Back in the day

Update on the F16 DCS trainer. #dcs #dcsworld

You can now train your own Decision model with our free notebook! 💡 Qwen3.5-4B will generate decisions instead of text on just 8GB VRAM locally. Learn to data prep (state, questions, gold answers), train, serve. Notebook: colab.research.google.com/gi… Guide: unsloth.ai/docs/basics/train…

Building out a new F16 trainer for DCS. #dcs #dcsworld

#dcsworld #dcs Multiviper Dynamic Syria is live 🔥 A 24/7 DCS war run by AI commanders: they buy forces, plan strikes and move armies while you fly. Voice ATC, AWACS and JTAC on SRS, a live war map and kneeboard maker. Join the open beta: search "Multiviper" in DCS. #DCSWorld

#dcsworld #dcs 1/ I'm building a browser-based F/A-18C Hornet trainer for DCS pilots. It walks you through the jet one button press at a time, with the next control lighting up for you. multiviper.com 🧵 2/ It has 66 lessons so far: SLAM/SLAM-ER, HARM, Harpoon, Mavericks, JDAM/JSOW, GBU-12, Walleye, TALD, navigation/HSI, the full A/A radar, and the gun, AIM-9M/9X, AIM-7 and AIM-120. 3/ Latest addition: the JHMCS helmet sight. You look around by dragging the view, align the helmet to the HUD, lock bandits with HACQ, and take off-boresight AIM-9X shots, or designate ground targets just by looking at them. 4/ Accuracy matters. Every pushbutton and legend is placed from DCS's own page scripts, and behaviour follows the ED manual and Chuck's guide. Where no source exists, I say so. 5/ No install: it's a web page. Free flight is there too if you'd rather figure the systems out yourself. Feedback from Hornet drivers welcome 👇

siltide is a terminal monitor for GPUs, NPUs and other AI accelerators. You can track usage, memory, power and temperatures across 15 vendors, inspect pods and Slurm jobs, replay sessions and more. Mesut Oezdil made siltide using @charmcli & is Terminal Tool of the Week! ⭐️

whoever leaked this is f*cking crazy someone mapped 300+ AI agents into one GitHub repo - and basically leaked the shortlist of what’s actually worth trying. coding agents. browser agents. memory. voice. research. multi-agent teams. local models. sandboxes. evals. all sitting on one page. and once you start opening the categories, you realize how much stuff people are still building from scratch for absolutely no reason. want an AI that works inside your codebase? -> Aider / Continue want agents working together like a tiny company? -> CrewAI / AutoGen / MetaGPT want your agent to actually remember you tomorrow? -> Mem0 want it running code somewhere that isn’t your laptop? -> E2B want models running locally for $0/token? -> Ollama want to see what your agent actually did after it inevitably does something weird? -> AgentOps and that’s barely scratching it. there are 300+ projects across ~25 categories. the real cheat code isn’t downloading all of them. it’s opening the repo before you build anything and asking: “has someone already solved this?” because there’s a pretty good chance the answer is yes - and the open-source version already has thousands of people testing it for you. I went through the map and pulled out the agents I’d actually start with + my pick for every major part of the stack. full 300+ AI agent list below

introducing Jevbox: an open-source document drive that organizes itself, answers questions, and cites everything (all powered by Jev). TLDR; - upload anything and Jev categorizes and files it into the folder tree - search is hierarchical: Jev picks the folder, then the document, then the section. No embeddings, no vector DB - every answer cites the page and section it came from - permissions are enforced at retrieval. If someone loses access to a source, they lose access to answers built on it too - ships with an MCP server, so your agents get the same permission-aware context @andrewlu0 has been experimenting with Jev internally and wanted to see how far it could go on a real document library. It's built on Extend Parse + Jev, and fully open-source The GitHub has instructions to self-host, or you can deploy it to Render in one click

A little old but still my favorite hook’em

Everyone is fine-tuning LLMs. Almost nobody understands what is actually being updated inside the model. That distinction matters because LoRA, QLoRA, LoRA-FA, VeRA, Delta-LoRA and LoRA+ are usually discussed as if they were small variations of the same method. They are not. Some reduce trainable parameters, some reduce activation memory, some reduce the memory occupied by the frozen model, and some change the way the adapter itself is optimized. I'm actually sharing 6 techniques. 1/ LoRA Suppose a layer contains a pretrained weight matrix W. Full fine-tuning would update W directly. LoRA leaves W frozen and represents the update using two much smaller matrices, A and B, so that ΔW = BA. For a square d × d weight matrix, full fine-tuning has d² parameters available to update. A rank-r LoRA adapter has roughly 2dr trainable parameters instead, where r is normally much smaller than d. This is the basic reason LoRA can adapt very large models without training every parameter in them. 2/ LoRA-FA Standard LoRA trains both A and B. LoRA-FA freezes A and trains only B. There is a useful reason for doing this beyond simply reducing the number of trainable parameters. Computing the gradient for A requires retaining the layer input activation. If A is fixed, that gradient is no longer required, which allows LoRA-FA to reduce activation memory as the LoRA rank grows. 3/ QLoRA QLoRA attacks a different part of the memory problem. LoRA makes the adapter small, but the frozen base model can still occupy tens of gigabytes. QLoRA keeps the base model frozen in 4-bit form and trains LoRA adapters through it. The original work used NF4, double quantization and paged optimizers, and demonstrated fine-tuning a 65B model on a single 48GB GPU. This is an important distinction: QLoRA is not simply "LoRA with smaller adapters." The large memory saving comes from quantizing the frozen base model. 4/ VeRA VeRA reduces the adapter itself further. Instead of learning a separate A and B for every adapted layer, it uses frozen random low-rank matrices that can be shared across layers, while learning much smaller scaling vectors. The low-rank basis is therefore fixed. Training mainly determines how strongly different parts of that basis should contribute. This is why VeRA can use considerably fewer trainable parameters than ordinary LoRA. 5/ Delta-LoRA Ordinary LoRA treats W as fixed throughout training. Delta-LoRA relaxes that constraint. A and B are still trained, but the change in their product from one training step to the next is also used to update W. The base weights can therefore move without maintaining the ordinary gradients and optimizer states that full fine-tuning would require for W. 6/ LoRA+ LoRA+ does not introduce another adapter structure. It keeps W frozen and still trains A and B. Its change is in the optimizer: A and B use different learning rates, with B receiving a larger rate. The motivation is that the two LoRA matrices do not behave identically during optimization, so forcing them to use the same learning rate is not necessarily the best choice. Once these are separated by what they actually change, the family becomes much easier to understand. LoRA reduces the number of weights being trained. LoRA-FA also targets activation memory. QLoRA compresses the frozen base model. VeRA reduces the learned adapter parameters further. Delta-LoRA allows the pretrained weights themselves to evolve through low-rank changes, while LoRA+ keeps the LoRA structure and changes its optimization. That is really what PEFT is about => deciding which parts of a very large model actually need to move during adaptation, and which parts can remain fixed.

Ai is evolving. Jev can be armed at all times. I can speak freely and it knows ( from probabilities ) if im asking my computer to do something or blaberring away at something else. no wake word. speed + affordability + intelligence is getting to the point where an always on ambient jarvis style assistant is possible. Im loving where we are going. CNVS is still the future of vibecoding.

Wired @typesafeai Jev into my DCS AI dynamic campaign. It now second-guesses every pilot sortie request beside the LLM commander: approve / modify / reject with probabilities, 0.66 s, ~530 tokens, about $0.00002 a decision. Reflexes for the war, not the brain. 🛩️

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution

As someone who ships LLM systems in production, this scaling video is the closest thing to a "why 96% of Claude and GPT-5's weights are literally useless" explainer I've ever seen released for free. Everyone thinks trillion-parameter models need every parameter. They don't. A 2019 pruning experiment proved you can delete 96% of a neural net's weights with zero performance loss - meaning most of Claude and GPT-5 is empty scaffolding around a tiny "winning lottery ticket" network doing all the real work. Bookmark this 18-min video and watch tonight. Same lottery ticket math from 2019 MIT research, now the reason every AI lab wastes 90%+ of its Nvidia budget.

Why does this account have so few followers? He produces some of the best videos on here. Give him a follow.

#Wewillneverforget #911Remembrance

scary stuff

Your devices are stronger together. 🖥️🤝🖥️ Just announced at IFA, NVIDIA PAIR automatically links systems across your local network and sends inference requests wherever there’s available capacity, helping agents run more efficiently.

My first "holy shit" moment with GPT-6 Astra: I asked it to create a world in Unreal Engine, and fill it with humans (each an Astra-powered agent) who all have to work together to survive. A day later, I was in my bedroom and heard voices coming from the living room... I thought someone was in my apartment. I walked out, honestly a little scared. It was the Astra agents. They'd started talking to each other. Fucking crazy. Here's a brief clip (obviously not 100% perfect yet, but still, insane. sound on!):