Tenyks

Google is shipping again as if there’s no tomorrow! 🚢 Here are Google’s top five recent innovations: 1. Imagen 3: An advanced image generation model capable of producing high-quality visuals across various styles, from photorealistic landscapes to textured oil paintings. 2. Gems: A platform for creating custom AI experts tailored to specific tasks or topics. Users can design their own Gems by providing instructions and a name, then interact with them as needed. Examples include: A "Learning Coach" Gem that breaks down complex topics for easier understanding. A "Coding Partner" Gem that enhances coding skills and assists with project development. 3. Gemini-1.5-Flash (0827): This updated model has made significant progress, jumping from 23rd to 6th place overall in the Chatbot Arena rankings. 4. Gemini-1.5-Pro (0827): The latest version demonstrates notable improvements in coding and mathematical capabilities compared to its predecessors. 5. Native Prompt Gallery in Google AI Studio: This new feature allows users to experiment with: Long-context applications Native multi-modal inputs (including image, video, and audio) Structured outputs Various other advanced AI functionalities 🚀 If you’re interested in multimodality, foundation and frontier models, don’t miss our post where we show you how to cascade foundation models (SAM 2 + GPT-4o).

Will you be in Germany this week? 🇩🇪 If so, come pay us a visit at AIAI Berlin where we will be talking all things Gen AI and Computer Vision. 🤖 We are super excited to attend another sure-to-be fantastic AIAI event running 11th-12th September and will be waiting for you at Booth 5. Alongside the booths there is a jam-packed two days of talks and lectures - all taking place at Waldorf Astoria! Not in Berlin? Not to worry - we are happy to jump on a video call anytime to explore how Tenyks' dynamic insights and visual search can help you boost model performance. (https://www.tenyks.ai/)

Seek and You Shall Find 🔎 📽 Scouring through large scale video collections is a time drain. Wouldn’t it be awesome if there was a simple way to extract insights from videos? Well, that’s exactly the feature we have been busy developing at Tenyks! Check out the following video to see our platform in action. Key Points: - Over 16hrs of footage searchable in seconds with fine-grained sensitive activity detection, all using Tenyks Video Processing pipelines - NO models were fine-tuned in the making of this video (we employed Foundation Models) - Data is taken from the publicly-available MEVA dataset (link) Fancy giving video search a go for yourself? Reach out for a chat and we’ll be happy to get you going (https://www.tenyks.ai/)

Tenyks on the Big Screen 🎬 Although we're not (yet) at 'social network'-style blockbuster level, thanks to Cambridge Enterprise, we were able to create this incredible video documenting the partnership between our two organisations, and the history of Tenyks. We hope you enjoy watching the video as much as we enjoyed making it. Should you be interested to learn how to utilise Tenyks' state-of-the-art Visual Search Engine to deploy your models into production 10x faster then send us a line, or book a free demo: https://lnkd.in/dQtbtEX6

🚀 Segment Anything Model (SAM 2) is arguably one of the best models Meta has released to date. 🎥 In our SAM 2 Series, we not only explain what SAM 2 is but also put it to the test alongside GPT-4o by cascading these foundation models through visual prompting. 🔧 In Part 1, we show you how to set up SAM 2 and how to run it on your own custom videos. Check it out for yourself using the Jupyter Notebook and Blog posts below: 📓 Jupyter Notebook: https://lnkd.in/eqJXNepx 📝 Blog post: https://lnkd.in/eqUWset4

🚀 Segment Anything 2 (SAM 2) is here, a real-time object segmentation model for images and videos: 🖼️🎥 SAM 2 is a unified and versatile model capable of segmenting any object in images and videos using in-context learning. That means it’s working even for previously unseen objects and visual domains, without requiring custom adaptation. 📂📊 The SAM 2 model, code, and weights are being shared with an Apache 2.0 license, along with the SA-V dataset containing 51,000 videos and 600,000+ masklets. This means that both open-source and commercial use are encouraged, while ensuring credit is given to the original authors and changes are documented. An interesting resolution to concerns regarding open-source 🤔 🎬🛠️ SAM 2 has potential applications in video effects creation and improving annotation tools for computer vision systems. 👷♀️Check out what you can do with SAM 2 on the Tenyks platform by signing up for a free account: https://lnkd.in/euig_G85

What’s the link between renowned competitive shooter Lenny Basham and the Tenyks Platform? I’m afraid that’s not a joke and there is no punchline…..🤦♀️ BUT there are some great ways that the Tenyks platform aligns with Lenny’s shooting philosophy! And given that Lenny’s path led him to an Olympic gold, we’re optimistic about what this means for our customers 🥇 So what are the similarities, and why are they relevant? - Focus on Process, Not Outcome: Lanny Basham's mental management system emphasizes focusing on the process rather than the outcome. In ML, it's crucial to concentrate on the process of data preparation, model selection, and validation techniques rather than obsessing over the final accuracy from the get-go. A well-structured process using the Tenyks platform leads to better and more predictable outcomes. - Ultra Specific Goal Setting: Setting clear, achievable goals is another lesson from Basham. In ML, setting specific objectives for your projects, such as improving model accuracy by a certain percentage or reducing error rates, can provide clear direction and motivation. Specifically, picking data slices in the platfrom that have poor performance and laser focusing only on them, helps in prioritizing tasks and measuring progress effectively. - Adapting to Conditions: In shooting, adapting to wind conditions, lighting, and other environmental factors is crucial. Similarly, ML models must be adaptable to different datasets, biases, and real-world conditions. Flexibility in approach and the ability to adjust your models based on feedback and changing scenarios are vital for long-term success. In the Tenyks platform you can quickly and easily seek out images in unusual conditions and ensure your model can perform regardless. - The Importance of Practice and Repetition: Thousands of repetitions for muscle memory led to Basham’s Gold and this is not something that can be avoided in the case of ML. Model tuning is an ongoing process that needs to be repeated tirelessly. By using the Tenyks platform you can ensure that your repetitions are adding value in the most important areas. - Visualisation and Positive Self-Image: Basham talks about the importance of visualising the results, and using mental simulations to course correct. The Tenyks platform allows you to visualise distributions, and have both high level understanding of how models works, as well as specific error patterns. We may not all be able to win Gold in Paris this summer, but you will be able to benefit from the impeccable accuracy of Tenyks. Sign up to the Tenyks platform to get instant actionable insights from your data, or reach out to speak to a team member.

Llama 3.1 405B: Pushing AI Frontiers While Enabling Practical Applications While everyone's talking about Meta's groundbreaking 405B parameter model, let's focus on its real-world potential. What sets Llama 3.1 apart is its sophisticated architecture, featuring multi-lingual support, advanced quantization for efficient deployment, and cutting-edge data preprocessing. But beyond the impressive specs, let's explore what developers can actually build: - Real-time and batch inference: Enable live video analytics or process large datasets - Supervised fine-tuning: Create domain-specific chatbots or customize content generators for niche industries - Continual pre-training: Develop AI assistants that stay updated with the latest information in rapidly evolving fields (e.g., medicine) - Retrieval-Augmented Generation (RAG): Build powerful question-answering systems for complex technical documentation Developers: what groundbreaking applications do you envision building with these advanced tools? Interested in RAG, Foundation Models & Visual Prompting? Learn how these technologies are about to disrupt Computer Vision (https://lnkd.in/dPnf6xpe)

New Tenyks Team Member Alert 📣 🥳 We are overjoyed to welcome Dimitris Tagkoulis to the Tenyks fold. Dimi is combining his love of tech, sales and product to truly thrive as our Founding Account Executive. Dimi joins us with a wealth of experience in search and AI due to his impressive stint closing more than $1m ARR at Algolia, and building out a similar-sized pipeline at Workable. But Dimi brings with him a lot more than an impressive CV - he wowed us all with his integrity, honesty and quality of character throughout the interview stages and during reference calls. These are all essential traits of a Tenyksian 💪 Oh, and, did we mention? As if he doesn’t already sound cool enough, Dimi also used to be an F3 racing driver 🏎 🤩 We are confident that Dimitri will bring the chequered flag to Tenyks GTM and don’t doubt that he could get anyone revved up about the potential Tenyks has to transform your unstructured data into actionable insights - send him a line to find out more 🏁

Calling all V7 users 📣 We are super excited to reveal that Tenyks now integrates directly with V7 🎉 Tenyks and V7 are united in our missions to make AI more reliable and accurate in allowing you to harness the full potential of your visual data. Gone are the days of manually transferring between the platforms! With this new integration you will save even more time, money (and a probably few headaches!) by processing your data 10x faster, easily identifying edge cases and creating endless possibilities to rapidly search rare objects. This means shipping reliable models into production faster and with an increase on ROI of labelling efforts! Win-win 🥳 If you’re interested to turn unstructured visual information into structured intelligence, while seamlessly using tools you already love (oh, hi V7 👋) then send us a line, or book a demo on our website (https://lnkd.in/dMHKFrhA)