lxe

37 ms·

511.
by lxe 3y ago
It took me a while to grok the situation solely because of how bizarre the pushback from platform.io people is. I understand (but would still be vupset) if rpi org was disallowing this on the grounds of commercial reasons, not the other way
512.
by lxe 3y ago
No local inference support as of yet at least. Here are a few things we’d be excited to support… Local inference through WebGPU Local inference through stable-diffusion-webui Desktop installation Contr
513.
by lxe 3y ago
Parsing pdfs or web semantically is really not an easy job, as I found in my own foray into LLM sumamrization.
514.
by lxe 3y ago
Can I just have a newsletter to subscribe? Why do I need an app?
515.
by lxe 3y ago
Oooh. Are there any performance gains over using ggml itself?
516.
by lxe 3y ago
I've been working with langchain and llamaindex and did notice that it's a pretty hefty abstraction on top of pretty simple concepts and I also eventually ended up dropping both and simply write the underlying code without the fra
517.
by lxe 3y ago
Senior Hacker News Commenter
518.
by lxe 3y ago
Disappointing. Yet another industry leader sewing public FUD for some reason. Why not bring rational discourse into the conversation around software safety and ethics. Automation has been the driving force of industry since the industrial r
519.
by lxe 3y ago
A very natural step for Vercel. Not having a prescribed statefulness/storage solution has been a pain since they've pivoted into strictly serverless.
520.
by lxe 3y ago
Sites like these have been popping up everywhere through the years. Why not simply use Yelp or Google? What are the moderation and verification challenges? What about legal/defamation issues? How would one trust each review?
521.
by lxe 3y ago
The "this incident will be reported" message always struck me as having the same vibe as the "provided by the management for your protection" labels they have on toilet seat covers.
522.
by lxe 3y ago
What's the history behind this? I doubt there was legacy or backwards compat reasoning? Allowing for such a loose and wide interpretation makes for complicated parsing and numerous exploits.
523.
by lxe 3y ago
I made a thingy that receives pictures of the earth from a geostationary satellite: https://gist.github.com/lxe/c1756ca659c3b78414149a3ea723eae2
524.
by lxe 3y ago
This isn't me, this is ggerganov's work. His demo is using gpt2, and whisper models ( https://ggml.ggerganov.com/ ).
525.
by lxe 3y ago
SIMD support in wasm is a bit weird. You'll need to compile into multiple targets, and many compile flags are not supported: https://emscripten.org/docs/porting/simd.html
526.
by lxe 3y ago
I've noticed that the learning curve stary fairly flat when it comes to understanding weights, and layers, and neural networks, heck, even what gradient descent is for... but then when it comes to actually understanding why optimizatio
527.
by lxe 3y ago
This has Three-Body Problem vibes :)
528.
by lxe 3y ago
It's hard to death-hug these wasm demos, as they are all static files that can be easily served via CDNs
529.
by lxe 4y ago
Engineering curiosity and fun
530.
by lxe 4y ago
For "actually serverless" voice chat, check out https://whisper.ggerganov.com/
531.
by lxe 4y ago
I didn't read the article solely because of the headline. Any statement like this made about any new tech development is automatically false.
532.
by lxe 4y ago
Sounds like a CNS injury with symptoms that are pretty much mirror benzodiazepine and alcohol withdrawals. No matter how unusual or rare it is to experience this, it's really worth surfacing cases like these. I'd rather not take t
533.
by lxe 4y ago
This is cool, but why does it require a login? I would guess it uses an LLM under the hood with a single-shot prompt to refine a given prompt.
534.
by lxe 4y ago
1. Use more whitespace 2. Go to https://dribbble.com/ and look for inspiration 3. Read https://css-tricks.com/ and try to reproduce some of these things 4. Get into computer graphics in general. Mess around
535.
by lxe 4y ago
Was interested in reading this until got the Medium banner. We need to stop using Medium.
536.
by lxe 4y ago
This one is my favorite!
537.
by lxe 4y ago
Finally, an actual battery breakthrough! Most "new battery" media has been mostly fluff for decades.
538.
by lxe 4y ago
Thanks. Appreciate the bug reports!
539.
by lxe 4y ago
Hey there pretty neat! Thanks for testing this out. I feel like the next logical step is to try to get it to work on a mobile device.
540.
by lxe 4y ago
Ship it. Can’t wait to let this thing loose via agent-based auto-gpt commit spree. In all seriousness, I do expect better results from a 1.3B parameter model. I’m guessing the quantization, the baseline model, or my fine tuning plays a role

More ›