Serg (@FraudGnosis) header
Serg (@FraudGnosis) profile picture

☀️🌊usmc vet/ using ai to help me create things in 3D both digital and physical / working with a 5090 & Bambu X2D. I also use AI to conduct investigations

Austin, TX
Joined March 2021
Photo by Serg (@FraudGnosis)Photo by Serg (@FraudGnosis)Photo by Serg (@FraudGnosis)Photo by Serg (@FraudGnosis)Photo by Serg (@FraudGnosis)Photo by Serg (@FraudGnosis)Photo by Serg (@FraudGnosis)Photo by Serg (@FraudGnosis)Photo by Serg (@FraudGnosis)Photo by Serg (@FraudGnosis)
Time to buy another ssd, I was already considering anyways. If this works it could be game changing
SSD is about to become the only limiting factor for local AI Most frontier open models are ultra sparse MoEs Qwen 3.8 Flash Next = 180B total, ~6B active per token (~3%) GLM 5.3 Flash = 321B total, ~18B active per token (~6%) DeepSeek V4.1 Flash = 552B total, ~16B active per token (~3%) GLM 5.3 = 753B total, ~40B active per token (~5%) Kimi K3.1 Pro = 2.8T total, ~104B active per token (~4%) When ~95% of a model sits idle on every token, keeping all of it in VRAM makes no sense Local AI has been evolving toward this for years: 2023-2025 - Layer offloading (Ollama, classic llama.cpp) This chopped whole layers across GPU and CPU, leaving the GPU waiting on CPU once VRAM filled up Oct 2026 - RAM expert cache (llama.cpp --moe-cache) This shifted from whole layers to individual experts, caching hot ones in VRAM and pulling misses from RAM. Huge leap for decode, but still capped at the size of your RAM Now - SSD expert streaming Early tools like Colibri and Strata proved direct NVMe streaming actually works on consumer hardware. Instead of hitting a VRAM or RAM wall, your SSD becomes an active part of the engine However most people hear "run from SSD" and assume 2 tok/s, which isn't entirely wrong. A few problems still exist: - Basic caching kicks out experts you're about to need again - Waiting on disk reads one layer at a time leaves the GPU and SSD sitting idle - Getting speculative decoding to work well with SSD streaming is hard - Long prompts are tough to keep fast without wiping out your hot cache - Most early tools are experimental engines built for only a couple models Making expert streaming feel fast takes solving all of those together At Magnitude, we've been working on this for the past 2 months. Here's how we solved it: - Smarter caching that keeps your session's most-reused experts in GPU memory - Speculative lookahead that pulls upcoming experts off SSD before the GPU needs them - Direct NVMe reads scheduled across layers so your SSD never sits idle - Streaming prefill that reads each cold expert once without wiping your hot cache - Built into a full engine that tunes itself to your chip across Macs and GPUs All of this together means DeepSeek V4.1 Flash on a single gaming GPU at ~40+ tok/s... We're excited to release it on Monday GitHub here: github.com/magnitudedev/magn…
44
It feels like an AI-generated vibe for UFC now. Everything's so fake. You know the fighters can't say anything, and they get paid peanuts, what really bumps people out is when a fighter loses and then now they lose half their check, which was already peanuts to begin with. Dana's out of touch and still bald, and Hunter is a parasite to the UFC.
UFC just is not hitting the same anymore just have to be honest with myself and stop being in denial
59
We are advancing our civilization
🚨 #BREAKING: Mathematicians conflicted over the emerging field of “cuckulus” whereby mathematicians prompt AI to solve unsolved problems while they just sit back and watch.
41
This is exactly the kind of progress I want to see from American AI labs. I think we’re overlooking a much bigger argument for local AI, though. If corporations, governments, and private individuals are going to have increasingly powerful AI systems of their own, shouldn’t the rest of us at least have some form of intellectual self-defense? We already have an entire civilian market for physical defense. Why shouldn’t we be developing one for intellectual defense as well? A capable local model might be the closest thing we have to a gun for your brain. And I think we need to start framing it that way.
This prediction of mine wasn't accurate and for that I apologize Inkling 2 Mini gives you Opus 4.5 intelligence and capabilities on an RTX 3080 (~300$ USD a year ago) Local AI is winning BIG TIME
1
1
77
In American company giving it a shot this is what I want to see my full supports behind it. I’m about to bench market on my 5090 and do some experiments see what I can come up with in content to post here and spread the word
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index. thinkingmachines.ai/news/ink…
56
If only I got this much engagement on my own posts ha!
82
You can run locally a GPT-6 Luna level model on your RTX 5090 at home.
Qwen3.8-27B beat GPT-6 Luna on Terminal-Bench 2.1 🔥 Qwen3.8-27B just jump from 5.6% to 12.0% on Terminal-Bench 4.1 performance A model DaisyRose-27B made after 18 days of RL on 144 H200s. trained on 1,200 curated long-horizon agent tasks with sessions stretched to 1M tokens, Built on Qwen3.8-27B using their Miles training stack. efficient gains in post-training across agentic and coding benchmarks. - huggingface.co/RadixArk/Dais…
15
40
1
998
31,843
If you could visit the Akashic Records, what would you bring back? Enjoy this retelling of an actual perceived experience I had on one of my first trips. It is my honor to finally be able to share this with the world.
168
301
23
2,746
80,510
Are we all just cattle?
38
72
21
774
30,885
They have Americans by the balls because they know that they can do as much corruption as they want. You’re still gonna buy into Anthropics IPO you’re gonna be dumped on when you least expect it.
Nvidia founder Jensen Huang's ouster removes a major threat to AI company Anthropic as it prepares to go public this week at a possible $1.4 trillion valuation, Business Insider reports. The two have clashed for years over China. Anthropic CEO Dario Amodei wants advanced US chips kept out of China, comparing the sales to "selling nuclear weapons to North Korea." Huang wanted to keep selling to China, calling the comparison "lunacy" and warning that cutting China off pushes it to build its own chips. Huang was also funding cheaper open AI models that compete with Anthropic's. Anthropic has paid $1.63 million to two lobbying firms tied to Secretary of State Marco Rubio and Susie Wiles. Rubio's March 30 memo led to Huang's downfall, and on Monday the administration closed a loophole that let Chinese firms rent Nvidia chips outside China. Follow: @AFpost
152
If you didn’t know, it was a op. Here is another clue. Let’s have the new CEO give out a pay raise on the verge of one of our biggest scandal and during a time where charismatic CEO who built the whole company has been fired.
JUST IN: Nvidia ($NVDA) interim CEO Colette Kress promises company-wide pay raises and additional equity grants months ahead of schedule as the chipmaker scrambles to retain employees following Jensen Huang’s departure, according to WIRED.
62
Serg retweeted
Tips for ADHD #2: You have a Ferrari for a brain. You cannot afford to live small. Otherwise that incredible machinery, without real data and feedback from the world will turn inwards and wreak havoc on yourself, giving you autoimmune issues, anxiety, and god knows what else. You have to move fast and live large, no choice.
Tips for ADHD #1: Steer yourself without planning, by reviewing the day's actions. Planning tyrannizes over yourself. No need. Collect as much data and feedback as you can through knocking off as many actions as you can, then review. Just like how a Chess player gets better by analyzing their own games. You will learn so much about yourself and make faster progress than anyone else hopping from plan to plan and never learning much about themselves even after 10 years.
3
133
7
1,963
48,161
I hate Mexicans who are still romanticizing losing against England at the Azteca. You guys are a bunch of lames, both the real Mexicans, and the Chicanos/pochos. All you foos gotta stop romanticizing being a loser. Just because you saw a good production. Have some self-respect where is your competitive genes at? Why the hell are you celebrating losing?
66
Replying to @kangminlee
Wes Huff has continuously misrepresented church history for years and manipulating truth to fit his calvanist view and has taken the twisting of history to try to undercut Catholicism and Orthodoxy. He has been called a biblical scholar, was a PhD student (never got his PhD) and was known as a "great debater" for beating one guy who is an idiot, which landed him on Joe Rogan. Well, someone looked into him more finally. Looks like his uploaded work that got him through his masters and his PhD work were full of plagiarized material misrepresented as his own as part of his work. He has since run damage control, lied about it the plagarized work being rough drafts, but it was all actually his final work for his degrees. It was even uploaded on his websites to show off. Not only that, but he has no scholarship. He has not done any actual debates. He has countlessly ignored Catholics and Orthodox trying to debate him or discuss his opinions and twistings of church history and misrepresentations. He's just a random dishonest protestant that lacks integrity in his work, has avoided all issues surrounding that for years, and yet became some cult hero because he was on Rogan and talks about old historical biblical artifacts.
14
4
142
5,019
The sudden hate for veterans on X this week is absolutely PSYOPS. It was intentional to prep for TCAVA going to the Senate this week. So here’s my question for those of you who support the Bill, do you still support it after watching the type of tactics that are being used to attack veterans? 
This sudden social media rash of "veterans are parasites" feels inauthentic and coordinated. I wonder what the objective is?
28
43
4
168
8,155
Replying to @AllThingsCivil
Yep. I certainly don’t take any of my benefits for granted, and I consider them a privilege afforded to me by the taxpayer. However, I won’t let anyone characterize them as “free” fringe benefits or suggest I’m somehow less deserving because others, whether by choice or circumstance, didn’t serve and therefore don’t have access to the same benefits. I think civilians who look at veterans’ benefits through the lens of what they themselves aren’t receiving are directing their frustration at the wrong people. The problem isn’t that veterans have these benefits. It’s that so many civilians don’t have comparable protections and opportunities. The answer shouldn’t be a race to the bottom. The loudest voices here on social media advocating for reducing or eliminating veterans’ benefits aren’t solving the inequities civilians face. They’re simply advocating for creating another group of people with fewer protections, without improving anyone else’s circumstances.
7
7
98
5,175
VETERANS, YOU EARNED YOUR BENEFITS. USE THEM! CHOOSE.VA.GOV 🇺🇸
251
577
116
3,690
107,964
I don’t give two shits about the ammo guy, but I do wanna point out that it’s very easy when the VA discussion comes up to act like it’s some free government handout. The Pentagon estimates that since Vietnam military service has been multigenerational at about 75 to 80% of members. That means a small subset of families subsidize the nation’s war fighting efforts. So, of course, it’s easy for most Americans to complain when they’ve literally never had skin in the game nor do they plan to.
47
58
8
710
15,100
Replying to @WatcherGuru
The comback we didn't know we needed :
26
123
8
2,993
131,098