Felipe Augusto Chies

"If you are the smartest person in the room, then you are in the wrong room." - Confucius I was definitely in the right room of Data Science Leuven accompanied by three wise people. First of all, I had the pleasure to meet Nikki Bueno de Mesquita, data product manager in practice, probably one of the first with this job title, providing a lot of practical experience and down-to-earth opinions on data product management. Next to Nikki, our panel discussion was moderated by Jules de Bruin: who better to interview people on data products than someone who gets paid to perform research to data products on a daily basis? On the other hand, his job was made easy by the very engaged audience bringing up a lot of great questions. And then last, but not least, you had Jean-Georges Perrin, a fellow Data Mesh Learning Community MVP who doesn't need introduction, but still I will honor him with a shout out as being one of the driving people behind Bitol, open standards for data engineering, starting with data contracts. Recording will be shared soon and I promise to add an extensive summary to my blog, but I have some takeaways that are too good to not already share: 🏠🔑 A data product is the combination of data with all that you need to use it. As such it brings technical and data ownership together. Data products emerge for many reasons, ownership is definitely one of them. Data contracts on the other hand, rise to increase maturity. ❌ Data products, compared to data sets produced by a central team, give you the right to fail. Never looked to it like that, but JGP is right. A product is never perfect from the start: you iterate on the product, over and over, to make it more desirable and usable. Centrally created datasets were assumed to be the truth and quite often, being a bottleneck, the central team had too much work to quickly iterate and fix mistakes. Being a fan of "fail fast, learn fast", this again strengthens my believe in product thinking for data. 💰 It's all about value. Both Jean-Georges and Nikki, and myself, stress that a data product should have a purpose, a reason of existence, tied to some value for the company. But Nikki describes it best: "If I don't succeed getting value out of a data product in 3 months, I should shift focus, as otherwise I won't get any more funding". Got to love scale-up mentality in companies like Otrium, but they are right! Focus on what brings value to your organization. Even though I was on a panel, I feel like I have learned more than I could share, so many thanks for the learning opportunity! #datamesh #dataproduct #datacontract

55

6 commentaren

Loving this data: In Spain, Amazon has created more jobs than any other company in the past five years 👷‍♂️ 👷‍♀️ In fact, AWS plans to invest €15.7 billion over the next decade, expanding its AWS Europe (Spain) Region in Aragón 🚀 This investment will support an estimated 17,500 full-time equivalent jobs in local businesses on an annual average basis, and will contribute an estimated €21.6 billion to Spain’s gross domestic product (GDP) through to 2033.

51

Learn how to calculate the number of tokens in your text input before sending requests to Gemini with the Vertex AI SDK for Python, as explained by Laurent Picard. #generativeai #vertexai #googlecloud #largeLanguageModels #Gemini https://lnkd.in/eyU6xRzV

1

VentureBeat writes about Microsoft's Phi-3 model enabling new generative AI applications on-device with its high quality outputs despite the relatively small model size https://lnkd.in/eqR3swC4. #microsoft #phi3 #generativeai #largelanguagemodel #artificialintelligence #venturebeat

6

❓How to create the #genai agent? 🤘 Use #langchain and create tools! ❓Anything else? Maybe something more human friendly? 🤘Use AI Agents for Amazon Web Services (AWS) #bedrock! ❓What If... I wan't to become smarter? Use 𝐅𝐋𝐀𝐏! (flow-adhering planning) 🔈 Ok, what is it and why you are bothering me with such a thing? Well, in a shortcut. Everything we put into LLM (in particular autoregressive transformers) need to be at first encoded (embeddings, tokens) with some attention logic. The process of selecting the best output in response to provided input (prompt) is so called decoding. There are many strategies to find the most appropriate response (e.g. token) and the most popular one is greedy search (token with highest probability is chosen) - which is the default one in many APIs we use nowadays ("generate()" function). There are also methods called nucleus sampling, beam decoding, temperature/random sampling (introduce some randomness in generated text)....and ...FLAP is the new one proposed by Amazon Science. It is an alternative decoding strategy, which uses a scoring system that encourages selection of tokens that adhere to the API and workflow dependency graphs (let's call it an agent workflow). Just Read More 😊: https://lnkd.in/drYhYkkR #aws #science #ai Tameshi

11

4 commentaren

🚨 Do you know how to create real-time #deepfakes? 🚨 One of the techniques used is #FaceSwapping! In 2021 and 2022, my colleagues Samuel Ménard, Liyun He Guelton, and Johan Maes worked on identifying machine learning-based solutions to create and detect these kinds of deepfakes. Fast forward to today, the techniques and solutions for Face Swapping have evolved significantly and are easily accessible! I recently created this video that showcases the ease of creating #deepfakes. What surprised me the most was being able to achieve this in real-time on my current PC (with a simple Nvidia graphics card), there were still some delays between the audio and visual synchronization. Even if tools are available for extracting facial images from videos, manually defining face outlines, and creating Digital Face Manipulation (DFM) models, creating high-quality #FaceSwapping models may require a significant time investment: just for a single face, you must let the GPU work for at least several days, or weeks! What do you think about this video? Can you spot a deepfake? Leave me a comment! Acknowledging the advancements in anti-spoofing solutions is crucial to maintaining #Trust in #Digital. As technologies continue to progress rapidly, it's vital to stay ahead and identify innovative solutions against deepfakes. #Worldline #GenAI #DeepfakeTechnology #RealTimeDeepfakes #MachineLearning #DigitalSecurity.

48

4 commentaren

I'm very happy to announce that 6 of our open-source models are now listed on the Amazon Web Services (AWS) Marketplace. In just a few clicks, AWS customers can deploy and run them on Amazon SageMaker in 16 AWS regions (Americas, Europe, Asia Pacific). 🔥 Arcee Nova 72B 🔥 Llama Spark 8B 🔥 Arcee Spark 7B 🔥 Arcee Argent 7B 🔥 Arcee Scribe 7B 🔥 Arcee Lite 1.5B Follow these simple steps: 1️⃣ Go to the Arcee.ai Marketplace page: https://lnkd.in/e4KYsTq5 2️⃣ Find the model you want to deploy and open the model page. 3️⃣ Read the model information, and subscribe to the model (they're all free): this should only take a few seconds. 4️⃣ Once your subscription is complete, select a built-in deployment option or, preferably, clone our Github repository (https://lnkd.in/e_wqdERZ) and run the sample notebook mentioned in the Marketplace listing. That's it! You'll be testing our models in minutes in your AWS account and without having to provision any infrastructure or write any code. As the model endpoints are compatible with the OpenAI Messages API, you'll be able to reuse your existing OpenAI prompts out of the box 😉 PS: https://lnkd.in/e_wqdERZ includes additional AWS developer resources, such as notebooks and CloudFormation templates to deploy our models from the Hugging Face hub, from S3 artifacts, or from your own model packages. Plenty of options to enjoy state-of-the-art open source small language models! #ai #opensource #slm

107

5 commentaren

What about developing an AI-powered API using Generative AI? Now posting an English version of an article I published some days ago in Portuguese, talking about the experience of developing software using AI tools such as Amazon Q Developer, AWS Bedrock, and Amazon Titan from Amazon Web Services (AWS). In this article, I bring an overview of this process and soon I will publish a video showing the step-by-step of this development.

3

1 commentaar

Are you curious about Generative AI ? Google Cloud Skills Boost offers a hands-on course to get you started with Generative AI, no experience needed !! Most important !! This course if FREE .. Now how cool is that 😎 Why spend time on this course ? - It gives a good and quick overview of Large Language Models (LLMs) and the core technology behind Generative AI. - Free Hands-On Labs to understand and experiment using the different products and services available on GCP - You earn a free badge to showcase it on your profile https://lnkd.in/difWMXXk https://lnkd.in/dcetFQ84 #generativeai #googlecloud #machinelearning #ai #cloudcomputing #freelearning #gcp

5

Working in AI at Google, we're witnessing firsthand the unfolding AI revolution. To demystify the practical applications of generative AI I've joined forces with Floor Eigenhuis, Luc de Jager and Mohamed Fawzi to create an immersive narrative that follows a fictional cloud engineer through a typical day to discover how generative AI is reshaping the professional landscape. Join us at the Google Cloud Summit in Amsterdam on Thursday to meet "Aimee," a cloud engineer, as she leverages AI to: 📧 Efficiently manage her inbox and tasks. 🤖 Extract insights from videos using Gen AI. 🔒 Protect sensitive data with AI-driven security measures. 📊 Evaluate and compare the performance of different Gen AI models. 🔍 Build a text and image search application using vector embeddings. #GoogleCloudSummit #GenerativeAI #AICloudEngineer #Amsterdam

49

2 commentaren

Google Cloud Vertex AI is rolling out multiple updates to make AI accessible for every enterprise. Up to 85% lower input costs and 80% lower output costs, and 100+ new languages coming to Gemini 1.5. #GoogleCloud #VertexAI #Gemini

2

1 commentaar

🌐 🌍 Microsoft has announced a monumental investment of 14.7 billion reais ($2.7 billion) to enhance Brazil's cloud and AI infrastructure over the next three years. This huge commitment not only accelerates AI innovation in Brazil but also signals a transformation for the entire region. As we see rapid advancements in technology and artificial intelligence across LatAm, it’s clear that the market is growing—but there are still countless opportunities to explore! We are excited to see the positive impact this will have across Brazil and Latin America! 🌍💡 https://lnkd.in/gMThuFSe

5

And if you got interested about AI agents after reading my post, make sure to take a look at Semantic Workbench, a new experimental tool from the Semantic Kernel team to streamline testing and debugging AI agents https://lnkd.in/dcSNHYBS #ai

19

How can your retail business use AI to drive new business value? To find out, read this report—which summarises the retail highlights from #GoogleCloudNext '24: https://lnkd.in/eSFjfHHW

3

#MLOps #GenAI You want to implement AI - and do it securely, and with confidence in the outputs. Then you might want to have a look at this great and well structured talk from my colleagues Mikhail Chrestkha and Chase Lyall. If you're an expert, check your work. If you're getting started - start here!

9

Explore the innovative GitHub Copilot, your coding companion that empowers developers, including those with disabilities, to code more efficiently. Alexandre Costa, a Microsoft Brazil employee with a visual disability, demonstrates how Copilot simplifies coding tasks, making technology more accessible and inclusive. Share your Copilot experiences below! #InclusiveCoding #Copilot Learn more:

3

Capgemini leading the industry with innovation in acceleration using Generative AI!! Learn more about it from Weiwei Feng below 👇 #genaigovernance #acceleratewithgenai Luc Ducrocq Prasad Shyam

17