Your curated collection of saved posts and media
@emilylambert @ManusAI You can try it out at https://t.co/D9ZOmxDM4W very cool thing too is that you can actually control this from your phone while you're afk and use websites like sentry to debug tricky production issues on your bed like i do let me know if you face any issues! https://t.co/yFQ6IZiIFz
I realised a lot of people didn't know that @ManusAI ships with a browser extension much like Claude for Chrome It's pretty sick and u can even automate everything from your phone as long as your computer is open :) recorded a quick demo, dm if any issues or problems~ https://t.co/9lyY4BMBN5
@emilylambert Have you tried the browser extension from @ManusAI? Would love to know how we might be able to do better :)
Did you know that Grok 3, Grok 4 and Grok 4.1 were all released within one year? https://t.co/jBFi0bUMf2
Make life multiplanetary https://t.co/pq1746EU77
BREAKING: UAE Defense Minister, Deputy Prime Minister, and Dubai Crown Prince ๐ฆ๐ช HH Sheikh Hamdan bin Mohammed welcomes Elon Musk in Dubai https://t.co/Xqsyr8SdtZ
I keep telling people they cannot comprehend how bad things are. People don't want to believe me. It's not just welfare fraud. It's everything. Everywhere. Something terrible has happened to this country. https://t.co/PwonlpwNoa
๐บ๐ธ SAN FRAN BLACKOUT โ WAYMO FROZE, TESLA DROVE Waymoโs robotaxis got a little too real last night - by completely shutting down when San Franciscoโs power outage knocked out traffic lights. Meanwhile, Teslas on FSD? Kept rolling. No drama, no headlines - just handling chaos like itโs a walk in the park. This is what happens when you train your AI on billions of real-world miles instead of coddling it in a simulation padded with perfect data and wishful thinking. Waymo bet on maps and order. Tesla bet on mess - and won. When the lights go out, the difference isnโt theoretical. Itโs traffic. Source: @Tesla_AI, @elonmusk
UN forecasts for Europe: This graphic was not created by the UN. However, the forecast was. Depending on the scenario, foreigners are expected to make up 17%, 26%, 40%, or even 75% of the population by 2050. Don't let anyone fool you. This is the biggest crime of the 21st century. It is a crime because it is not happening naturally. Political resistance against it is censored and repressed.
ใๅฅฝ่ฉๅๆณจไธญใ ใใใใใใใใคใฉในใ ใใพใใใๅงใกใใ ใใใผVer.ใ ๅๅๅถไฝ๏ผBINDing https://t.co/ot6bEDoDmr #ใใคใณใใฃใณใฐ #ใจใญใใ https://t.co/HPVnTuVN8d
ใๅใๆซ้ฒ็ฎใใใใณใในๅใ ใ้ซๅณฐใใใฌใใใใคใฉในใ ใใฏใฉใชในใ #ใใคใณใใฃใณใฐ #ใจใญใใ https://t.co/vKFD0Zc1VY
๐๐๏ธ Ancient Egypt had symbols that feel like interfacesโฆ not just art. Today Iโm obsessed with UDJAT (Eye of Horus) โ some people call it an โactivation codeโ ๐โจ Hereโs the mini ritual people swear by ๐๐ ๐งช UDJAT 1-minute test: ๐ Find a quiet moment ๐งโโ๏ธ Relax your shoulders ๐ฌ๏ธ Inhaleโฆ exhaleโฆ slow your breath ๐๏ธ Focus between your eyebrows (third eye point) ๐๏ธ Say: โOO-JAHโ 3 times โก๏ธ Observe: pressure, vibration, calm, sharper intuition, vivid imagery, orโฆ nothing at all No claims, just curiosity ๐๐ก Sometimes the mind is the portal. Sometimes the voice is the key. ๐๐ถ And sometimes itโs simply your nervous system responding to attention + breath (still fascinating!). ๐ง โจ Would you try this experiment? โ YES / โ NO If YES, report back what happened ๐๐
Neat ideas to improve multi-agent LLM systems. Most frameworks rely on static workflows: fixed role assignments, linear task flows, and limited communication between agents. When tasks involve ambiguity, changing context, or uneven agent performance, rigid pipelines break down. An agent analyzing a financial disclosure might miss new information that contradicts earlier assumptions. Factual errors propagate downstream without correction. This new research introduces an adaptive coordination framework with three core mechanisms: 1) Dynamic routing: This allows agents to reassign subtasks at runtime based on confidence, complexity, or workload. An agent encountering a technical legal paragraph can defer to a compliance-focused peer instead of producing a subpar result. 2) Bidirectional feedback: Enables downstream agents to issue revision requests upstream. A QA agent detecting an inconsistency between two document sections can trigger clarification from the summarization agent, preventing error propagation. 3) Parallel agent evaluation: For high-ambiguity tasks, multiple agents tackle the same subtask independently. An evaluator scores each output on factual correctness, coherence, and relevance, then selects the best result for downstream use. On the SEC 10-K analysis, the full system achieved 92% factual coverage and 94% compliance accuracy, compared to 71% and 74% for static baselines. Revision rates dropped by 74%. Redundancy penalties fell by 73%. Human coherence ratings improved from 3.2 to 4.7 on a 5-point scale. The ablation study revealed shared memory and feedback loops as the most critical components. Removing either caused coverage and coherence to drop by over 20%. Paper: https://t.co/x86kb54zQD Learn to build effective AI Agents in my academy: https://t.co/JBU5beIoD0
Huge if true. MiMo-V2-Flash rivals top-tier open-weight models such as DeepSeek-V3.2 and Kimi-K2, despite using only 1/2x and 1/3x of their total parameters, respectively. Those results look pretty strong for an open-weight model. https://t.co/1k5XMWZMye
Brilliant post on using coding agents. The workflow described here is as close as it gets to my own. From creating rules and skills to optimizing workflows, testing, and more. https://t.co/Wm8aOMq22U
ใๅใๆซ้ฒ็ฎใ ใๅฏพ้ญๅฟRPGXใใใ ใๅคงไบบใใใใใ ๅๅๅถไฝ๏ผMegumu #ใฎใใกใ ใใฌ #ใจใญใใ https://t.co/ZdD7ebxbFM
ใใใใใใใใฟใใใใคใฉในใ ใๅพ่ผฉใกใใใ ๅๅๅถไฝ๏ผใกใญใณใใผใบ #ใญใฑใใใใผใค #ใจใญใใ https://t.co/WfcAkqovNV
Kling Team (Kuaishou Technology) introduces MemFlow MemFlow uses adaptive memory to dynamically retrieve relevant historical frames, ensuring consistent & efficient long streaming video narratives, even with event or scenario switches. https://t.co/GE6pc6WgdE
โUnderstand indigenous healing will also heal other nations and peoples. We've all been living on broken ugly systems built on genocide and slavery. Strengthening and empowering others, in whatever capacity, also helps the self.โ โ Red Haircrow ๐ชถโจ #Wednesdayvibe https://t.co/JIA0vf1Jnm
ใๅฅฝ่ฉๅๆณจไธญใ ใใใผในใใใฏๅๅ ใ็ฌนๆฃฎใใขใจใใใใคใฉในใใใตใญใฅใในใใผใใฉใคใ ้ปๆณๆฒขๆๅญฃไน ๅถๆ ver.ใ ๅๅๅถไฝ๏ผใใใใ https://t.co/ot6bEDoDmr #ใฆใงใWF #WebWF https://t.co/aaMOpusZV2
ใๅใๆซ้ฒ็ฎใ ใใใใใใใใคใฉในใ ใ่ธใๅญใ ไผ็ป้ฒ่กไธญ๏ผ #ใฎใใกใ ใใฌ #ใจใญใใ https://t.co/APEz9eOdyz
#nativeplants #powwowlife #nativeamericanstyle #nativebeauty #nativeheritage #nativeamericanearrings #cherokeecheif #nativespeaker #usa https://t.co/PsQ3K7Q1FL
๐Lovart Christmas Sale Is Here Christmas Unlimited: Dec 20 โ Dec 26 (UTC) Holiday Special: Up to 60% OFF & 365 Days of Zero-Credit Magic. Models included: NanoBanana Pro, NanoBanana, Image 1.5, Seedream 4.5, Seedream 4,Midjourney v7, Kling O1, Kling 2.6, Wan 2.6. https://t.co/1rScjd3vKi
Itโs ALIVEEEE https://t.co/37tQ1ZmjjP
Itโs ALIVEEEE https://t.co/37tQ1ZmjjP
โOpen source for the people. Closed source for the shareholders.โ โ Lisan al-Gaib https://t.co/n4HN688ho0
๐จ Qwen-Image-Layered is LIVE โ native image decomposition, fully open-sourced! โจ Why it stands out โ Photoshop-grade layering Physically isolated RGBA layers with true native editability โ Prompt-controlled structure Explicitly specify 3โ10 layers โ from coarse layouts to fine-g
โOpen source for the people. Closed source for the shareholders.โ โ Lisan al-Gaib https://t.co/n4HN688ho0
Reachy Mini aka Jarvis is alive! The instructions manual is very easy to follow and the native App is awesome. On to the next phase ๐ Unboxing and build video comingโฆ https://t.co/ilPvUJIc7Z
i am sooo bored i am about to duplicate `fineweb` dataset on Hugging Face! ๐ https://t.co/ELXY5M0MTf
i am sooo bored i am about to duplicate `fineweb` dataset on Hugging Face! ๐ https://t.co/ELXY5M0MTf
@ns_schoe @BenCarr630567 @ManusAI Here's where you can find the chat mode toggle https://t.co/8yrfPkKkSw
