Hamza Tazi

One more intresting presentation of Paul Tremblot de la Croix "How Generative AI Can Help Startups Succeed" at the #VDS2024. And my note on how AI can drive success: 💬 speakers emphasized that data is the foundation. Generative AI applications are only as good as the data they're built on. The ability to combine real-time data with generative capabilities is a game-changer for startups. 🕵‍♂️ shared the importance of customization. Don’t adopt generative AI just because it's trendy. Tailor it to your business needs, ensuring it’s the right tool for your use case. 👮‍♀‍ discussed data privacy and security concerns, emphasizing GDPR and other regulations. Startups must handle customer data carefully and integrate strong security protocols when deploying generative AI tools. Key points: 1. Generative AI + RAG = Personalization. By using AI with RAG, startups can retrieve relevant real-time information or past interactions, ensuring personalization not just based on predictions but also on actual preferences, history or external knowledge sources. 2. Customization & Use Case Fit. It's crucial to pick the most efficient AI model, considering cost, latency and accuracy. Generative AI isn't one-size-fits-all! 3. From MVP to Production. Building an AI-powered product is just the first step. Measuring, tracking and fine-tuning AI in production is essential for maintaining customer trust👌 I appreciate your meaningful insights shared about scaling AI strategically. Also on keep security tight and focus on solving real customer problems and think about there needs 😇 #AI #Startups #VDS2024 #DataPrivacy #Cybersecurity

4

You MUST see this. TL;DR We've reached a true breakthrough in clinical AI, developing CARE1 - the world's first clinical foundation model with an application submitted to FDA review. Truthfully this is just the tip of the iceberg, we've used a fraction of our data and two years from now we'll see these as the first baby steps. You can read more about it here: https://lnkd.in/drhPYTtX Shameless plug - this recent breakthrough is just one example of the impactful work our teams do every day. We're looking for highly experienced algorithmic experts, both ICs and tech leads, to join us and take a pivotal role in this and other high impact projects. Feel free to reach out to me directly. This is first and foremost thanks to the amazing team working on this effort - Itay Katzir the mastermind who's been planning this for years building paradigm after paradigm to enable this. From the algorithmic engine, through NLP, through scalable training pipelines and more. Ariel Persiko - truly a one man algo team, the exact person you want on this type of breakthrough project. Dor Alon - AI Infrastructure expert, built components that were critical to this endavour. And of course some of our amazing leaders who were deeply involved in invaluable strategic guidance - Uri Goren, Ronny Levanda, Tamar Lin. Not less than that - I think this is a true lesson of how good founders really move the needle even at a 450 person company. Elad Walach our CEO was maybe the first person who started pushing strongly for us to invest in this, and Michael B. Braginsky our legendary CTO who continues to amaze me even after 8 years with his ability to inspire us with unachievable goals which we are then able to achieve.

72

1 Comment

🚀 New Post Alert 🚀 To ASK’EM and beyond (1 of 2) A path to scalable, data-driven decisions I recently came across an article by Mary K. Pratt featuring ASK'EM—an embedded product survey system I created at Indeed to give product teams targeted, actionable insights that directly connected to user needs. ASK'EM was a three-and-a-half-year passion project, so seeing it recognized as a case study was one of those 🤯 moments! Today, I’m building on that foundation at Splunk. We’re using my AAIM UP with U3 approach to deliver products that are Useful, Usable, and Ubiquitous (U3) and aligned with Upward Potential(UP)—creating a continuous loop of Monitoring, Improving, Alerting, and Attributing (AAIM) to keep product decisions relevant and impactful. Interested in how these approaches can turn data into a powerful growth engine for your products? Check out my latest post! Here’s a preview of my guiding principles: 🌟 Build on a durable foundation that targets what matters most 🌟 Align goals and metrics with customer needs and business priorities 🌟 Meet customer needs with intuitive, adaptable products ‘Til next time, I’m Bianca 👉 Read the full post here  https://lnkd.in/eZuvi25B 👉 Read about my approach here  https://lnkd.in/ezpcvjdj 👉 Read the original article by Mary K. Pratt here https://lnkd.in/eUY6EaZx #DataDrivenDecisions #MeasureWhatMatters #ASKEM #UserFirst

11

I'm currently going through some very old SnagIT screenshots from 2015. I'm trying to recall which hype cycle this is from. Which hype cycle was relevant in 2015? #SelfServeAnalytics? Had we reached #BigData yet? The progress in "#reusability" has been remarkable for the cartoon industry. 😂 #SameHypeDifferentDay

2

While you probably don't have to worry about aliens invading on the fourth of July, you should be aware that text embeddings may expose your private information. Fortunately it's an addressable risk. Our Senior AI Scientist, Joseph Ferrara, PhD, covers how to mitigate the problem on a pdf. Broadly, the steps are: -Extract the text -De-identify the sensitive info -Chunk the resulting text -Embed it in a Pinecone database -Perform some test queries If you want to see more detail and sample code, check out the how to guide the comments. #RAG #AI #embeddings

15

3 Comments

In the current landscape, we've witnessed significant advancements in Large Language Models (LLMs). Precise assessment of these models is crucial for model selection, performance evaluation, and identifying and addressing biases. Let's talk about BLEU (Bilingual Evaluation Understudy), a widely used method for performance evaluation.  This metric, an extrinsic evaluation tool, compares model-generated outputs with one or more reference texts to gauge similarity. Scores range from 0 to 1, with higher scores indicating superior performance. This approach proves invaluable for evaluating the performance of translation models. Scores below 10 are negligible, while those exceeding 50 denote good translation quality. The BLEU formula comprises two key components:  1. Brevity Penalty: Penalizes overly short-generated translations compared to the reference length using an exponential decay 2. N-Gram Overlap: Assesses the match between unigrams, bigrams, trigrams, and four-grams in the generated output and the reference translations. To prevent overcounting, the n-gram counts are capped at the maximum occurring in the reference. This overlap is calculated as a precision-based metric, with the geometric mean of these four precision values being taken. Important considerations when utilizing BLEU for evaluation: 1. BLEU is corpus-based: It's not suitable for assessing individual sentences. 2. Lacks differentiation between content and function words: For instance, removing a function word like 'the' incurs the same penalty as replacing a significant term like 'Earth' with 'Mars'. 3. Limited in capturing sentence meaning: It only considers up to four-grams and overlooks long-range dependencies.

14

Is an individual Enterprise LLM License a justifiable expense for your team members? I've seen this logic circulating: "If someone earning ≥$62,400/yr can save just one hour in a month using a #GenerativeAI tool, it justifies the cost of a $30/month #EnterpriseLLM." This assumes time directly equals productivity, and that a user doesn't already have an hour+ of free time in a month. We know this isn't always the case. Two quick factors to look at first: 1. The litmus test: Will time saved be reinvested into productivity? Is that productivity worth more or less than an hour of time on paper? 2. This simple formula: S>=max(C/H,F) In words: The number of hours saved (S) must be greater than or equal to either C / H (cost divided by hourly rate) or F (existing free time), whichever is larger. For the record, if we're using these tools effectively, an hour saved each month for is on the extreme low end of what I would expect for most business roles...

8

9 Comments

We're thrilled to announce that Epsilla (YC S23) now supports Claude 3.5, the latest LLM by Anthropic that’s set to outperform GPT-4o! 🚀🤖 How does Claude 3.5 really perform against GPT-4o? We've recorded a fun video where Claude 3.5 and GPT-4o go head-to-head, tackling the question: Compare GraphRAG with traditional RAG, grounded by 361 research papers on Retrieval-Augmented Generation published since 2024. Curious about which model reigns supreme? Watch the video and join the conversation! For more in-depth insights, check out the detailed comparisons here: Claude 3.5 Insights: https://lnkd.in/gYQBKqrz GPT-4o Insights: https://lnkd.in/gyq38cwA Which one do you think is better? Please share your thoughts in the comments. #Epsilla #Claude35 #GPT4o #AI #ML #Anthropic #LLM

51

My intern at Schneider Electric and student at Northeastern University, Udisha Dutta Chowdhury, is presenting our work at PyData NYC tomorrow! Her talk is titled "Adopting Open-Source Tools for Time Series Forecasting: Opportunities and Pitfalls".  In addition to introducing sktime and skforecast as case studies, she will also guide the attendees through the broader considerations of tool selection for forecasting. We will emphasize the importance of Data Understanding, Data Preparation, and Backtesting in building effective and reliable forecasting pipelines. If you are in the conference, please come say hi! The talk will also be posted on YouTube later. #techtalk #algorithms #forecasting #machinelearning #datascience

27

1 Comment

I’m thrilled to collaborate with Josep from DataBites to share a comprehensive guide of top resources in ML. Josep Ferrer is a veteran technical writer of machine learning having published more than 50+ articles in Kdnuggets and towardsDataScience. Together, we've curated an essential list of resources to help you build a strong foundation in ML. The guide is divided into two parts. In Part 1, we share key theoretical resources: • Top-rated online courses • Must-read books. • Popular podcasts and tutorials to get you started. These high-quality resources are designed to equip you with the knowledge and confidence to tackle the challenges ahead in your ML journey. Checkout the article here: https://lnkd.in/gVDS_Uaw Stay tuned for Part 2 on Saturday, where we'll dive into practical, hands-on learning resources, some classic projects, and advanced techniques to help you apply your knowledge and build a standout portfolio. Don't miss out! Check out Part 1 now and take the first step. -- 𝘐𝘧 𝘺𝘰𝘶 𝘭𝘪𝘬𝘦𝘥 𝘵𝘩𝘦 𝘢𝘳𝘵𝘪𝘤𝘭𝘦, 𝘢𝘭𝘴𝘰 𝘤𝘰𝘯𝘴𝘪𝘥𝘦𝘳 𝘴𝘶𝘣𝘴𝘤𝘳𝘪𝘣𝘪𝘯𝘨, 𝑙𝑖𝑛𝑘 𝑏𝑒𝑙𝑜𝑤 𝑚𝑦 𝑝𝑟𝑜𝑓𝑖𝑙𝑒 𝑝𝑖𝑐𝑡𝑢𝑟𝑒 ☝️ #MachineLearning #LearningJourney #MLResources

69

20 Comments

I wrote a blog post on a thorny but underappreciated problem in ML-based marketing personalization: event attribution. It's how you go from the actions you take and customer behavior you observe to creating training experiences for your model. Good attribution makes a huge difference in driving uplift, there are a lot of degrees of freedom in configuring it, and at the same time, it's rarely discussed in ML/bandits/RL literature.

47

4 Comments

LangChain's modularity and flexibility enable the seamless orchestration of various LLM components, making it an ideal framework for constructing intricate conversational flows. Paired with LangGraph's visual design capabilities for managing these flows, development processes are notably more intuitive and streamlined. Read my take. #machinelearning #LLM #Langchain #Langgraph

13

AI teams I've spoken to are spending way too much money building massive internal benchmarking datasets for their product before releasing to a single user 🙁 Of the hundreds of AI engineers we spoke with, the most effective ground truth datasets were only between 50 - 150 examples in size. Common features? Small quantity, high quality, and extremely iterative. There were only two companies that had ground truth datasets with 1000+ examples, both of which threw it all away once real users started trying the product. “We threw our benchmarking data away after we saw real user queries” - AI Engineer, Tech Consulting Firm ❗ Benchmarking datasets are inherently flawed because they NEVER correlate with how users actually use your AI agent product. They also change as new user behaviors, requirements, and product features are implemented. Therefore, don't overthink it. The best practice we’ve seen is deploying early and collecting real user queries and responses over time as part of a continuous benchmark creation process. The dataset is a living thing (just like any product), and spending too much time crafting the perfect benchmark set before releasing is wasted effort. Another emerging practice is using evaluation models (model as a judge) to give first pass heuristics on performance. We're really excited about this approach because evaluation models are highly flexible, and decrease the cost/reliance on large benchmarking datasets. 🚀 Let me know if this experience resonates! Twitter: https://lnkd.in/eAsQvtiu

24

2 Comments

As the CEO of an AI-driven health-tech company, Labops, I’m constantly evaluating how technology can create the most impact. While Large Language Models (LLMs) are making headlines, I believe Small to Medium-sized Language Models (SMLs) deserve more attention—especially for businesses operating in niche domains like healthcare. Why do I say this? 1️⃣ Tailored Solutions: In healthcare, where precision and specificity are critical, SMLs (40–50B parameters) are more adaptable to domain-specific needs. 2️⃣ Cost-Effectiveness: For startups, running and fine-tuning a SML is more economical than relying on large, resource-intensive LLMs. 3️⃣ Ownership & Flexibility: Open-source SMLs allow businesses to build and deploy models on their terms, giving greater control over data and performance. At Labops, where we’re building solutions to transform diagnostic services, the ability to leverage AI that aligns with our unique requirements is crucial. SMLs strike the right balance between scalability and practicality, making them a natural fit for startups like ours. LLMs have their place, but they’re not always the answer for every use case—especially in sectors where focus and efficiency matter more than sheer size. If you’re in healthcare or another specialized industry, I encourage you to explore the potential of SMLs. They might just be the competitive edge you’re looking for. What’s your perspective on SML vs. LLM? Let’s discuss! 💬 #AI #HealthTech #LLM #SML #Startups #Innovation #Labops

5

1 Comment

"Data democratization" (an iconic buzzphrase) has been around for a long time. Tons of companies have promised to deliver it... with mixed results. Lots of dashboards, heavy implementation, data siloes, the list goes on. But advancements in AI have made "simple data access for all" a reality. Check out Bessemer Venture Partners's Data Shift Right roadmap. Thanks for the shout out Lindsey Li, Yael Schiff, and Amit Karp.

14