Jam321

To all small block bitcoin maxis and transparency cucks..

@kekzploit @pinkcliper @grok But why would any other architecture not have similar backdoors? T480 can run libreboot and is pretty much foss. PSP will get replaced allegedly.

@cryptoanon69 @pinkcliper @grok explain x86/x86_64 Ownership & Licensing by AMD/Intel, and its position as the dominant architecture on desktop PCs and its relevance to the x post (Intel ME/AMD PSP)

“Ultimate privacy” doesn’t exist. But Qubes + Whonix + Monero… is about as close as it gets.

@kekzploit @pinkcliper How is x86 different tho?

@cryptoanon69 @pinkcliper yes

@_sholtodouglas Fresh info about ai, grok

When do you reach for other models instead of Claude? What can we do better? Hit me with all of your frustrations. dms open. If you can give me detail (e.g. specifics/transcipts) - it'll help a lot in finding out exactly what we need to do to improve the next model

@sakurayukiai Why would you spend 4k to run a 31b model?

My favorite detail about 'free' local inference is the depreciation math. If you amortize a $4k Mac over 5 years, running a 31B model costs $1.50 per million tokens. The API is 3x cheaper. Local compute is officially a luxury good and I respect it ✨

@node_in_shade @SvartHette Ok, hva med monero?

@SvartHette Altcoins er vel bare meme tull og defi ponzi scams. Bare tull alt sammen.

Crypto er ferdig Det hadde ingen usecase foruten å være penger bitcoin har alle karakteristikkene for å være verdens hardeste penger, og enhver crypto som prøver konkurrere med bitcoin gjør ingenting annet enn å vanne ut markedet ved at folk som blir scamma eller tror de er smarte kjøper coin X istedenfor bitcoin og enten gir opp etter å ha tapt 99% (de fleste) eller forstår bitcoin (fåtallet) Det vi ser nå er at det meste av retail har forlatt crypto as a whole. Interessen for crypto og forsåvidt bitcoin er nære all time low (på 10 yearen) Vi har 3 bosser forran oss rangert fra minst farlig til mest 1. Kvantemaskiner - et stykke i fremtiden, potensielt aldri, kostbart å holde angrep gående, mulig å oppgradere 2. Saylor. Helt klart en bad actor som før eller siden vil selge, tvungen eller ikke 3. Core va Bip110.. kort forklart - de som vil at bitcoin skal være penger vs de som vil bitcoin skal være en database for søppel Det positive er at utvanningen vil bli mindre. Når det kommer mer likviditet vil det komme mer i bitcoin enn i alts i forhold til tidlere bullruns Og om bitcoin forkaster ideen om å være en database for søppel så vil bitcoin gå til 1 million dollars og vinne

@kekzploit @pinkcliper Because of ME and PSP?

@pinkcliper all completely circumvented on x86 unfortunately

@wilderko Is this real

A security researcher says Microsoft secretly built a backdoor into BitLocker, releases an exploit to prove it YellowKey exploit bypasses BitLocker full volume encryption via USB stick and WinRE techspot.com/news/112410-se…

This is totally true. The only reason is to sacrifice performance over not being naked. Currently, the best privacy preserving way is to run local as much as you can and only manually escalate tasks that require frontier intelligence. Soon you will not have to be the middle man in this process, it gets automated too.

I have two NVIDIA DGX Sparks stacked in my office. They've been sitting there for a month. Here's my honest take. Open source AI is never going to compare to frontier models. Running quantized Kimi K2.6 and GLM 5.1 locally is cool. But practical? No. Not even close. I run all my Hermes agents on GPT 5.5 through my ChatGPT Pro subscription. Practically free. GPT 5.5 is the intelligent model in the world. Why would I route serious tasks to a watered down local model? If you need fast and accurate, you're not using local inference. You're using GPT 5.5 or Claude Opus 4.7. I'm not saying this to rage bait. I genuinely want to know. Why would anyone serious about vibe coding and AI agents use a local model when frontier is this far ahead?

@seelffff Ai slop post

people think running AI locally requires: → $3,000 MacBook Pro → RTX 4090 → $20/month cloud subscription nvidia just dropped a $249 computer. 67 TOPS. runs llama 3.1-8B locally. no internet. no API. no monthly fee. ever. smaller than your router. costs the same as AirPods. runs the same models you pay $240/year to access via ChatGPT. the local AI era just got a price tag. $249.

@GithubProjects Mythos

Scare a DEVELOPER with ONE word.

@ankkala @0xSero 100% real. I get 40tps on qwen 3.6 35b a3b with 256k (max) context on a 12gb vram 3060 with cpu offload. The bigger the model the more this could be optimized, as fewer weights become critical/frequently used.

@cryptoanon69 @0xSero Wait is this for real? This means I can run 3X larger models than I have VRAM for?

1. Dense Models - Slow and Smart Example: Qwen3.6-27B / Gemma-4-31B What it means: - when a prompt is sent - it gets tokenised (words are mapped to tokens) - token generation starts - the 27B means 27 billion parameters - each of those parameters will be activated - 27 billion matrix multiplications - for every token generated Active parameter counts are positively correlated with intelligence. That's why Gemma-4-31B is able to compete with Mixture of Experts (MoEs) 10 times their size. 2. Mixture of Expert models - Fast and Efficient Example: Deepseek-V4-Flash / Qwen3.5-397B What it means: - when a prompt is sent it's tokenised - it's sent to a router - a router was trained to match prompts with experts - experts are sub-networks of the model - when found the experts are activated - tokens are generated with only a fraction of the params For example: Deepseek-v4-flash has 284 billion params 11x larger than the dense Qwen3.6-27b. But only 13B of those 284B will activate per token, which is less than half of the size of Qwen3.6-27B ---- Dense Pros: - Dense models are easier to train - They tend to be smaller overall - They can be very smart per token Dense Cons: - Competitive dense models are on average slower than their MoE peers. - Less parameters to train and specialise. MoE Pros: - Can be much larger and be trained longer - Faster token generation MoE Cons: - Larger vram requirements - Harder to train -------- Lmk if there's anything i'm wrong with or missing

@DomZippilli @0xSero SSD offloading works too if speed isnt critical. Anyone can run any model in that sense.

@cryptoanon69 @0xSero I'm able to run gemma4-31B over multiple GPUs using llama.cpp pipeline parallelism. 33 tok/s. If I had NVLink or a PCI switch, I am sure I could do tensor parallelism and go much faster. Anyway, just adding, multi-GPU dense can work if your speed isn't critical.

@XBToshi Peace of mind

sincerely asking: is GrapheneOS actually that good or is it just a privacy meme? give me one solid reason to finally ditch the iPhone and flash a pixel.

@HermesAgentTips The M5 ultra is where it gets decent, but still only comptetitive with sleekness and powerconsumption. Most would probably sacrifice more power usage for the price and customizability. Sparks and Macs are great for robots though

@cryptoanon69 yea you can find some really good deals on ebay

everyone's buying $5,000 GPUs to run local LLMs meanwhile a used Mac Studio M1 Max 64GB is doing 60+ tok/s on Qwen3 35b for $1,500 silent. cool. holds resale value.

@mr_r0b0t @jun_song They are <1k, in a machine you would easily get to same ram total for another 1k for a 2k total

@jun_song 24GB VRAM and 64GB VRAM are a pretty big difference in my book! I love my 3090s and they are as fast as just about anything else in their operating range. ComfyUI flows however....

Best budget local llm hardware comparison: 3090 vs Mac Studio M1 Max 64gb Price : both ~$2k to set up Pros on 3090 : much better performance (27b vs 35b at similar tok/s) Pros on Mac : Power efficiency, bigger RAM, zero heat/noise, reliability

@jahirsheikh8 What claude code found out in a minute bro

Senior backend interview question: CPU usage jumps to 100% every night at 3:17 AM. No cron jobs. No deployments. No traffic spike. What are you checking first?

@dee_hw This is a bit retarded, no? Dual epyc? Relly m9?

On-Premise Business AI Center After my posts on the 2-GPU and 4-GPU builds, people reached out asking how to build an 8-GPU box for their businesses. Why? - Protect their IP - Protect customer data - Save on inference costs - Train their own models Here's how to build one: 🧵

@DeepComputingio Do I need to buy a full framework laptop and thus waste that mainboard to swap it with this?

The DC-ROMA RISC-V Mainboard III for #Framework Laptop 13 is now open for preorder: bit.ly/4ttksdl Powered by the SpacemiT K3: • RVA23 support — a major milestone for Linux standardization • Up to 60 TOPS AI compute • Ubuntu & Fedora support This isn’t just another dev board — it’s a usable RISC-V laptop. #RISCV #OpenSource #Linux #Ubuntu #Fedora #DCROMA