omneity

21 ms·

361.
by omneity 3y ago
There is a significant difference. Video has a temporal component (frames tend to be correlated with previous ones), and vision LLMs do not have some sort of hidden states to keep track of that temporal component. Using captions to bridge t
362.
by omneity 3y ago
If keeping up with Google dev blogs is important to you, at Monitoro[0] we support alerts even if the website doesn't offer RSS. You can also catch specific updates, for example if a new post mentions a Google tool you're using at
363.
by omneity 3y ago
In terms of “airdroppability to disaster zones”, hardly anything beats cardboard beds[0]. It is true that the field hospital in TFA is a much comprehensive solution. I just wonder what an equivalent would be if designed with similar minimal
364.
by omneity 3y ago
How does this 2-bit quantization method compare to HQQ which was posted yesterday? https://news.ycombinator.com/item?id=38563537
365.
by omneity 3y ago
I hope for Godot to remain true to itself over time. Unity was the indie upstart at one point, just like Godot is now. Penpot/Figma story all over ... I hope not.
366.
by omneity 3y ago
There are minor differences, but the similarity is indeed striking.
367.
by omneity 3y ago
You should add attribution on the poster to GP as it is an original creation of theirs and their kid.
368.
by omneity 3y ago
Thanks a lot! I will reach out to you soon via the email on your profile.
369.
by omneity 3y ago
Oh wow. Could you talk more about your use of all these GPUs? And how you handled all these disparities for real usage?
370.
by omneity 3y ago
I’m working on something in this space that doesn’t exhibit the issues you mentioned, and it could potentially be useful to you. Can I get in touch with you for a demo and to get your feedback?
371.
by omneity 3y ago
I use Maroofy quite a lot and it was funny to discover many ripoffs and plagiarism between songs by complete chance.
372.
by omneity 3y ago
It seems to be written in 1952, reminiscing what it was like around 1900.
373.
by omneity 3y ago
Fascinating perspective. I wonder if such an argumentation could stand on legal grounds.
374.
by omneity 3y ago
This is excellent, I looked for something like this a few years back to explore what a minimalistic connected computing stack could be, and the only options were dead kickstarters or hacking something from blackberries or nokia c3s with hit
375.
by omneity 3y ago
Do you mean PID controllers?
376.
by omneity 3y ago
When I was a kid in school and got bored I would write html on a paper notebook. It got tedious quickly and sorta made me adopt a haml like notation for brevity. Around the same time I had a pocket pc that I would take to vacations. It had
377.
by omneity 3y ago
I feel like a modern version of this with a small LLM and a vectorized wikipedia, maybe even TTS, would make for a very interesting offline device. Humanity’s knowledge in a box, or as the Internet Box as the IT Crowd called it.
378.
by omneity 3y ago
txtai looks cool. Did you use it before in scenarios with all 4 components? I’m curious about which use cases it enables.
379.
by omneity 3y ago
This was a brilliant read for me back in the day (still is!) and gave me the inspiration to write a post[0] about detail abstraction, market forces and why you can't fit big cups in coffee machines. [0]: https://omarkama.li&
380.
by omneity 3y ago
Look at that! This might be the missing piece. https://ai.meta.com/research/cicero/
381.
by omneity 3y ago
GPT-4 is unlikely to be 1.7T params. This is a number floating around in the internet with no justification. The largest US open model is Google’s Switch-C which is 1.6T and only because it is a Mixture of experts model, i.e. it is constitu
382.
by omneity 3y ago
What makes you think this technology is out of control? If anything a 1T param model would only run in like 5-10 computers in this world. As much control as it gets. By the way this is doomer rhetoric. Bashing scientific advancements as dan
383.
by omneity 3y ago
To take the analogy of a company, the problem here is that management is really bad. What you described is rather akin to hiring better workers, but we need better managers. Whether it’s a single or multiple models is more of an implementa
384.
by omneity 3y ago
This does work to a certain extent, but doesn't really converge for significantly more complex tasks. (Source: tried to make all sorts of agents work on complex problems in a divide and conquer fashion) They eventually ... lose the thr
385.
by omneity 3y ago
Nice eye! As for your last question yes that exists. There are two models from Meta that do exactly this, instruction based iteration on photos, Emu Edit[0], and videos, Emu Video[1]. There's also LLaVa-interactive[2] for photos where
386.
by omneity 3y ago
I haven’t found that option. I know it exists in Gpt4all though. Personally I use a locally served frontend to use ChatGPT via API.
387.
by omneity 3y ago
I really like LM Studio and had it open when I came across this post. LM Studio is an interesting mixture of: - A local model runtime - A model catalog - A UI to chat with the models easily - An openAI compatible API And it has several plug
388.
by omneity 3y ago
Yeah, functional and/or declarative programming can be surprisingly easy for beginners. They can just describe what they want and then see the magic unfold.
389.
by omneity 3y ago
I'm working on making it possible to feed LLMs live data from websites. It is challenging because injecting webpages into LLM prompts directly has a mixed result, especially if there's a smidge of ambiguity or complexity in the we
390.
by omneity 3y ago
Congrats for launching Pickcode! The attention to detail is great and shows that you care. As an aside on the topic at large, I find there are (at least) two schools of thought in the visual programming space, when intending to make program

More ›