Image created with OpenAI GPT-Image-1. Image prompt: over-the-top 1990s pro-wrestling promo poster, split-screen arena featuring “Modal Monster” juggling video, text, and audio icons mid-roar; RGB floodlights, grainy print texture, vivid neon titles
So to recap: – Yesterday, frontier closed model equivalent reasoning model from Qwen, – This morning, frontier closed model equivalent reasoning vision capabilities from stepfun – sometime today(?) a frontier video model from wan? All open source What is America doing?”” / X https://x.com/Teknium1/status/1948744914876920039
In 2022, @GoogleDeepMind launched Ithaca to help restore, place and date ancient texts. Now, they’re working with collaborators to introduce Aeneas, a new AI model that contextualizes ancient Latin inscriptions. 📜 Learn more ⬇️”” / X https://x.com/Google/status/1948039522194718799
Introducing the first model for contextualizing ancient inscriptions, designed to help historians better interpret, attribute and restore fragmentary texts. – Google DeepMind https://deepmind.google/discover/blog/aeneas-transforms-how-historians-connect-the-past/
Neat example of AI in the humanities. A Google model trained on Latin text fills in lost parts of Latin inscriptions & identifies related texts Historians increased their accuracy by 44% when working with the AI (Though AI alone beats historians, historian + AI was usually best) https://x.com/emollick/status/1948063719042498587
Our new state-of-the-art AI model Aeneas transforms how historians connect the past. 📜 Ancient inscriptions often lack context – it’s like solving a puzzle with 90% of the pieces lost to time. It helps researchers interpret and situate inscriptions in their past context. 🧵 https://x.com/GoogleDeepMind/status/1948037924882133390
Train and deploy robot skills in the cloud… with just one click. Gr00t‑n1.5 is now live on Phosphobot, making training and inference simpler than ever. Example prompt: “”Grab food and place into bowl.”” Tips for better results ✅ Record longer episodes (~30–40s) ✅ Target https://x.com/IlirAliu_/status/1947721603082817884
Meta FAIR recently released the Seamless Interaction Dataset, the largest known high-quality video dataset of its kind, with: 4,000+ diverse participants 4,000+ hours of footage 65k+ interactions 5,000+ annotated samples This dataset of full-body, in-person, face-to-face https://x.com/AIatMeta/status/1947692466205037006
timescope: testing if large models understand long videos or they just claim to do so 🤠 they randomly insert needles (short videos/static images) in long videos and ask questions about the needle itself 🤯 Gemini seems to be the best! very cool work by @orr_zohar et al 👏 https://x.com/mervenoyann/status/1948049876228452788
RT @liliang_ren: We’re open-sourcing the pre-training code for Phi4-mini-Flash, our SoTA hybrid model that delivers 10× faster reasoning th…”” / X https://x.com/algo_diver/status/1946397862767767921
RT @liliang_ren: We’re open-sourcing the pre-training code for Phi4-mini-Flash, our SoTA hybrid model that delivers 10× faster reasoning th…”” / X https://x.com/ClementDelangue/status/1946246738823545317
Data-efficient and Accurate Vision Models from Synthetic Data
https://microsoft.github.io/DAViD/
DAViD: Data-efficient and Accurate Vision Models from Synthetic Data”” TL;DR: Training only on high quality human centric synthetic data; diverse in terms of poses, environments, lighting, and appearances, and not tailored to any specific evaluation set. https://x.com/Almorgand/status/1947669634800398607
Dream, Lift, Animate: From Single Images to Animatable Gaussian Avatars https://research.nvidia.com/labs/dair/dream-lift-animate/
How soon until AI can continuously fuse together all sensor data into a persistent 4D model of reality? https://x.com/bilawalsidhu/status/1947474834973131158
Huge. Take any image (real or synthetic) and turn it into a multi-part 3D object using @Scenario_gg. https://x.com/bilawalsidhu/status/1947673321014735099
The value of generating multi-part 3d meshes cannot be overstated — much easier to rig and animate things without a ton of manual work. Loving what scenario has been doing, esp since going back to their 3d roots! https://x.com/bilawalsidhu/status/1946364106606256281
VoluMe – Authentic 3D Video Calls from Live Gaussian Splat Prediction https://microsoft.github.io/VoluMe/
Robots learning in simulation, and then mastering the real world. [📍 Bookmark GitHub Repositories] Today’s video shows a full Sim2Real pipeline with the OMY robot. ✅ Compact 6-DoF manipulator built for AI and robotics research ✅ Trained in Isaac Sim, validated in Gazebo, https://x.com/IlirAliu_/status/1946504663349481768
Robots won’t master the real world by training in The Matrix. Real data (not shortcuts like simulation or proxy datasets) is the key to building foundation models that TRULY generalize. @svlevine calls these shortcuts “sporks”… clever but flawed stand-ins that try to https://x.com/IlirAliu_/status/1947199867799122306
ReasonVQA drops a giant 4.2M‑question stress‑test on Visual Question Answering, forcing models to pull facts from Wikidata and link them to what they actually see. The authors wrap up by stressing that ReasonVQA is a low‑cost, 4.2M‑question benchmark that forces models to pull https://x.com/rohanpaul_ai/status/1948319999636148226
🎉 Big news! We’ve raised $110M from new and existing investors, including @nvidia & @Snowflake This funding reinforces our position at the forefront of AI innovation, with exciting releases like Reka Vision, Reka Research & Reka Flash 3.1 Read more 👇 https://x.com/RekaAILabs/status/1947689320594157668
DrafterBench shows top LLMs still stumble on basic drawing edits, scoring roughly 80 rather than 100. The benchmark targets the everyday pain of engineers who now fix PDF plans by hand. The study builds a 1920‑task suite covering 12 edit types across text, tables, and vectors. https://x.com/rohanpaul_ai/status/1946907944773443888
The problem is not just the proliferation of devices that let you record people without their knowledge, but the fact that multimodal LLM let you use recordings in ways that neither law not society anticipated. Everyone has an easy way to mine hours of footage. No forgetting. https://x.com/emollick/status/1948100250175942713
Gemini 2.5 Flash-Lite is now stable and generally available – Google Developers Blog https://developers.googleblog.com/en/gemini-25-flash-lite-is-now-stable-and-generally-available/
Gemini 2.5 Flash-Lite, our most cost-efficient and fastest 2.5 model yet, is now available — complete with the same capabilities that make our 2.5 model family so useful, like multimodality, thinking and connection to tools like Search. Learn more 🔦⬇️”” / X https://x.com/Google/status/1947689382892204542
Gemini’s native text to speech (TTS) capabilities are available for scaled production use 🗣️! Both 2.5 Flash and 2.5 Pro are available and are powerful for use cases like NotebookLM style podcast content. https://x.com/OfficialLoganK/status/1947328086577492309
Tavily seems pretty nice for a web search/web extraction tool. Anyone have a better one?”” / X https://x.com/Teknium1/status/1946079899544158352
Github 👨🔧: Automate browser-based workflows with LLMs and Computer Vision Github 👨🔧: Automate browser-based workflows with LLMs and Computer Vision Skyvern automates browser-based workflows using LLMs and computer vision, replacing brittle, code-defined XPath interactions. It https://x.com/rohanpaul_ai/status/1948712082431447194
Dubbing for Everyone: Data-Efficient Visual Dubbing using Neural Rendering Priors https://dubbingforeveryone.github.io/
Can a simple bird’s-eye view make LiDAR robots navigate more accurately than ever before? [📍Bookmark the paper & code for later] EV-LIO(LC) is a new LiDAR-Inertial Odometry framework that blends Bird’s Eye View (BEV) image representations with point cloud registration and loop https://x.com/IlirAliu_/status/1948077653468029386
🎞️MoVieS: Motion-Aware 4D Dynamic View Synthesis in One Second”” TL;DR: Feed-forward framework that jointly reconstructs appearance, geometry and motion for 4D scene perception from monocular videos in one second. https://x.com/Almorgand/status/1946251119325274153
HairCUP https://bjkim95.github.io/haircup/
It’s been fun collaborating with @superwhisperapp, a blazing fast AI speech-to-text app, with support for local and cloud models. By putting @vercel CDN in front of their model API, they’re seeing up to 350ms gains in some geos 🤯 To be clear, this is 350ms+ faster by *just* https://x.com/rauchg/status/1947019072908230674
Elon Musk’s Neuralink says owned by ‘disadvantaged’ persons in filing https://www.cnbc.com/2025/07/17/elon-musks-neuralink-says-owned-by-disadvantaged-persons-in-filing.html
Multimodal models still leak harmful text when attackers mix tricky words and images. AutoSteer adds a lightweight layer-aware safety prober and refusal head to frozen multimodal LLMs, driving attack success below 5% while leaving regular performance unchan A safety awareness https://x.com/rohanpaul_ai/status/1946527437913035110
Scale your production apps with the stable version of Gemini 2.5 Flash-Lite. ⚡ It’s faster than our 2.0 Flash models, more cost-efficient, and outperforms 2.0 Flash-Lite across coding, math, reasoning, and multimodal understanding. Start building → https://x.com/GoogleDeepMind/status/1947689582012633542
The new OpenAI editing update is incredible, so I made a little app to try it out. Just enter your API keys and you’re good to go. You can create or edit any image. I love having this saved on my home screen so I can use it on the go. https://x.com/skirano/status/1946642722430804413
A joint OpenCLIP (3.0.0) and timm (1.0.18) release day today. It’s been a quarter since the last OC release, so what’s new? PE (Perception Encoder) Core support was the headline feature. Using the timm vision encoder for the PE models, I adapted the weights from @AIatMeta so they”” / X https://x.com/wightmanr/status/1948108826206707744
DAViD: Data-efficient and Accurate Vision Models from Synthetic Data https://microsoft.github.io/DAViD/
Can “screenshotting a page and feeding it to the LLM” can replace existing document processing solutions? I saw some discussion on this on HN, and was inspired to write a blog post on this. TL;DR you absolutely should use LLMs for doc processing, but you need a lot of tuning to https://x.com/jerryjliu0/status/1948475176062255504
Deep-learning system teaches soft, bio-inspired robots to move using only a single camera https://techxplore.com/news/2025-07-deep-soft-bio-robots-camera.html
What if robots could actually feel what they touch, and get things right… even when vision fails? [📍Bookmark paper & code for later] Robots usually struggle with delicate tasks like inserting test tubes or transferring fluids. This new handheld gripper might change that. A https://x.com/IlirAliu_/status/1945751072221417661
Paper propose an efficient framework to compress multiple video-frame features before feeding them into large multimodal models, thereby mitigating the severe token explosion arising from long or dense videos. The study proves a state space compressor packs 256 frame videos into https://x.com/rohanpaul_ai/status/1947232835753926670
RT @vllm_project: The @huggingface Transformers ↔️ @vllm_project integration just leveled up: Vision-Language Models are now supported out…”” / X https://x.com/ClementDelangue/status/1947775555387916397
The @huggingface Transformers ↔️ @vllm_project integration just leveled up: Vision-Language Models are now supported out of the box! If the model is integrated into Transformers, you can now run it directly with vLLM. https://x.com/vllm_project/status/1947756551663718754
Chinese company Robot Era has unveiled their next-gen robot L7, a 5′7″ tall humanoid. The company has also showcased the ERA-42 Vision-Language-Action model running on L7 robot to autonomously execute dexterous tasks. https://x.com/TheHumanoidHub/status/1947564066488389939
Robots usually need tons of labeled data to learn precise actions. What if they could learn control skills directly from human videos… no labels needed? Robotics pretraining just took a BIG jump forward. A new Autoregressive Robotic Model, learns low-level 4D representations https://x.com/IlirAliu_/status/1947346268637769909
