ThreadWeaver Adaptive Threading for Efficient Parallel Reasoning in Language Models https://t.co/LzYpML0iSs

discuss: https://t.co/RYaNe0NByW

Ilhan Omar: "The U.S. government will only do what Somalians in the U.S. tell them to do. They will do what we want and nothing else. They must follow our orders." https://t.co/ZPey4EqNfN

We are still here. The leaders we choose are continuing with the legacy. https://t.co/Zo4WLu75Ou

Divide in Canada is by DESIGN!! Federal Gov is FAVOURING NEWCOMERS while treating Canadians like peasants !! https://t.co/G1HBqg9vJr

@jpdenantes44 @JohnLee90252472 @JimBeattie18 @DEOLINDAMA93701 @ceconomou56 @lillianromerof @BrindusaB1 @AnnaCountessK @mariatontini1 @jimwyman5 @Mirosla29898402 @maype7 @Marzena99557145 @TopBananaMall Bonjour cheríe Jp! Bonne dimanche ♥️🌹🙋‍♀️ https://t.co/8cLGT8mcFs

Native American Restitution #nativeamerican #nativestyle #americanindia... https://t.co/z8fom0nfrF via @YouTube Truth!

The new head of the British secret services is the granddaughter of a Ukrainian nicknamed “the Nazi Butcher”. He was so terrible, enjoying killing himself, that even his fellow Nazis were appalled. Blaise is now finishing her grandfather’s fight against Russia, on the side of Ukrainian Nazis. In short, the MI6 chief's "Nazi Butcher" grandfather would have been particularly proud of what she is doing.

Wan-Move Motion-controllable Video Generation via Latent Trajectory Guidance https://t.co/BQFd0kd9p3

discuss: https://t.co/LtHrunvO1w

model: https://t.co/L1t0M5MW9S

Wow cool! New research on, literally, "why ice is slippery". :D https://t.co/08dwHzpC6C

my brother-in-law's alfred hitchcock-themed airbnb in small-town california is killing it https://t.co/IdldTZkYWy

my brother-in-law's alfred hitchcock-themed airbnb in small-town california is killing it https://t.co/IdldTZkYWy

Arc Raiders: Game in AI debate in running for top Bafta award https://t.co/TeTFDNTqGU @ARCRaidersGame @tommyr345 @bbcnews

Accenture and Anthropic launch multi-year partnership to move enterprises from AI pilots to production https://t.co/8AFdgWKbMp @accenture @AnthropicAI

🖌️ EDIT TEXT is here. Like you always wanted. Accurate Recognition: Reads text in complex layouts, wild fonts, even half-hidden areas. Flexible Editing: Edit, delete, or rewrite any text in the image while keeping the original style, just like a document.

thank you Woke Elon https://t.co/DfPME4LSKU

🚨 POST-QUANTUM CRYPTOGRAPHY: THE ENCRYPTION WAR HAS BEGUN 🚨 Your passwords are not “safe for years” anymore. They’re already broken – you just don’t know it yet. 🧨 Right now, powerful actors are using a strategy called “Harvest Now, Decrypt Later”: 🗓️ 2024: they quietly record your encrypted data ⚡ 2030+: they use quantum computers to crack today’s encryption in hours ➡️ Your secrets today = public knowledge tomorrow What’s at risk? 1️⃣ Your bank accounts – from “secure” to “steal everything” 💸 2️⃣ State secrets – diplomacy, defense, critical infrastructure 🏛️ 3️⃣ Your health data – perfect blackmail material 🧬 4️⃣ Cryptocurrency – keys broken, assets worthless overnight ₿🔥 The old crypto is not enough. We need new quantum-resistant shields: 🧊 CRYSTALS-Kyber – encryption defender 🔶 CRYSTALS-Dilithium – signature guardian 🦁 SPHINCS+ – strong, stateless signatures 🧱 Lattice-based schemes – math fortresses for the post-quantum era This isn’t sci-fi. It’s a timeline problem: 🔐 Classical decryption: millions of years ⚛️ Quantum decryption: ~8 hours If you build in: • AI • Web3 / Crypto • Fintech • Healthcare • Governments / critical infrastructure …you cannot ignore post-quantum cryptography anymore. Now is the time to: ✅ Audit your encryption ✅ Track PQC standards (CRYSTALS-Kyber, Dilithium, SPHINCS+, lattice-based) ✅ Start a migration roadmap before it’s too late The Quantum Apocalypse is not a movie. It’s a roadmap – and the countdown has already started. ⏳ Save & share this post with someone who still thinks “SSL padlock = safe forever”. #QuantumComputing #Cybersecurity #AI #Web3 #Blockchain

Nearly a third of American teens interact with AI chatbots daily, study finds https://t.co/SnSTc6sau0 @claresduffy

When a chatbot’s advice is a matter of life or death, how can we leave AI to the free market wild west? https://t.co/O5lG8nbvNf

Introducing Jais-2 - a major leap for Arabic AI. Co-developed with @G42ai , Inception, and @mbzuai, Jais-2 is a family of Arabic models built end-to-end on Cerebras from pretraining to inference. Jais-2 70B runs at 2,000 tokens/sec – 20x faster than leading LLMs. It brings deep cultural grounding and dialect awareness to more than 400 million Arabic speakers worldwide. Learn more: https://t.co/eej9cyti8K

the future of software is personal. as personal as bird watching. if the app you want doesn't exist, make it. software isn't owned by programmers anymore. it's owned by people who know exactly what they need. bird watchers. knitters. runners. parents tracking their kids' art. people cataloging their vinyl. the more niche, the better. n=1 software. made by you, for you.

Minimax @Hailuo_AI 2.3 has seemed to fly under the radar, but at the moment it is one of our favorite models available. Here we ran a test comparing it with @Kling_ai 2.6 because we wanted to see how it performs in a side by side test with a model we believe is one of the best on the market. Here are the different prompts we ran… Prompt 1: Slow camera push-in as a futuristic flying transport vehicle lifts off from a platform. Prompt 2: Cinematic, medium-shot of a woman frantically running away from something in the forest, the camera is tracking the woman as she runs. Prompt 3: close-up shot of an orangutang eating a banana at a zoo Prompt 4: two scientists are working in a laboratory. Suddenly, something in front of them begins smoking and burning. Both scientists run out of the shot as they run for their lives. Prompt 5: two knights (on horseback) gallop across an open field, as the knights ride across the field the camera cranes up (revealing more of the wide open landscape). Verdict 🏆: I was intentional about testing each model thoroughly in a unique way with each prompt, whether it was testing physics with a lab fire or action shots with a jungle chase, before drawing my conclusions. I was impressed with how Hailuo handled skin texture and prompt adherence. Similarly, I continued to be impressed with Kling’s physics and camera movements. Surprisingly, Hailuo 2.3 won 2 out of 3 tests; making it the technical winner. However, I would still lean towards Kling as the greater model given what it has to offer compared to Minimax Hailuo. Needless to say, the competition was closer than I expected. If you haven’t been paying attention to Minimax Hailuo 2.3, it might be worth a look.

lmao Woke Elon got my back https://t.co/o61iJrCk8O

🚀 Introducing VoxCPM 1.5! We are taking realistic speech generation to the next level with major upgrades in audio quality and stability. 🎧✨ What's new: 🔊 44.1kHz Hi-Fi Audio: Now supports high-sampling-rate waveform decoding (up from 16kHz). ⚡ High Efficiency: Generates 1s of audio with only 6.25 tokens (prev. 12.5), maintaining speed with higher quality. 🛠 Dev-Friendly: New scripts for LoRA & Full Fine-tuning to enable deep customization. 📈 Enhanced Stability: Minimized artifacts and improved performance on long-form audio. Experience the new standard for TTS. 👇 Huggingface: https://t.co/eBcaVeD6ba Github: https://t.co/VjpJxoyoLr #VoxCPM #TTS #AI #OpenSource #OpenBMB

Pipecat 0.0.97 release. Some highlights: Support for @GradiumAI's new speech-to-text and text-to-speech models. Gradium is a voice-focused AI lab that spun out of the non-profit Kyutai Labs, which has been doing architecturally innovative work on neural codecs and speech-language models for the last two years. Continued improvements in the core text aggregator and interruption handling classes, both to fix small corner cases and to make behavior as configurable as possible. This is the kind of often-invisible work that underpins Pipecat's ability to support a wide range of models and pipeline "shapes." Models stream (or don't stream) tokens differently. Different use cases need to make different engineering trade-offs in the service of natural, low-latency interactions. Similarly, continued steps towards full support of reasoning models. Mostly, reasoning models haven't been used in voice AI pipelines, because we are generally prioritizing low latency. But, increasingly, we are using multiple models in parallel in voice agents. Thinking fast and slow, as it were. Using reasoning models requires updating `LLMContext` abstractions to thread thought signatures into the conversation context, and handling function call internals slightly differently. Access to word timestamps from the @cartesia_ai speech-to-text model. The Smart Turn model service now defaults to the new v3.1 weights and uses the full current utterance rather than only the most recent fragment.

Last month, Google dropped something interesting: five AI Agent papers released across five consecutive days, one per day, each digging into a different part of how agents should be built, evaluated, secured, and deployed. No big splash, just a steady rollout of more than 250 pages of highly technical material. Here’s the distilled version of what those five drops covered: Read. Learn. Bookmark.

Put together some notes on Mistral Vibe, Mistral's new CLI coding agent (think Claude Code / OpenAI Codex) which is Apache 2 licensed, written in Python+Pydantic+Textual and has a neat set of system and tool definition prompts that are fun to read https://t.co/PccM3HkMKr

One month ago, we released a 41-page ebook on context engineering. Turns out, we had way more to say. Our new blog post dives deeper into the discipline of treating the context window as a scarce resource and designing everything around it (retrieval, memory, tools, prompts) so your LLM spends its limited attention budget only on high-signal tokens. 𝗧𝗵𝗲 𝗦𝗶𝘅 𝗣𝗶𝗹𝗹𝗮𝗿𝘀 𝗼𝗳 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴: 1️⃣ 𝗔𝗴𝗲𝗻𝘁𝘀: Orchestrate decisions and manage information flow dynamically 2️⃣ 𝗤𝘂𝗲𝗿𝘆 𝗔𝘂𝗴𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻: Refine user input for different downstream tasks 3️⃣ 𝗥𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹: Optimize chunking strategies to balance precision and context 4️⃣ 𝗣𝗿𝗼𝗺𝗽𝘁𝗶𝗻𝗴 𝗧𝗲𝗰𝗵𝗻𝗶𝗾𝘂𝗲𝘀: Guide the model on how to use retrieved information 5️⃣ 𝗠𝗲𝗺𝗼𝗿𝘆: Design layered systems (short-term, long-term, working) that don't pollute context 6️⃣ 𝗧𝗼𝗼𝗹𝘀: Enable real-world action through the Thought-Action-Observation cycle The blog includes a complete walkthrough of building a real-world agent with 𝗘𝗹𝘆𝘀𝗶𝗮, our open source agentic RAG framework. You'll see how built-in tools (query, aggregate, cited_summarize) and custom tools work together in a decision-tree architecture with global context awareness. Read the full blog post here: https://t.co/uuDcf3o0K2

If you found this helpful, follow us @AINativeF for more insights. A like or share on the first tweet would mean a lot—thank you for your support! Image Credit: Flux https://t.co/RF8pCvxzz6

If you found this helpful, follow us @AINativeF for more insights. A like or share on the first tweet would mean a lot—thank you for your support! Image Credit: Flux https://t.co/Vb0lWSzWfv