petu

7 ms·

61.
by petu 3mo ago
It probably would be called 'deepseek-v4-flash-0731' in API edit: nope, at least deepseek kept "deepseek-v4-flash" and just updated model underneath. I guess preview is no longer worth serving with that release and you&#
62.
by petu 3mo ago
Maybe .sexy TLD owner decided that "onlyfans" is a "premium" domain, with special pricing. And registrar isn't aware of premium domain pricing (at least for that TLD).
63.
by petu 3mo ago
Unused RAM is wasted RAM. So not really Apple magic, about every OS uses "free" memory as disk cache. Try to leave only a gigabyte or two free, speed likely would drop dramatically. Edit: or do some calculation / logging of e
64.
by petu 3mo ago
You don't get to take distilled model home, it all stays with Google. It's "optimize your costs in our garden" product.
65.
by petu 3mo ago
LLMs can't reliably answer what they are (w/o getting that info from system prompt/tool call/etc), so yea. https://xcancel.com/teortaxesTex/status/2026130112685416881 I think I've seen sam
66.
by petu 3mo ago
They're not saying anything about Anthropic serving costs in that quote, just calculating what MTok price is for running agents. Next sentence after your quote: > Agentic workloads have extremely high input-to-output ratios (our Cla
67.
by petu 3mo ago
I think internal use is allowed at any scale in the license? > 4. The requirements set forth in Sections 2 and 3 do not apply to: (a) internal use of the Software, defined as any use that does not make the Software, its outputs, or its u
68.
by petu 3mo ago
You need whole weights in VRAM for optimal performance. Don't be confused by "experts" in the name -- you don't get to load static subset of experts and blast next 100 tokens with them. In typical MoE model they get swit
69.
by petu 3mo ago
First sentence is about token efficiency.
70.
by petu 3mo ago
I think in case of Firefox it's done purely with JS ( https://mozilla.github.io/pdf.js/ ), no native code. So that feels fine (assuming it runs in same sandbox like any page), just bundled HTML/JS being changed
71.
by petu 3mo ago
To be fair Deepseek V4 now is just a preview, which likely implies training isn't completed.
72.
by petu 3mo ago
Deepseek V4 paper is just ~three months old
73.
by petu 3mo ago
I've misread that you're having consistent problems across different models, sorry. If it's just 10 Pro, then google says Qualcomm was having bad years (I've heard about Snapdragon 888 fiasco, but apparently it extended
74.
by petu 3mo ago
It's also 2.8x parameter count (1T -> 2.8T), likely higher activation per token (50B?).
75.
by petu 3mo ago
Nowhere close to what you asked, but Nothing already seems to be one of the most "open" devices today. That's for simple fact of having offline/unsigned EDL (Qualcomm's Emergency DownLoad) and files for it getting c
76.
by petu 3mo ago
> it seems that it still runs hot according to some reviews I've seen on the yt. Never noticed it being even warm in normal use, consistently cold. > this is especially true during hot summers Sounds like not a phone problem -- v
77.
by petu 3mo ago
As other commenter said OnePlus is subsidiary of Oppo. Over past few years they were loosing autonomy / reusing more and more work by Oppo. If you're worried about the firmware, then current day OxygenOS is just rebadged ColorOS.
78.
by petu 3mo ago
Most people use official clients and AFAIK they don't really accept any external contributions? All clients were fully developed by employees.
79.
by petu 3mo ago
packing multiple trits together e.g. 5 trits (243 states) into a byte gives 1.6 bits per trit: https://compilade.net/blog/ternary-packing
80.
by petu 3mo ago
Human could write that many comments to get enough base64 text for a design. Maybe to even get some of the highlighted characters in places they want (roughly equally spaced apart).
81.
by petu 3mo ago
Indeed, I had a brain fart. Thank you. But doesn't that make situation even worse? They likely would be 100B+ market cap/total asset value if spun off into independent company, but not able to generate even 1B in annual profit? Ac
82.
by petu 3mo ago
While it can be disregarded just as major update CS2 is formally 2023. Dota is live service game, getting major patches each year. Patch notes for version released 3 months ago: https://liquipedia.net/dota2/Version_7.41
83.
by petu 3mo ago
3% is lower than buying US Treasury bonds? Effectively zero risk zero effort return vs running a business. https://home.treasury.gov/resource-center/data-chart-center/...
84.
by petu 4mo ago
Nvidia reuses numbers for workstation cards, so there's multiple vastly different 'RTX 6000' cards: Quadro RTX 6000 (Turing / 2080 Ti / 24GB @ 672.0 GB/s) RTX A6000 (Ampere / 3090 / 48GB @ 768.0
85.
by petu 4mo ago
Maybe I didn't put it clear enough: __just the first__ GrapheneOS requirement leaves us with three brands. There's 20+ more items on the list after that, Nothing and Fairphone likely fail on them.
86.
by petu 4mo ago
Pixels are consistently "third party Android builds friendly", plus GrapheneOS has a list of required security features (beyond their control): https://grapheneos.org/faq#future-devices e.g. first one in the list:
87.
by petu 4mo ago
How's that relevant? There's no Steam subscription of any kind to access your library. Whether you're spending $0 on average year, or $300 -- you get to access your Steam library the same.
88.
by petu 4mo ago
There's talks about HBF. F for Flash -- HBM packaging and bus width, but using NAND memory. e.g. https://www.sandisk.com/company/newsroom/press-releases/2026...
89.
by petu 4mo ago
That's attribution to Qwen team. There (is/was) no attribution to Nex team (they've released a model based on Qwen 3.5 397B as well). As per OP link Nex claims that what Rio team released (so far) is just linear interpolation
90.
by petu 4mo ago
You're kinda talking about Claude being used for planning/architect role, while local LLM is just executing it (performing edits) -- at least in such form it's a thing, yes.

More ›