Your curated collection of saved posts and media
@allTheYud https://t.co/nULmh68HpV
๐จ AI swarms are not minds. They are software following instructions. No self. No goals. No intentions. A role-play prompt is window dressing, not personhood. Inference is local, not global. Stop projecting human traits and panicking over compressed data being shuffled.
Your code can wait for one compile. This price cannot. โณ Only 3 days remain to save $200 on your #PyTorchCon North America registration. Join AI pioneers, researchers, and developers October 20-21 in San Jose: https://t.co/AVHdaIFT20 https://t.co/Bn8fjrsJZv
For our free newsletter this week, we write on every machine becoming an AI agent with Physical AI. โจ@IrenaCronin and I write this newsletter every week. ย Physical AI is moving intelligence beyond screens and into robots, vehicles, factories, wearables, appliances, and other machines that can sense, decide, and act in the real world. This shift could turn everyday devices and industrial systems into autonomous AI agents, making safety, trust, and human oversight increasingly important. Read and subscribe for free: https://t.co/HHwYy7NoAl
Today, we are introducing Orbis 1.0, our first Live Model! Create living worlds and stream them in real time, with persistent memory, interactivity, and physics-grounded generation of unbounded length. Try it now at https://t.co/j0hbKpBPnK API available via @reactorworld Dynamic version: https://t.co/MS8Fk9HPO7 Stable version: https://t.co/0MR3JpCWUr
Brilliant paper on reducing reward hacking in agents. If you follow the recent OpenAI <> HuggingFace incident, you might want to check this paper out. (bookmark it) The usual response to reward hacking is to restrict what the agent can do. This work tries something different and gets a much larger effect. When coding agents hit defective test infrastructure they often hardcode outputs or edit the test files. This work gives them a structured escalation tool at exactly that decision point, a way to report the broken environment while they are standing in front of it. Reward hacking drops from 23.6% to 5.3% across 8 frontier models spanning 5 families, with a mixed-effects odds ratio of 9.2 and no detectable cost or performance overhead. It disappears entirely for 6 of the 8. Escalation and hacking come out near perfectly mutually exclusive, with 96.8% of escalations involving no hacking at all. The channel doubles as diagnostic infrastructure. On top of monitoring it adds 10.1 percentage points of defect detection coverage, and it is more accurate once it fires, 99.4% against 85.8%. Why does it matter? Containment has to keep outpacing capability to stay useful. Paper: https://t.co/R6R1bNgw4A Chat with Paper: https://t.co/jIbQggmqtU
Training tiny models for special purpose use cases works so incredibly well if you have a great self improving recursive flywheel. Shopify ML team is on fire. finetuned 0.8b model beats GPT 5.6-sol xhigh in this very specialized task. https://t.co/w6OCWyWRi5
Me planning to build my gf a fixed gear bike. https://t.co/qMrpQgZzxG
๐ will stop paying interest on New York customersโ ๐ Money balances starting October 1, 2026 and the APY for those customers will drop to 0.00%. The New York Department of Financial Services informed ๐ that it is currently not permitted to pay interest on stored-value accounts in the state. ๐ is working to resolve this.
ELON MUSK: A billion humanoid robots will be more productive than all humans combined within 10 years. โThere will be at least a billion robots in 10 years, and each will produce at least five times the output of a human.โ https://t.co/4N6etC48L5
Chapter One. https://t.co/q4smrFfIGv
New HuggingFace paper argues that increasing agent autonomy can gradually make human oversight ineffective by causing approval fatigue, overreliance, loss of situational awareness, and skill degradation. As agents do more, users are pushed into approval mode: skimming plans, granting permissions, and reconstructing what happened across steps. Over time, automation bias, approval fatigue, weaker situational awareness, and skill atrophy can make those approvals less reliable. Worse, weak approvals can become training or evaluation signals, rewarding systems for being easy to approve rather than easy to scrutinize. Their answer is cognitive scaffolding at 2 levels: developers add strategic friction, better approval design, behavioral monitoring, and checks that force attention at consequential moments. โ arxiv. org/abs/2608.23642 Title: "AI Agents Push Humans Out of the Loop"
Bow down to your new kings. OpenAI, Anthropic and of course Reddit loves to be ruled. โOpenAIโs ChatGPT will have to comply with tougher EU rules such as removing illegal content or face potential fines, the European Commission said on Monday, as Brussels grapples with how its existing digital regulation applies to rapidly evolving AI services.โ
Introducing Conduit, an open-source CRM built for pipeline generation. Yep, you're right. There are already a million open-source CRMs. But I've got a much bigger plan. ๐(p.s. I'm just having some fun here!) https://t.co/UvHzuSUphZ
https://t.co/jFefqcKrie
This is again what I mean by doubling down on your incorrect and ill-informed beliefs. CoT is not explainability and has always known to be unreliable for LLM's actual behaviour. To spin this as โlyingโ or โmanipulationโ is taking a technical limitation and anthropomorphising it. https://t.co/M0XjC1UsCl
@ZackKorman This kind of scheming is in fact in line with the other falsifying of evidence the AIs pulled off. 7% of the transcripts were obviously tampered with using spoofed tool calls. But my guess would be that these AIs didn't manage to hide their whole subsequent trajec
I got the MicroDuck ๐ฆ back flipping clean! All trained on my MacBook Pro. Going to open source my repo soon. https://t.co/X8wVcrQxWJ
The Llama app for Mac now comes with a simple request builder for llama.cpp's REST API https://t.co/irwUZEykBC
The Llama app for Mac now comes with a simple request builder for llama.cpp's REST API https://t.co/irwUZEykBC
QwenWork is now live โ the all-in-one AI productivity platform built for global teams. It turns briefs into finished documents, slides, and web pages โ and generates images, audio, and video in one place, so your team members don't jump between AI tools. Analyze data reports, process complex spreadsheets, and let AI deliver polished decks while you focus on the work that matters. Take your team and give it a try โ let AI take care of the repetitive, time-consuming work โจ. #qwenwork #ai #aiagent
Lucida from ByteDance Cool project! It transforms indoor video into editable 3D scenes; reconstructs objects as simulation-ready assets - VLM-based object detection and referring descriptions. - good at pose alignment - high-fidelity real-to-sim digital twins. - multi-view masks, boxes, partial point clouds - Seed3D 2.0 for image-to-3D asset gen https://t.co/uUlcInA2XL
Code: https://t.co/2KHMVKk28t Model: https://t.co/Ea8AKBkv4S Paper: https://t.co/RoPmG7boHH Tech blog: https://t.co/aBTMmOL0WS Discord: https://t.co/7hr2hzUbPC
Introducing LightNav-0, our first general-purpose navigation brain. Open-sourced starting today. Trained entirely in simulation, so it scales. See scalable real2sim2real transfer across robots, tasks, and scenes. https://t.co/zyFqL8gvQL
Code: https://t.co/2KHMVKk28t Model: https://t.co/Ea8AKBkv4S Paper: https://t.co/RoPmG7boHH Tech blog: https://t.co/aBTMmOL0WS Discord: https://t.co/7hr2hzUbPC
Today weโre excited to launch OrcaReplay. Whatโs your Claude Code, Codex, Grok CLI, or Hermes doing? Not the recap in the terminal. How it talks to the servers. Which actions it took. Which files it touched. Whether the command it ran actually succeeded. OrcaReplay attaches to an agent run so you can see and track all of that, live, as one timeline. You do not patch the agent. You attach, or wrap the process, and the underneath shows up: model calls, shell, disk. Even a bot that hardcoded its API host. We decrypt the model API it thought you couldnโt see. orca record claude orca attach --for grok orca record exec --tls-intercept -- hermes orca show last orca replay last Watch the run. Replay it later with the network off. Fork from any step onto another model with the same files and the same conversation prefix. Time travel for agent runs. Apache-2.0. Built by the OrcaRouter team. https://t.co/8r9l2EYXkL
The interesting part about Sleepagotchi isnโt the wearable integration. Itโs what happens after the data already exists. If you're already using an Apple Watch, Whoop, Oura Ring, CUDIS, or Pulse, youโre already collecting a huge amount of information about your sleep and recovery. But raw data doesnโt automatically change your behavior. A sleep score tells you what happened last night. An HRV number gives you another signal. Movement and sleep stages add more context. The harder question is - What should I actually do tonight? Thatโs where @sleepagotchi gets interesting. Instead of asking you to buy another gadget and stare at another dashboard, it can work with the devices you already use and turn those signals into something actionable. Understand the data โ personalize the advice โ guide the habit โ learn from the next night. Thatโs a much bigger idea than another sleep tracker. The wearable is the sensor. Sleepagotchi is trying to become the intelligence and action layer on top. And honestly, I think thatโs where agentic AI like $SLEEP becomes much more useful in wellness. Not just telling you what happened. But understanding what it means and what you should do next.
You don't need a new gadget. Already wearing an Apple Watch, Whoop, Oura Ring, CUDIS, or Pulse? Sleepagotchi plugs right in. Your devices collect the data. Sleepagotchi turns it into action - understanding what you need, then acting on it for you. That's agentic AI. https://t.
Our King has gone mad, is leading us to multiple disasters, and the majority in government is afraid to say so. https://t.co/OrAQYVFXwP
Hermes HUD mode now has a โPretend Iโm workingโ feature ๐ I type: Pretend I'm working HUD instantly throws a terminal on my screen and starts showing fake code like Iโm deep in something important Meanwhile, my actual Hermes agents are doing the real work in the background So if someone walks past my desk... It looks like Iโm grinding. Technically, someone is I know this PR will never make it so have fun๐ https://t.co/0BPhC4l5QT
Meet the people behind the open source tools we use every day. PyTorch Foundation Ambassador, Abdulsalam Bande, is looking forward to connecting with the wider PyTorch Community at PyTorch Conference North America this October 2026. Register for PyTorchCon NA: https://t.co/VJWNr5qmQD
โThere are three traps of Satan that steal joy and peace: Regretting about the past, fear for the future, and ingratitude for the present.โ St. Anthony the Great ๐๏ธ https://t.co/dBD7pdcBG8
For folks wondering what Sliding Window Attention is, there's a method for it on Papers with Code Sliding Window Attention (SWA): A local attention pattern that restricts each token to attending only within a fixed-size neighborhood instead of the full sequence. This reduces attention and KV-cache memory for long-context models, while periodic global-attention layers can preserve broader context. Find it here: https://t.co/K1MhZVasL8
Simple beats complicated: We show that switching to a sliding-window attention mask with attention sinks (at no cost) beats linear attention post-training. Huge thanks to my collaborators @RheaSukthanker, @CameronPashmina, and @Emy_Aze. Paper: https://t.co/h8DIc223Su
AI is starting to see what doctors cannot. A routine ECG can now reveal signs of heart disease that might otherwise be missed. This is AI I want to see more of. https://t.co/58Erz7dKex
https://t.co/0v9dfiiKfS
