declaredapple

23 ms·

61.
by declaredapple 3y ago
Almost certainly not. It's absolutely just sampling with temperature or top_p/k, etc. Beam searches would be very expensive, I can't see them doing that for chatgpt which appears to be their "consumer product" and o
62.
by declaredapple 3y ago
You can but it'll never be the same as the base model. That said it appears they also released the base checkpoints that aren't fine-tuned for alignment
63.
by declaredapple 3y ago
Congrats on the launch and thanks for the contribution! This looks like it's on-par or better compared to mistral 7B 0.1 or is that 0.2? Are there plans for MoE or 70B models?
64.
by declaredapple 3y ago
> "don't sell F22s to anyone else" What they actually said (abbreviated) was anything with more then > "600 Gbyte/s [memory transfer speed]" or 4800TOPS Page 103 https://public-inspection.feder
65.
by declaredapple 3y ago
It's just frustrating for me to see people write rules, and then say "hey not like that!". It wastes everyone's time, the point of regulations is to draw the line and have no ambiguous meaning. It's like putting up
66.
by declaredapple 3y ago
I don't get it, the rules were "Do not cross this line", nvidia, being a business that makes products that comply with regulations, made a product that stopped just short of the line. It's like putting up a speed limit s
67.
by declaredapple 3y ago
> so presumably the intent was to stop export of GA100/GH100 silicon altogether They should have just written this in the regulations. Instead they're making nvidia waste resources making a product that complied with regulatio
68.
by declaredapple 3y ago
I actually have had this happen and it's infuriating. I've had queries of copy+pasted errors with zero results, but playing around with it a bit just to find a github result that was only like two words off. > it’s terrible but
69.
by declaredapple 3y ago
> No the hell not. It should do what I tell it to. That's what I said. "then evaluate the query anyway." I should have added "original" to that statement > If that is nothing, that's what it should show.
70.
by declaredapple 3y ago
I don't understand what you mean. I'm saying if you have a query that returns a similarity score, I don't want only results with > 0.1. I want all the results returned
71.
by declaredapple 3y ago
What are you talking about? Have you ever worked with full text search? I'm saying I don't like high cutoffs of similarity scores. I have no idea what you're talking about. Very very few queries should have _literally zero re
72.
by declaredapple 3y ago
> if you query is bad or too specific. Then it should suggest a better one and then evaluate the query anyway. > This is what any decent search engine should do -- return nothing WHY?! That's the opposite of it's job! I have
73.
by declaredapple 3y ago
> I assumed MOE was where each expert analyzed the problem in a different way Uh sorta but not like parent described at all. You have multiple "experts" and you have a routing layer(s) that decide which expert to send it to. U
74.
by declaredapple 3y ago
They have an esp expansion board but the default flipper is an stm microcontroller + CC1101, I was actually surprised it didn't use an esp32.
75.
by declaredapple 3y ago
I disagree mainly because google, aws, apple, etc. All have similar, or even more access to GPU compute and funding for it, and in google's case also has been one of the main research contributers, yet they still struggle to touch GPT4
76.
by declaredapple 3y ago
> What we're trying to find is a business that, for one reason or another -- it can be because it's the low-cost producer in some area, it can be because it has a natural franchise because of surface capabilities, it could be b
77.
by declaredapple 3y ago
I'm curious what the impact of this would be on non-flippers. The flipper is just some common off-the-shelf radios and microcontroller, albiet in a nice package. That's it. Like common radios found in your everyday stuff, like s
78.
by declaredapple 3y ago
> One thing you could do with MoE is giving each expert different subsets of the input tokens. Don't MoE's route tokens to experts after the attention step? That wouldn't solve the n^2 issue the attention step has. If you
79.
by declaredapple 3y ago
> ZeroWater makes a carafe that reduces total dissolved solids to 0 ppm, including PFAs and microplastics, or so they claim. ZeroWater uses an ion exchange filter - It's my understanding that many (most?) micro plastics are not diss
80.
by declaredapple 3y ago
Keep in mind that a lot of the microplastics come from "macroplastics" (?) that have broken down. There's few things that destroy them further. That and many of the effects may not occur until a sufficient amount of microplas
81.
by declaredapple 3y ago
is binwalk still used these days?
82.
by declaredapple 3y ago
> Nothing suspicious about that at ALL no sir nothing to see here. Yes, very suspicious that he didn't want to issue CVEs for checks notes Two DOS attacks that only apply to users that explicitly enabled experimental QUIC support
83.
by declaredapple 3y ago
> These include, at a bare minimum, adequate contacts with the forum for the defendant to reasonably anticipate being haled into court there. This case does not come close to clearing that threshold. Neither party is based in Texas and t
84.
by declaredapple 3y ago
> but don’t have the capital to manufacture it myself You're looking at 10k to file it, and then all it gives you is the ability to sue. If [insert company] violates it, you could easily be looking at 100k+ for litigation that you m
85.
by declaredapple 3y ago
I'm on board with this. It shouldn't exist for software, I'd be happy to see it go for hardware too. I'd be more on-board with the idea if non-software patents only had say a 5 year lifespan
86.
by declaredapple 3y ago
For security purposes you should of course assume they do indefinitely. In practice I'm pretty sure every US ISP does, at least on the timescale of 30d+. I don't think the US officially has a retention requirement but I think Cana
87.
by declaredapple 3y ago
> At that point it becomes "building value for the shareholder" and "minimizing risk for the shareholder". Generally people refer to this as Enshittification > if you ruin the experience for the user then you'
88.
by declaredapple 3y ago
> And most IPV4 addresses are vague and only give a coarse location of where you are That's not true, you can request/compel Frontier to tell you what specific customer that IP was assigned to at a given time. > and they&#x
89.
by declaredapple 3y ago
> There's also no ways to really make a file system faster when dealing with many small files and directories. Depends on the file/directory structure. Traditionally back in the CGI days URI's were links to files on a serv
90.
by declaredapple 3y ago
You either wouldn't be liable anyway (not responsible for what people use it for) or will still be held liable (took no measures to prevent malicious use).

More ›