Cristian Mora’s Post

Microsoft Purview DLP now has the capability to review scanned documents (PDF, JPEG, PNG) using OCR technology (optical character recognition ). Even more impressive, it can detect handwritten text!! https://lnkd.in/dk5ZAGHA #TechInnovation #MicrosoftPurview #DLP #OCR #HandwritingRecognition

Here is a recent preprint for #task #oriented #communication design of joint source-channel coding optimized for #hardware parallelism and #FPGA implementation in 6G! Thanks Hungpu(Hong-fu) chou for inviting me to your interesting research! Happy to see all the community effort to push this line of research! #AI #telecommunications #wireless #task #oriented #goal #oriented

Transcribing badly scanned documents with Vision LLMs unlocks a lot of use cases 🔥 OCR / solutions like unstructured often don't work well. Use our prompt to transcribe any document (EN/DE/RU/ES) to a format that's easily understandable in downstream LLM tasks: https://lnkd.in/dXcVMVaJ #transcription #gpt4V #promptengineering

Some things I found over the last days that may be interesting for you :) 4M: Massively Multimodal Masked Modeling by #Apple https://lnkd.in/dn--2Qhr RouteLLM: A framework for serving and evaluating #LLM routers "Trained routers are provided out of the box, which we have shown to reduce costs by up to 85% on widely-used benchmarks such as MT Bench while maintaining 95% GPT-4 performance." https://lnkd.in/dX3_QT4f a friendly guy who gave us #Mac user a small open source PDF tool for semantic search: https://lnkd.in/dvakXdtW he used this model: https://lnkd.in/d2saNZsn

It's time to stop wasting time entering data manually. Using cutting-edge OCR technology, the Images-to-text Converter can reliably and rapidly extract text from photographs and scanned documents. Reduce your time spent typing and devote it instead to more pressing concerns.

🔍 Unleashing the Power of OCR: Simplified Guide with Code Samples in Python Hey Folks 👨🏻💻 Excited to share with you my latest blog post that dives into the fascinating world of Optical Character Recognition (OCR)! 🚀 In this comprehensive guide, I demystify OCR techniques and provide simple implementations with code samples. 💻 Optical Character Recognition (OCR) is a technology that enables the conversion of scanned images and PDFs into editable and searchable documents. OCR technology is widely used by businesses and individuals to convert paper documents into digital formats for easy storage, retrieval, and sharing. The technology works by digitizing text and images and then using algorithms to recognize and translate the characters into a machine-readable format. OCR technology has greatly simplified document management and made it easier for people to access and use information in a digital format. 🔎 What's Inside? ✅ Understanding OCR: Learn how OCR technology converts scanned or photographed text into machine-readable text, revolutionizing the way we interact with text. ✅ Practical Use Cases: Discover how OCR finds applications in various industries, from digitizing printed documents to automating data entry and enhancing accessibility. ✅ Hands-on with Python: Explore step-by-step instructions and code samples using popular OCR libraries like Tesseract, Keras OCR, and EasyOCR. Unlock the power of text extraction with ease! 👩💻👨💻 Whether you're a tech enthusiast or simply curious about OCR, this blog post is designed to make the topic accessible to everyone. No technical jargon just simplified explanations and practical examples. 💡 So, if you've ever wondered how OCR works or wanted to leverage its potential in your projects, this blog post is a must-read! 💡 📚 Dive into the world of OCR now and unlock the power of transforming images into valuable textual information. 🌐 Read the full blog post here: https://lnkd.in/dARn7YT3 #OpticalCharacterRecognition #OCR #BlogPost #Blogging #DataCapture #DataProcessing #DigitalTransformation #image #processing #and #computer #vision #photography #photooftheday #love #technology #records #instagood #instagram #follow #photo #tech #hiyanshbheda #life #company #nature #original #with #business #pictureperfect #for #software #photographer #sunlight #like #programmer #art #machinelearning

For Image Encoder, the image resolution size and the data set on which the models were trained varied between the CLIP and AIM models. The following table shows the results of each ablation.Table.... #Apple39s #April #Gunton #language #Large #Matthew #MM1 #models #multimodal

My current Rust toy project: transform jpeg A to appear identical to jpeg B without decoding the image bytes and animate the transition. Jpeg images aren't stored as RGB triplets - the pixel data is compressed, so modifying encoded bytes directly leads to glitchy/weird transformations. More on Jpeg encoding: https://lnkd.in/ggR6fyDT The first image is the original - the second image was changed by a single byte and saved, which resulted in cascading glitches in the lower third of the image. I want to do this thousands of times and make a gif. I already got it working-ish with random adjustments, but it was too slow. Working on a new approach.

Instruction Tuning the Largest Pretrained Retrieval-Augmented LLM This exciting new paper from NVIDIA introduces Retro 48B, the largest LLM pretrained with retrieval. Continues pretraining a 43B parameter GPT model on additional 100B tokens by retrieving from 1.2T tokens (using the Retro augmentation method). The Retro 48B model shows significant perplexity improvement over its GPT 43B counterpart. Scaling the Retro model to 48B means it can be instruction-tuned more effectively. This work applies instruction tuning to Retro 48B and demonstrates significant improvement (+7%) over the instruction-tuned GPT on zero-shot question-answering tasks. The important insight from this work is the potential benefit attained from pretraining with retrieval. Results highlight the promising direction to obtain a better GPT decoder for QA through continued pretraining with retrieval before instruction tuning. (paper in the comments)

11/10-11 Angela’s AI Newsletter: Follow me and stay updated on today's trending topics on Twitter! Andrej Karpathy Discusses Security and Memory Management in LLM OS 🔗: https://lnkd.in/gPGXP8UD 📌: Andrej Karpathy delved into the security and memory management aspects of LLM OS. He emphasized the importance of distinguishing between trusted and attacker-controlled memory in the OS, highlighting the challenges in protecting sensitive data like SSH keys. NousResearch Introduces Advanced Long-Context Language Model Yarn-Mistral-7b-128k 🔗: https://lnkd.in/gZRQHFgp 📌: Yarn-Mistral-7b-128k by NousResearch is a state-of-the-art language model designed for long-context processing. An extension of Mistral-7B-v0.1, it has been further pretrained for 1500 steps on long-context data using the YaRN extension method and supports a context window of up to 128k tokens. Exploratory Analysis of GPT-4's Value in Medical Diagnosis 🔗: https://lnkd.in/gvtU2GUT 📌: This analysis demonstrates GPT-4's impressive performance in medical diagnosis, where it outperformed 99.98% of human journal readers in complex clinical cases. However, it also highlights GPT-4's limitations and the need for optimization in medical applications, emphasizing the potential for further advancements in AI-assisted diagnostics. Open source Python library: enabling web interaction with multi-modal LLMs like GPT4 🔗: https://lnkd.in/gT9UhFTi 📌: Introducing Tarsier, an open source Python library to enable web interaction with multi-modal LLMs like GPT4! Abacus AI Launches Cost-Effective Retrieval APIs for Open Source LLMs 🔗: https://lnkd.in/gdpNmvSQ 📌: Abacus AI has launched open-source retrieval APIs, offering an economical and flexible solution for developing RAG applications with custom LLMs, including options for fine-tuning and selecting various open-source LLMs. Aleksa Gordić Creates an All Open-Source Jarvis for Images: allows real-time voice control 🔗: https://lnkd.in/gu5ArJ_p 📌: Aleksa Gordić shared his latest creation: a fully open-source Jarvis for image manipulation, running locally without external APIs or OpenAI. This system allows real-time voice control over images, including searching, selecting, segmenting, modifying via inpainting, and even conversing with them. All demonstrated features were recorded in a single take, with only the diffusion model generation step edited for brevity. By Angela Chen & ChatGPT. Contact Email: hanzhespring@gmail.com #ai #chatgpt #artificialintelligence #openai #elonmusk