lxe

13 ms·

181.
by lxe 2y ago
I've been a fan of the Author's other theme: Orchis https://github.com/vinceliuice/Orchis-theme
182.
by lxe 2y ago
Unless I can right click (or long press) and copy or save and then paste somewhere else, I don't want it.
183.
by lxe 2y ago
If you give a blind researcher this task, they might have trouble seeing the gorillas as well. Also the prompt matters. To a human, literally everything they see and experience is "the prompt", so to speak. A constant barrage of i
184.
by lxe 2y ago
The app is actually really good too.
185.
by lxe 2y ago
The chat UI's content_filter is not something the model responds with. Once the content_filter end even is sent from the server, it stops generation and modifies the UI state bailing out. You can probably use the API to bypass this fea
186.
by lxe 2y ago
You can also intercept the xhr response which would still stop generation, but the UI won't update, revelaing the thoughts that lead to the content filter: const filter = t => t?.split('\n').filter(l => !l.include
187.
by lxe 2y ago
Seems to be significantly slower than on the official deepseek API. But at least it's the full model!
188.
by lxe 2y ago
Once again, DeepSeek-R1-Distill-Llama-8B, is not DeepSeek-R1
189.
by lxe 2y ago
Both ollama and groq are spreading this misinfo for some reason.
190.
by lxe 2y ago
Pretty impressive TPS numbers for CPU-only
191.
by lxe 2y ago
These mini models are NOT the DeepSeek R1. > DeepSeek-R1-Distill models are fine-tuned based on open-source models, using samples generated by DeepSeek-R1. The amount of confusion on the internet because of this seems surprisingly high.
192.
by lxe 2y ago
I mean, almost ALL opensource models, ever since alpaca, contain a ton of synthetic data produced via ChatGPT in their finetuning or training datasets. It's not a surprise to anyone who's been using OSS LLMs for a while: almost AL
193.
by lxe 2y ago
I'm developing an intuition to how and what to ask in order for the LLM's answer to be helpful. Once you start spinning your wheels, clear context, copy what you need, and start over.
194.
by lxe 2y ago
This is one of those special HN posts that demonstrates outsized excellence on the author's behalf. Watched the video and I'm very impressed.
195.
by lxe 2y ago
Before pervasive GPS, it took me very little time to actually learn and internalize a route. Now it takes a lot longer to remember it when you're constantly guided. Same exact thing is happening with guided reasoning we get with LLMs
196.
by lxe 2y ago
Not sure why so many top comments dismiss the severity of this. This is just exactly the type of attack that give law enforcement or a malicious actor a way to establish proof of whereabouts.
197.
by lxe 2y ago
Wow! Nokia understood exactly what was happening and what needed to be done but failed to execute.
198.
by lxe 2y ago
Not just same bands, but same noise and artifacts too. They copypasted the data?
199.
by lxe 2y ago
I think most of the instruction fine-tuning methods for oss models stem from Alpaca, so it should be included: https://crfm.stanford.edu/2023/03/13/alpaca.html And the one referenced in there on synthetic da
200.
by lxe 2y ago
Instead of crossing your eyes and attempting to focus, what can really help relax your ocular muscles is to do the opposite: look "past" the images into the distance until the images overlap.
201.
by lxe 2y ago
I was about to post this same exact post :) Was the high score holder on there for a few years.
202.
by lxe 2y ago
> I don’t have my email on my gaming PC, nor do I have it on my work laptops. What? You have your email on literally every device -- be honest.
203.
by lxe 2y ago
Looks like NFT nonsense but without the blockchain
204.
by lxe 2y ago
This slowly increases your blood pressure and heart rate.
205.
by lxe 2y ago
You have to check out the author's other youtube videos exploring this tower in depth. Fascinating stuff! https://www.youtube.com/watch?v=6_u8x8V4YYs https://www.youtube.com/watch?v=wJ-XqQchwEw
206.
by lxe 2y ago
This reads like a bit from Silicon Valley the TV series... is this parody?
207.
by lxe 2y ago
Been on Iosevka for ages. Nothing else comes close for me. Not even berkeley condensed
208.
by lxe 2y ago
A used 3090 is $600-900, performs better than 7900, and is much more versatile because CUDA
209.
by lxe 2y ago
Wake words are different from "listen to everyhing until name is called". A wake work is needed for both privacy and technical reasons -- you can't just have alexa beaming everything it hears to amazon. So instead it uses a l
210.
by lxe 2y ago
Here's what I'm looking for in a voice assistant: - Full privacy: nothing goes to the "cloud" - Non-shitty microphones and processing: i want to be able to be heard without having to yell, repeat, or correct - No wake wo

More ›