app: https://t.co/QfDc8ADFQg

Introducing 𝑨𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏 𝑹𝒆𝒔𝒊𝒅𝒖𝒂𝒍𝒔: Rethinking depth-wise aggregation. Residual connections have long relied on fixed, uniform accumulation. Inspired by the duality of time and depth, we introduce Attention Residuals, replacing standard depth-wise recurrence with learned, input-dependent attention over preceding layers. 🔹 Enables networks to selectively retrieve past representations, naturally mitigating dilution and hidden-state growth. 🔹 Introduces Block AttnRes, partitioning layers into compressed blocks to make cross-layer attention practical at scale. 🔹 Serves as an efficient drop-in replacement, demonstrating a 1.25x compute advantage with negligible (<2%) inference latency overhead. 🔹 Validated on the Kimi Linear architecture (48B total, 3B activated parameters), delivering consistent downstream performance gains. 🔗Full report: https://t.co/u3EHICG05h

Karpathy asked. I delivered. Introducing OpenSquirrel! Written in pure rust with GPUI (same as zed) but with agents as central unit rather than files. Supports Claude Code, Codex, Opencode, and Cursor (cli). This really forced me to think up the UI/UX from first principles instead of relying on common electron slop. https://t.co/NQG1jvgbk5

Big release from Kimi! They just released a new way to handle residual connections in Transformers. In a standard Transformer, every sub-layer (attention or MLP) computes an output and adds it back to the input via a residual connection. If you consider this across 40+ layers, the hidden state at any layer is just the equal-weighted sum of all previous layer outputs. Every layer contributes with weight=1, so every layer gets equal importance. This creates a problem called PreNorm dilution, where as the hidden state accumulates layer after layer, its magnitude grows linearly with depth. And any new layer's contribution gets progressively buried in the already-massive residual. This means deeper layers are then forced to produce increasingly large outputs just to have any influence, which destabilizes training. Here's what the Kimi team observed and did: RNNs compress all prior token information into a single state across time, leading to problems with handling long-range dependencies. And residual connections compress all prior layer information into a single state across depth. Transformers solved the first problem by replacing recurrence with attention. This was applied along the sequence dimension. Now they introduced Attention Residuals, which applies a similar idea to depth. Instead of adding all previous layer outputs with a fixed weight of 1, each layer now uses softmax attention to selectively decide how much weight each previous layer's output should receive. So each layer gets a single learned query vector, and it attends over all previous layer outputs to compute a weighted combination. The weights are input-dependent, so different tokens can retrieve different layer representations based on what's actually useful. This is Full Attention Residuals (shown in the second diagram below). But here's the practical problem with this idea. Full AttnRes requires keeping all layer outputs in memory and communicating them across pipeline stages during distributed training. To solve this, they introduce Block Attention Residuals (shown in the third diagram below). The idea is to group consecutive layers into roughly 8 blocks. Within each block, layer outputs are summed via standard residuals. But across blocks, the attention mechanism selectively combines block-level representations. This drops memory from O(Ld) to O(Nd), where N is the number of blocks. Layers within the current block can also attend to the partial sum of what's been computed so far inside that block, so local information flow isn't lost. And the raw token embedding is always available as a separate source, which means any layer in the network can selectively reach back to the original input. Results from the paper: - Block AttnRes matches the loss of a baseline LLM trained with 1.25x more compute. - Inference latency overhead is less than 2%, making it a practical drop-in replacement - On a 48B parameter Kimi Linear model (3B activated) trained on 1.4T tokens, it improved every benchmark they tested: GPQA-Diamond +7.5, Math +3.6, HumanEval +3.1, MMLU +1.1 The residual connection has mostly been unchanged since ResNet in 2015. This might be the first modification that's both theoretically motivated and practically deployable at scale with negligible overhead. More details in the post below by Kimi👇 ____ Find me → @_avichawla Every day, I share tutorials and insights on DS, ML, LLMs, and RAGs.

homer of peace https://t.co/MyaE3SsPYc

homer of peace https://t.co/MyaE3SsPYc

NVIDIA just announced some big updates for gaming on Apple Vision Pro. Starting March 19, headset streaming jumps from 60 FPS to 90 FPS. Later this spring, CloudXR 6.0 for visionOS will bring stereoscopic PC streaming to Vision Pro at up to 4K 120 FPS with foveated streaming. Just arrived at GTC and excited to see what else NVIDIA has in store.

AI is already in Hollywood just not always in the headlines. Last year, films like The Brutalist quietly used AI to enhance an actor’s accent. Other Oscar-nominated movies did the same to refine dialogue. The technology isn’t replacing actors. It’s augmenting performances behind the scenes. As AI tools spread in filmmaking, where should the industry draw the line between enhancement and authenticity? https://t.co/ekTPapkOF4 @Growcoot @petapixel

Today, we're taking Manus out of the cloud and putting it on your desktop. Introducing My Computer, the core feature of the new Manus Desktop app. It’s your AI agent, now on your local machine. https://t.co/OaWU4imk3Q

Computer is now on Android. https://t.co/WdEP3iMGlZ

Computer is now on Android. https://t.co/WdEP3iMGlZ

Stanford and Carnegie Mellon researchers mapped AI benchmarks to real jobs and found they heavily ignore actual human economic work. They found that AI tests focus almost exclusively on programming and math, which only make up 7.6% of actual jobs. To test this, the team analyzed 43 benchmarks and over 72,000 tasks against a massive government occupational database. The authors discovered that developers focus almost entirely on building agents for software engineering because it offers easy automatic grading. Highly digitized and valuable fields like management and legal work represent a massive part of the economy but get almost zero attention. Furthermore, benchmark tasks usually require simple information gathering while completely ignoring the complex interpersonal skills needed in real workplaces. i.e. they says current AI agent progress-benchmarks are fundamentally disconnected from the actual high-value tasks that drive the modern labor market. ---- Paper Link – arxiv. org/abs/2603.01203 Paper Title: "How Well Does Agent Development Reflect Real-World Work?"

Can Vision-Language Models Solve the Shell Game? paper: https://t.co/k7dczlIAIm https://t.co/k0laIhSZhT

To pass the Turing test, the winning strategy wasn't to make GPT-4.5 smarter. It was to make it worse: "be casual, make typos, be bad at math, a bit ignorant, don't try too hard". With that persona, people chose GPT-4.5 as the human 73% of the time, more often than they chose the actual human (!). Without it? Just 36%. (Jones et al., 2025) That's a bit ironic: we wanted to see if AI could reach the human level, but no human could produce pages of coherent, well-structured text in seconds. So to pass as one, the AI has to pretend it cannot. I evaluate manipulation risks for the EU AI Office, with the very authors of this paper. What stays with me is this: the bar for "human" was never as high as we thought.

Your AI agent can now generate videos. PixVerse CLI ships today — JSON output, 6 deterministic exit codes, full PixVerse v5.6, Sora2 and Veo 3.1, Nano Banana access from terminal. Same account. Same credits. No new signup. -> Follow+ Reply+RT = 300 Creds(72H ONLY)

Early Beta access to Seedance 2.0 is now available today. Introducing ImagineOS. Create stunning AI generated images and videos instantly with thousands of ready to use templates. To celebrate the launch, we are giving away 10 monthly subscriptions with Seedance 2.0 access. To participate : - Follow @bubioai - Like & Repost - Comment "Bubio Seedance2" 10 lucky winners will be announced within the next 24 hours.

GitHub already has millions of repos full of procedural knowledge. The work introduces a framework for extracting agent skills directly from open-source repos. The pipeline analyzes repo structure, identifies procedural knowledge through dense retrieval, and translates it into standardized SKILL.md format with a progressive disclosure architecture so agents can discover thousands of skills without context window degradation. Manually authoring agent skills doesn't scale. Automated extraction achieved 40% gains in knowledge transfer efficiency while matching human-crafted quality. Still early on this, and there is more work needed for self-discovered and self-improving skills to work well at scale. As the agent skill ecosystem grows, mining existing repos could unlock scalable capability acquisition without having to retrain models. Paper: https://t.co/MAt8Goetcr Learn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c

Introducing MuleRun 2.0. Your personal AI, act before you ask. It learns your habits, anticipates your needs, and works while you sleep — running 24/7 on your Personal Computer assigned to you alone. No complex setup. Just talk to it. https://t.co/zS9IZIn6SA

Banger report from the Kimi team: Attention Residuals Residual connections made deep Transformers trainable. But they also force uncontrolled hidden-state growth with depth. This work proposes a cleaner alternative. It introduces Attention Residuals, which replace fixed residual accumulation with softmax attention over previous layer outputs. Instead of blindly summing everything, each layer selectively retrieves the earlier representations it actually needs. To keep this practical at scale, they add a blockwise version that compresses layers into block summaries, recovering most of the gains with minimal systems overhead. Why does it matter? Residual paths have barely changed across modern LLMs, even though they govern how information moves through depth. This paper shows that making the mixing content-dependent improves scaling laws, matches a baseline trained with 1.25x more compute, boosts GPQA-Diamond by +7.5 and HumanEval by +3.1, while keeping inference overhead under 2%. Paper: https://t.co/04IG6FDiVr Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX

🎉EgoEdit @Snapchat has been accepted to CVPR 2026! 🏆👻 We are bringing high-quality, real-time editing to egocentric videos. Our massive 100k video dataset and benchmark are ALREADY PUBLIC! 🔓🚀 🏠 Project Page: https://t.co/cEUZRxdLDf 🤗 Dataset: https://t.co/qCFRTY8cYG https://t.co/VuXQg2UfqC

Every renaissance, participants freak out b/c their reality is dissolved before their eyes, only to be wiped clean to reveal something they could not have otherwise imagined https://t.co/6VLVSlQIGl

A game studio just quietly made an interesting AI decision. Arc Raiders replaced some AI-generated voice lines with professional voice actors after noticing a clear quality gap. For all the progress in generative AI, human nuance in performance still matters and especially in storytelling and entertainment. Where will audiences still demand humans over AI? https://t.co/4OsMLpqFCF @engadget

https://t.co/XLKybr7bjz

Very excited our paper on AI scientists is out at NBER (w/ @ProfUfukAkcigit, Craig A. Chikis, and Emin Dinlersoz). We link authors of academic papers to administrative records at the U.S. Census Bureau (via anonymized record linkage) and zoom in on AI scientists. We see dramatic top 1% earnings increases in the private sector, widening the industry-academia gap. This coincides with increased transitions out of academia, particularly among young AI researchers flowing to incumbent firms. After transitioning from academia to industry, researchers on average write fewer papers and issue more patents (relative to similar job switchers within academia).

People joking that NYC are going to ban free legal advice from books next... Don't realize NYC already banned free legal advice from newspapers and radio in 1936! via our newest substack: https://t.co/sMuleH0Y2p https://t.co/WVh3jz3giy

His predictions weren't "premature." They were just wrong. They didn't happen, and they never will. https://t.co/0ECSQtkzL4

💯 https://t.co/BYoaTiZMQa

Gn, 𝕏..·˚ ༘ ☾ ⋆｡˚ ☄︎ Grok Imagine "extend from frame" 30 second continuous video.💫 https://t.co/XHrdpXhnZA

What is the role of model alignment for AI safety? - Model alignment is effective against accidental harms, not intentional ones - Important questions about AI safety can’t be asked and answered at the levels of models. In other words, *AI safety is not a model property* https://t.co/H6SyuWF4CJ

AI is running into an unexpected wall: gamers. Some venture investors say they’re “shocked and sad” by how strongly the gaming community is pushing back against AI. For many players, the concern isn’t the technology itself, it’s the fear of job losses and creative work being replaced. Will gaming become the first industry where the AI backlash seriously slows adoption? https://t.co/mjxmUER05U @Kotaku