13

Source document

[2025-10-14 13:29] _slava_cm: generated pretty detailed personality prompts, which topics it's supposed to cover eg, tell me what I should buy if I am (customer profile) or compare these products, build me a bundle for (specific purpose) blah blah [2025-10-14 13:31] _slava_cm: Just figured out how I'll implement it actually, I'll combine it with direct inbox into one floating button and from there a little popup sidebar will open up with option for direct chat or AI assistance, important to have both because inbox has to sync with WhatsApp, messenger, provide order tracking capabilities etc. so I'll have both options available under one button. Just have to code that part but should be able to Vibe it [2025-10-14 13:35] _slava_cm: Overlapping issues has to be really minimal, content variety is pretty large so has to not impede anything the customer sees throughout the entire journey or on any page [2025-10-14 13:36] _slava_cm: or could just have it only visible on home page and product/collection pages, gotta figure out how to code that though still noob [2025-10-14 13:49] ajaxdavis: nice ux solve [2025-10-14 13:50] ajaxdavis: would take quite a bit of effort to reimplement both solutions into one [2025-10-14 13:51] ajaxdavis: must be pretty powerful with all the product manuals in there [2025-10-14 16:54] alluring_piglet_29962: Everything I read about this suggests it's pretty underwhelming. Less performant for local inference than strix halo boxes which are half the price. DGX Sparks can apparently be interconnected together in a way that strix halo can't, and obviously it's CUDA, so don't have to deal with the pain of ROCm. [2025-10-14 16:54] alluring_piglet_29962: Even so, I think in a couple of years there will be much better options available for local inference. [2025-10-14 17:42] foxhop.: Yeah 6-7k CUDA cores each. [2025-10-14 17:43] foxhop.: 10k is nice but these are just shy of that. [2025-10-14 17:43] foxhop.: with 128G vRAM. [2025-10-14 20:14] alluring_piglet_29962: https://old.reddit.com/r/LocalLLaMA/comments/1o6izz2/dgx_spark_vs_ai_max_395/ [2025-10-14 23:08] ajaxdavis: is local inference all that important or is it more important that you just are in control of the model etc (hosted gpus) because i feel like you can do the later, and all you need is an internet connection but you don't have the requirement of carrying around gpu/tpus [2025-10-14 23:29] ajaxdavis: https://wispbit.com/ <@1211062099137265723> [2025-10-14 23:42] ajaxdavis: https://accent-explorer.boldvoice.com/ [2025-10-15 00:17] alluring_piglet_29962: I'm not going to argue that local inference is rational. If I was doing something in anger I would likely end up calling OpenAI I assume. [2025-10-15 00:17] alluring_piglet_29962: That being said, I think rationality should be my servant and not my master. [2025-10-15 01:32] ajaxdavis: https://supermemory.ai/ cool idea for an ai startup, wouldn't mind doing something like that myself...