nodja

10 ms·

91.
by nodja 1y ago
> As things have shifted more towards mass consumption of model weights it's become less and less common to see. Not the real reason. The real reason is that training has moved to FP/BF16 over the years as NVIDIA made that more
92.
by nodja 1y ago
The wear on the parity drive is the same regardless of raid technology you choose, unraid just lets you have mismatched data drives. In fact you could argue that unraid is healthier for the drives since a write doesn't trigger a write
93.
by nodja 1y ago
If what you said were true then they would ban all porn and not just rape/incest/bestiality porn. They're banning specific genres of porn which makes it an obvious morality issue. I can't back this up with facts but the
94.
by nodja 1y ago
I'll take the bait. > I think the bulk of this is kind of a silly semantic argument > "After all, I didn't click 'Rent now'. I paid 60 bucks and clicked a button that said 'Buy'" > With all t
95.
by nodja 1y ago
I've found that the best experience of self hosting excalidraw is actually using it inside nextcloud, it's called whiteboard over there but it's actually excalidraw. Setup is bit finicky but workable if you understand how rev
96.
by nodja 1y ago
The title is just fearmongering, it's removing the driver from being automatically installed from windows update, not preventing it from being installed altogether. They're also not revoking the signatures either so downloading an
97.
by nodja 1y ago
> Apple Announces Foundation Models and Containerization frameworks, etc. This sounds like apple announced 2 things, AI models and container related stuff I'd change it to something like: > Apple Announces Foundation Models, Cont
98.
by nodja 1y ago
My intuition is that the harder it is for an LLM to do something during training the more actual compression/learning will be encoded in it's weights. With multi-token/diffusion it becomes much easier to "reward/los
99.
by nodja 1y ago
This is insanely fast, my guess is that the tradeoff here is that the GPUs will always be working at max capacity and there will be minimal compute savings from batching, which I realize now is not really a tradeoff. My only worry is that t
100.
by nodja 1y ago
2m27s for a partial response editing a 178 line file (it failed with an error, which seems to happen a lot with claude, but that's another issue). https://streamable.com/rqlr84
101.
by nodja 1y ago
Here's a video of what it looks like with sonnet 3.7. https://streamable.com/rqlr84 The claude and gemini models tend to be the slowest (yes, including flash). 4o is currently the fastest but still not great.
102.
by nodja 1y ago
I wish they optimized things before adding more crap that will slow things down even more. The only thing that's fast with copilot is the autocomplete, it sometimes takes several minutes to make edits on a 100 line file regardless of t
103.
by nodja 1y ago
It's more than that, or in this case, less. This is a build of windows targeting long term deployments and embedded system, so it's essentially a pared down version of windows to reduce memory requirements. One of those things is
104.
by nodja 1y ago
The store is also an app on windows and is sometimes an hard dependency to install apps that only exist on the windows store without having to jump through many hoops. It's usually part of windows itself in the regular retail builds of
105.
by nodja 1y ago
Dataset. To train these models you need inputs and expected output. For text-image pairs there exists vast amounts of data (in the billions). The models are trained on text + noise to output a denoised image. The dataset of sketch-image pai
106.
by nodja 2y ago
My understanding is that frontier researchers will work for companies that will let them publish papers and discuss them with their peers. When you're an engineer at the tier of these AI researchers, winning an extra 100k/year on
107.
by nodja 2y ago
I highly suspect that CoT tokens are at least partially working as register tokens. Have these big LLM trainers tried replacing CoT with a similar amount of register tokens and see if the improvements are similar?
108.
by nodja 2y ago
They both work on a sorted list of tokens by probability. top_k selects a fixed amount of tokens, top_p selects the top tokens until the sum of probabilities passes the threshold p. So for example if the top 2 tokens have a .5 and .4 probab
109.
by nodja 2y ago
It's open source. https://github.com/Lumon-Industries/Macrodata-Refinement
110.
by nodja 2y ago
While 192GB of ram is appealing, it's also quite expensive at $6000. For that price I rather buy a system with 5 used 3090s, which while being "only" 120GB of VRAM, you benefit from much faster tokens/s and prompt proces
111.
by nodja 2y ago
From my understanding this not only removes tokenization but also sampling correct? Sampling can be a pain point of LLMs, but they also can enable interesting usages, like forcing grammar so the model always outputs valid JSON or tuning tem
112.
by nodja 2y ago
I'm portuguese and there's an oddity in either this chart or my memory. When we transitioned to the euro it seemed most shops straight up converted from escudo to eurocent. So if something cost 50 escudos it would cost 50 cents. I
113.
by nodja 2y ago
That's part of it. Normally when you visit contentsite.com which serves ads from adsite.com. Adblocker rules can just block adsite.com and the ads won't be shown. CNAME cloaking would have the main site have a subdomain like adsit
114.
by nodja 2y ago
Gamemaker if you want no code. It's free if you're not planning to release games commercially. You can add code later if you want custom behaviors that gamemaker doesn't include out of the box. 2D only. Godot if you want codi
115.
by nodja 2y ago
It will be 100+ fps on a current mid-tier system easily, because this is not hardware emulation. Most emulators need to implement/compile a whole different architecture to x86, even if similar to x86 it causes issues because cache gets
116.
by nodja 2y ago
If you don't know what you're doing, on ios you need a separate video with instructions (or fumble around) to figure out how to install a third party app, while on android the prompts guide you. On android when you download an apk
117.
by nodja 2y ago
Somewhat inefficient for text, very inefficient for images, specially if you work in pixel space. The max context a model today has been trained is 1M tokens, which takes up a lot of memory. Even if context was not an issue, to generate a 1
118.
by nodja 2y ago
>Intel confirms oxidation and excessive voltage in 13th and 14th Gen CPUs Title is ambiguous, it seems to imply 14th gen CPUs have oxidation issues, the video explains that the oxidation issues are only on an early run of 13th gen. If yo
119.
by nodja 2y ago
Chat GPT4 has been in general both better and worse than 3.5. I sometimes start a conversation with GPT4 and then have 3.5 fix it in another window (by pasting the code and pretending it's mine). It feels like things have degraded for
120.
by nodja 3y ago
A DHT is a decentralized key-value database, it's most famous use being in the bittorrent protocols, it uses a routing algorithm to guarantee that you can find the peers that can retrieve the value of a known key, granted that you at l

More ›