idonotknowwhy » user profile

Member Since 02/11/2016
Last Seen Hidden
Badges 21 4

Recent Activities

Can you run your own vibecoded games on these without writing them for GBA etc?
16/09/2026 - 00:58
>Note that if your agent is Gemini then any issue will be automatically closed without a response. LMAO why?
01/09/2026 - 19:28
They were going to up the price by 30c soon anyway. At least the scheduled price jump to $6.70 in 3 months won't have any extra card fees.
22/08/2026 - 12:32
How was this ever $999? Nobody in their right mind would have paid that.
09/08/2026 - 23:23
>Really? Which part? Token efficiency differences? Yes, that's true and has been pretty much since GPT models first came out. Tokenizers…
04/08/2026 - 01:05
Most people use the subsidized coding plans, probably US$20 / month. And token efficiency is far greater with frontier models.
03/08/2026 - 22:47
Yeah but there are clearly people who want to run K3 locally, even at 0.3 t/s I saw people building raid-0 of PCIe5.0 drives to do this at…
03/08/2026 - 13:27
One finetuned, they're better at summarizing reddit/ozbargain discussions.
03/08/2026 - 10:47
You're not fitting an unlobotomized 14B in this landfill card lol
02/08/2026 - 00:58
Clicked thinking it was a NAS :(
02/08/2026 - 00:53
Wait, this is a thing that bothers other people? I thought it was just me lol. Never had a problem with them until maybe 1-2 years ago, now…
16/07/2026 - 23:02
"You can't order his magical AM5 unicorn from Dell or Lenovo. They sell corporate beige obedience boxes, not RGB gamer graves. Order him a…
13/07/2026 - 14:47
https://m.media-amazon.com/images/I/91nXMYOKy7L._AC_SL1500_.jpg That looks nothing like Halo to me. Is it supposed to be the silent…
06/07/2026 - 22:46
The cables aren't interchangeable with Type-4 though right? Seems I've trapped myself in an "ecosystem" with all my Corsair power…
21/06/2026 - 14:22
> people are making the MI50s work They "work", but it's a real PITA. And prompt ingestion is incredibly slow, they're not really viable…
01/06/2026 - 15:19
Yes, but your only option is llama.cpp and LLMs. You'll be in for a world of pain if you try to run diffusion models, finetuning, etc.
01/06/2026 - 15:17
https://files.catbox.moe/292igw.png
27/05/2026 - 09:35
You're absolutely right to call him out!
26/05/2026 - 02:57
>has a spare M2 slot So one could add an m2->OCuLink connect an eGPU?
12/05/2026 - 12:32
Remember LLMs hallucinate, so always.... ask another LLM to critique the first one! From Gemma-4: > Copilot is legally regarded. A…
30/04/2026 - 08:57
>99.9999% of Mac users Six significant figures for 12 people. Is that like someone's left pinky finger hovering over the Terminal icon?
19/04/2026 - 20:18
I would not be surprised; the observable universe is Turing-computable and a Matrioshka-brain-scale AI could run a lattice-QCD-level…
12/04/2026 - 22:50
Oh good, because I doubt any of those hallucinated keys would actually work lol
16/02/2026 - 19:23
>The issue is many people are mistakenly assuming it is 100% correct. >To make matters worse that wrong info gets repeated online as fact…
16/02/2026 - 11:46
>Logic says It will become sentient Not going to happen (at least with any of the current technology) > it's only a matter of time until it…
16/02/2026 - 11:22
Loading...