Jens Bohnwagner

rabbit inc. R1 and the Humane Pin are not working as planned. Maybe Siri will be better! 🤯 Making the Vision model able to control and understand UI is still a very hard task. We test so many things in AQ22 to make a model able to understand desktop UI. In April, Apple released the paper "Ferret-UI," a multimodal vision-language model that understands icons, widgets, and text on iOS mobile screens and reasons about their spatial relationships and functional meanings. It was a really interesting paper, and Ferret was one of the models that made me think that small on-device LLMs will be the future. Why It’s a Big Deal? This isn't just about making Siri smarter. With Ferret-UI’s deep screen understanding, it’s easy to see how it could become a full-on assistant that helps you navigate your phone entirely by voice or other means. Think of it as having an on-device helper that knows your screen as well as you do. Apple’s New Approach? What's also surprising is Apple’s openness with this project. They shared a lot of details about how they trained Ferret-UI, including the datasets and benchmarks they used. This level of transparency is rare for Apple and shows they’re serious about leading in AI research on small on-device models. Paper link in the comments. #ai #apple #largelanguagemodels #deeplearning

24

2 Comments

AI Video Generation: Initial Feature Implementation Complete I took some time over the weekend to implement the basic functionality of AI video generation using the recently released Luma AI API. 1. Users can input descriptions or keywords in Korean for the video they want to create. 2. The system then analyzes the input to generate a video prompt tailored for Luma AI. 3. Finally, the generated Luma prompt is used to produce the video. Total development time was about two hours. The API documentation was less than ideal, so I had to directly analyze the Node.js module code provided by Luma to make it work. The ultimate goal is to build a B-roll service. The name should give you a hint of what we’re aiming for. It’s an exciting time to be in tech. There are so many possibilities for new services to create.

4

🚀 Accurately detect objects in overhead imagery with the SOTA VajraV1 models using the VayuAI SDK: https://lnkd.in/gQBx5cab

1

1 Comment

🗝️ Need Skyvern to log into websites with TOTPs? Don't worry -- we've got you covered! We already supported QR-code based TOTPs, but email / SMS based TOTPs were the bane of our existence.. until Shuchang Zheng stepped in. Shuchang Zheng just launched a killer feature: TOTPs within Skyvern. You can now give Skyvern a TOTP retrieval URL, and when Skyvern is presented with a screen asking for a 2FA code, it'll call that URL to grab the code and automate whatever is behind that paywall! Want to learn more? Read our advanced feature documentation here: https://lnkd.in/gmKkdJ55

10

3 Comments

🔥 We're thrilled to announce the SRT MiniServer 2.5 Beta! 🔥 Here's a sneak peek at what's new in this beta release: ✨ Redesigned Remote Control: Experience a sleek and intuitive web interface for effortless control. ✨ UDP Multicast in SDI OutAddon: Enjoy enhanced flexibility with UDP multicast generation, now featuring powerful transcoding capabilities. ✨ RTT Reporting: Gain valuable insights into your network performance with comprehensive RTT reports, allowing you to identify and troubleshoot issues effectively. ✨ Enhanced UI Highlighting: Benefit from improved UI elements that provide clear warnings for latency and RTT mismatches, ensuring optimal streaming parameters. ✨ Advanced TS-recording (daily archive) Ready to explore the future of SRT streaming? Download the SRT MiniServer 2.5 BETA today and share your feedback! #SRTMiniServer #streaming #video #broadcast #videoengineering #livestreaming #remoteproduction SRT Alliance

11

1 Comment

Q: How do we connect the Meta verse (or Omniverse) to the Universe in real time? A: With an Event Mesh. Proof: This very cool paper and experiment is a great step forward in Metaverse manufacturing and real time factory floor connectivity and VR operations of the future. If you find this space interesting, you could read more in the below articles about the possibilities: https://lnkd.in/eHF9ptiw and https://lnkd.in/e-sb4Zmw

21

1 Comment

This video shows what #Plugnix is capable of - what those network diagrams mean in real life. This is a network you can instantiate anywhere in the world at the speed of software. Sincere respect for all the inventors who created the legacy technologies of present, but if we are going to win, we need an edge network revolution. No more boxes (contested logistics), and no more limited supply chain and high cost (custom hardware). Agnosticism. Speed. Security. CI/CD. Dynamic Network reconfig on the fly, over the air. For war and for peace. #Military and #PublicSafety. *We did not demo anything related to Ground Control Station capabilities, but the launch customer is big and important. IYKYK.* Shoutout to Air Force Research Laboratory funding the free UAS Tool, which is part of the enabling technology behind this stack. #CJADC2 #JFN #ATAK #WINTAK TAK Product Center Aaron Sikorski Paul Mandal Motorola Solutions TrellisWare Technologies

75

9 Comments

If you didn't make it to the Apteco global launch event last week, you might be interested in our Apteco Gen-AI Assistant announcement. Watch https://lnkd.in/gTdthUpG to see "One more thing" announced in the best tradition of tech reveals. The Apteco Gen-AI Assistant will change how our users interact with the Apteco marketing platform and their first party data. You can start learning about it now... #AptecoGlobalLaunch

17

1 Comment

🚀 Introducing version 3.2.0 of our Native #PlantUML Editor for #iPad with #AI capabilities. 
This update brings new features: 
- prompt refinements
- detection of use-case diagrams from drawing.
- generic fix and improvements. 
See it on App Store https://lnkd.in/dHED6DKp 👀

3

OpenAI has recently released Sora, their text-to-video AI model, to ChatGPT Pro and Plus users. This innovative model can generate high-quality video clips up to 20 seconds long at 1080p resolution from text prompts. Sora's capabilities include: Creating videos from text descriptions Animating still images Extending existing videos Filling in missing frames Generating seamless video loops Blending different video clips The model uses a combination of diffusion and transformer architectures, allowing it to understand and simulate complex scenes with detailed characters and backgrounds59. Sora builds on learnings from DALL·E and GPT models, using techniques like recaptioning to improve prompt adherence. While impressive, Sora still has limitations, such as accurately simulating certain physical interactions5. As AI video generation technology advances, it opens up new possibilities for creative expression and storytelling. However, it also raises important considerations about deepfakes and content authenticity. As we explore the potential of AI-generated video, it's crucial to consider both the creative opportunities and the ethical implications of this technology

6

1 Comment

"Upgrade your VR game with Razer's latest innovation! Say goodbye to discomfort and hello to immersive experiences. The new face gasket and head strap are game-changers. Have you tried it yet? Comment below! 🎮🔥 #VR #Innovation #Razer" Original article: https://lnkd.in/e2igT7fM

1

Use #MO71 simple column filters to narrow down the number of #IBMMQ objects or status you are viewing in list. Simple to use and very powerful. #WorkSmartWithMQ

6

CTRL Documentable now has local search for all generated #Unreal C++ and #Blueprint properties, functions, and nodes. (we'll be releasing the first version of this plugin within the week) #buildinpublic #unrealengine #ue5 #documentation #opensource #programming #development #gamedev #ctrlframework #indiedev

14

We just released a frame interpolation node on Substrate – fun to combine this one with image generation techniques. You can make videos like this with a simple two-node workflow! ControlNet (4 images) → InterpolateFrames

18

1 Comment

Building a tool to pick helpful augmentations pipeline is in the short term TODO list. If you can share how you choose augmentations and their parameters right now it would be very helpfull

13

We managed to train a very lite STT AI transcription model using my own Singaporean-accented voice recordings and synthetically derived data (from HELF AI) to get a rather promising 60% accuracy. This preliminary model gives me hope that HELF will be able to disrupt the incumbents and provide highly accurate medical AI transcriptions that are 1) accurate, 2) cheap, if not free, 3) works strictly on-prem so that your private healthcare information never leaves your own device. AI medical transcription is currently provided by tech giants and they are charging an arm and two legs for the service. Most require cloud access as well and have demanding hardware requirements. I want to make it cheaper, better and faster. Going to take a break now, hope our next training iteration will yield something even better! Support us to change Healthcare for the future! www.helf.co #AI #MedicalAI #Transcription #HELF #ML #health #healthtech #doctor

4

If your project requires a top-end performance of the camera system, take a look at IFF SDK: https://mr-te.ch/ With IFF SDK you will be able, for example, to capture raw 8K 10-bit images at 115 FPS or 4K 10-bit images at 425 FPS and process them (applying black level subtraction, histogram calculation, white balance, demosaicing, color correction, etc.) in real-time or write image streams from two such cameras (in raw, TIFF or DNG formats) to RAID storage with a 10 GB/s throughput and without frame loss. #iffsdk #mrtech #8k #imageprocessing #machinevision

9