tokenization you get a "vector of embedding vectors" i.e. an embedding for each token, and that gets processed b
838.
by PeterisP 3y ago
If performance and computing cost wasn't an issue (which it is!) then you could just have a page of code that continuously runs the LLM in a loop, outputting both the actual output and also some certain "short term memory state&qu
839.
by PeterisP 3y ago
Tab-delimited "csv" formats are quite common (e.g. the CONLL format family for many natural language processing tasks) and also supported by common tools such as MS Excel for decades already.
840.
by PeterisP 3y ago
The question isn't whether you can view the existing web content on the Vision Pro, the question is whether all of the Vision Pro SDK, sensors, and 3D functionality is exposed to the browser to enable VR-enabled webapps for the genres

More ›