Myttle
@xmyttle • 1,512 subscribers
losing sleep on @Polymarket research AI & predictions
Videos
HE STOPPED RENTING AI AND BOLTED FOUR USED RTX 3090s TO ONE BOARD watch the table: four five-year-old GPUs, 24GB of VRAM each. 96GB of local model memory sitting under one operator’s control. the build looks extreme. the subscription math is worse. then opened his statement and found $340/month going to seven AI companies. six cancellations saved $140 immediately. then he realized a single used 3090 for roughly $700 could already run Qwen 27B locally: - no token meter - no rate limits - no files leaving the machine - roughly 25–30 tokens per second you do not need the four-card monster in this clip. one card moves most daily coding, writing and document work off the cloud. the old flex was paying for every frontier model. the new flex is owning the compute they keep renting back to you.
Myttle74,168 görüntüleme • 1 ay önce
ANTHROPIC FOUND THE MOST EXPENSIVE PART OF AI, EVERY EMPLOYEE STARTS FROM ZERO. one person explains the project to Claude. the answer disappears inside a private chat. the next employee opens another Claude and repeats the same context. Claude Tag removes that reset. everyone in a Slack channel works with the same agent. It sees the thread, remembers relevant decisions, uses the tools assigned to that channel and leaves the output where the next person can continue. Anthropic says the internal version already creates 65% of its product team’s code. the model did not suddenly become 65% smarter. the context stopped disappearing between people.
Myttle30,351 görüntüleme • 1 ay önce
a guy put a local AI coding stack on an Android phone Termux on the screen Ollama underneath a local model answering from the device that tiny terminal is the shift. two years ago, local LLMs felt like punishment fans screaming bad models one token per second cloud looked inevitable now the same idea runs almost anywhere. - Raspberry Pi for tiny chat - MacBook Air for daily writing - RTX 3090 for serious inference - LM Studio when you want the clean GUI - Ollama when you want apps to connect the point is not replacing GPT-5.1 or Opus 4.8 on the hardest tasks. the point is owning the boring 80%: - drafts - summaries - code help - private docs - offline workflows no token anxiety no rate-limit ceiling no sensitive files leaving the machine cloud AI became the premium layer. local AI became the workbench.
Myttle27,108 görüntüleme • 2 ay önce
NASA DID NOT TRUST ONE SENSOR that is why Apollo landed. radar was noisy. gyros drifted. the computer had to guess where the spacecraft really was while every instrument lied a little. that same problem now sits on quant desks. not “where is the spacecraft?” but: what is the real relationship between these two assets right now? most retail traders freeze that relationship into one number. run regression get beta trade the spread pretend the ratio stays alive it does not. the market keeps moving. the hedge ratio drifts. your “market neutral” position quietly becomes directional. a Kalman filter estimates a hidden state from noisy measurements. it does not blindly trust the last tick. it does not blindly trust the old model. it blends both. every tick becomes: > predict the current state > observe the noisy price > measure the surprise > decide how much to trust it > update the estimate that trust-weight is the whole trick. if the old estimate is shaky and the new data is clean, the filter moves fast. if the old estimate is strong and the tick is noisy, it barely moves. that is why a static beta is a dead photograph. a Kalman hedge ratio is a live estimate. the same math that helped NASA fuse noisy signals now helps quants keep a pairs trade from lying to them.
Myttle20,673 görüntüleme • 2 ay önce
THE FIRST AI AGENT SHOULD LIVE IN YOUR DRAFTS FOLDER not in your sent folder. that is the difference. everyone wants the agent to “run the inbox.” bad first move. the safer version is smaller: read the unread emails find what needs attention write the draft save it wait that is Shadow Mode. the agent proves judgment before it gets permission. in this video, Claude processes Gmail and prepares drafts without pretending it should own the whole relationship. that is the correct order. first it watches. then it drafts. then a human approves. then, only after enough clean cases, one narrow category gets automated. pricing questions can move up. simple scheduling can move up. angry customers stay human. refunds stay human. anything with money, reputation or emotion stays locked. the goal is not to build an agent that acts fast. the goal is to build one that knows when not to act.
Myttle13,110 görüntüleme • 2 ay önce
this GitHub README is where the $5k AI OS starts one CLI > Drive > Gmail > Calendar > Docs > Sheets Claude Code stops being a chat box and starts touching the actual business. that is what clients buy. they buy the moment Google Workspace becomes executable. one operator built the first version in 3 hours. got paid $1,650 the same day. then the same setup turned into: - $5k installs - $2.5k-$5k retainers - $15k/month ops layers the margin sits in the boring wiring. context in markdown. connections through CLI. skills as repeatable SOP files. cadence while the founder sleeps. old agencies sold automation diagrams. new operators sell the command layer.
Myttle12,602 görüntüleme • 2 ay önce
Daha fazla içerik yok.