big momentum here.

AI, assisted programming is like a double, edged sword. It's great for speed, but I worry about the depth of understanding it fosters. Are we trading intuition for convenience? The long, term effects on code quality could be huge.

The long-term maintainability risk is real and almost nobody talks about it honestly. Code that AI generates to pass tests today becomes the hardest thing to debug in six months when the original context is gone and nobody on the team wrote it.

🤖 @mlconference connects people working with Generative AI and Machine Learning Engineering to share practical ideas and build real-world AI systems. 📅 June 1 - 5, 2026 📍 San Diego ➡️ mlconference.ai/san-diego/ 👉 Tech events in the US: devitjobs.com/events #DevITEvents

Applied Machine Learning Conference 2026 brings together builders, researchers, and innovators applying ML in the real world to share knowledge and advance AI. #AMLC2026 #MachineLearning #AppliedAI #MLConference #ArtificialIntelligence

Large Language Models (LLMs) like Claude and Grok are changing how we use AI in business. For advanced practitioners, making them work well for real-world, domain-specific tasks needs more than just prompt engineering. You need to fine-tune them, adapt them specifically, and use efficient training methods. In this hands-on workshop at #MLPrague, led by Elad Ben-Zaken and Oded Ovadia, you’ll go step by step: ▪️ Dataset preparation ▪️ Fine-tuning a pre-trained model ▪️ Evaluating results ▪️ Deploying the adapted model Everything is done in code, and you’ll be ready to use these techniques in your projects right away. 👉 Register at mlprague.com 🎟️ 1 workshop ticket gives you two sessions: one in the morning and one in the afternoon. #mlprague2026 #ai #MachineLearning #aiconferences#aiconferences2026 #mlconference #aiworkshops #prague

I’m back in Munich, the organisers of @mlconference ask me last minute if I could teach a 2-day AI course. Stupidly I said yes and then spent all weekend writing exercises. Monday done, time for a beer 🍺

Closing out the first decade of ML Prague conference with a compelling panel discussion featuring: ▪️ Stanislav Fort, Google DeepMind ▪️ Iryna Gurevych, Technical University of Darmstadt ▪️ Jon McLoone, Wolfram Research ▪️ And moderated by Jiří Materna (Scientific program & Co-Founder, ML Prague) #mlprague #mlprague2025 #mlconference #aiconference #machinelearning #AI #conference #prague #10years

⭐ Today at #mlprague - Afternoon talks ▪️ Distributed Collaborative AI with Applications to Drones Hava Siegelmann has addressed the challenges limiting drone autonomy, such as computational constraints, energy limits, and communication overload. She has presented sequence AI algorithms that improve compute and energy efficiency, enable rapid adaptation to dynamic environments, and allow the use of cheaper hardware. She has also introduced a new cooperative AI paradigm where drones act as lifelong learners, updating and peer-teaching each other without overwhelming communication needs — moving toward safer and truly autonomous systems. ▪️ How to feed your LLMs with data from the web Jan Čurn @apify has explained how to efficiently collect and prepare web data for feeding Large Language Models (LLMs) and RAG applications. He has addressed challenges like blocking, dynamic content rendering, and data quality, and has shown how to build robust web data extraction pipelines and clean HTML to avoid the "garbage in, garbage out" problem — backed by real-world application examples. ▪️ Fitting LLMs into a single GPU: Making neural networks smaller Vladimir Macko has tackled the challenge of fitting large neural networks into a single GPU by making models smaller and more efficient. He has presented state-of-the-art techniques in pruning and quantization, and has shared key insights from both academic research and industry projects. He has shown practical strategies for algorithm selection, toolchain optimization, and model evaluation to help machine learning practitioners shrink models without sacrificing performance. #mlprague #mlprague2025 #mlconference #aiconference #machinelearning #AI #conference #prague #10years

⭐ Today at #mlprague - Afternoon talks ▪️ Evaluating LLM outputs with humans and LLMs Ondrej Dusek has tackled how to effectively evaluate LLM outputs on text generation tasks. He has introduced an efficient human annotation framework and schema, and presented a new metric based on an ensemble of open-source LLMs that explains each annotated error. He showed how both methods achieve high correlation with human judgments and avoid data leakage by using fresh, unseen benchmarks. ▪️ Advances and Challenges in Topic Modeling of Text Documents Martin Neznal explored advances and challenges in topic modeling for text documents. He presented methods for improving clustering quality through better preprocessing, comparing different clustering techniques, and showed strategies to detect new clusters over time. He also addressed how to validate topic quality using both traditional metrics and LLM-based evaluation, and emphasizing the role of human feedback to refine and improve real-world topic modeling systems. ▪️ Towards Real-World Fact-Checking with Large Language Models Iryna Gurevych addressed the challenge of real-world fact-checking with large language models. She presented strategies to dismantle misleading narratives that misuse scientific publications and demonstrating how multimodal LLMs can detect misinformation based on visual content. She was also showing how to generate strong alternative explanations that counter false claims, addressing not just why a claim is false, but why it appeared credible in the first place. #mlprague #mlprague2025 #mlconference #aiconference #machinelearning #AI #conference #prague #10years

⭐ Today at #mlprague - Morning talks ▪️  Adversarial attacks on the largest language and vision models Deep neural networks — from computer vision models to massive language models — remain highly vulnerable to adversarial attacks, and there is still limited theoretical understanding and few reliable defenses against these threats. Stanislav Fort (Google DeepMind) explored the robustness of modern deep learning models, demonstrated practical examples of transferable attacks on large closed-source vision-language models, and drew connections between adversarial vulnerabilities and broader challenges in general AI alignment. ▪️  Training AI Models for Crime Scene Fingerprint Recognition Jakub Sochor (Innovatrics) addressed the challenge of training AI models for crime scene fingerprint recognition without ground truth annotations. By introducing innovative methods using synthetically generated fingerprint data, he is showing how AI advancements boost the accuracy and efficiency of latent fingerprint analysis in forensic investigations. ▪️  Understanding the neural networks through rule extraction Tomáš Pevný (Czech Technical University) is uncovering how neural networks store and process information by extracting decision rules from trained models. He is explaining why understanding these rules is difficult without knowing the data distribution, offering insights into both the robustness of neural networks and the ease of creating adversarial examples. He is also showing how decision rules compose during inference. #mlprague #mlprague2025 #mlconference #aiconference #machinelearning #AI #conference #prague #10years

⭐ Today at #mlprague - Afternoon talks (2) ▪️ Attentive interpretable models for scalable content recommendation in mobile games Martin Dlask presented a novel approach to content recommendation in mobile games using attentive interpretable models. He introduced a scale-adaptive algorithm for fair and relevant in-game recommendations and detailed the architecture of a scalable prediction system serving millions of players. He also shared practical strategies for building robust recommender systems, including tackling feedback loops, cold-start problems, and bias. ▪️ Evolution of Recommendation System: from ANN to Ensemble of Scorers Early neural network-based recommendation systems often struggled with efficiency, scalability, and interpretability, especially as data complexity grew. Raid Arfua demonstrated how moving to an ensemble of simpler scoring methods, using sparse matrix techniques and the Hyperbolic Score metric, improves performance and transparency. He also discussed building strong data platforms for rapid experimentation and proposed ideas for integrating LLMs into future recommender systems. ▪️ Estimating online behavior of ad hoc cohorts using context-dependent weighing of panel participants Estimating online behavior of ad hoc user cohorts is challenging due to biases, data sparsity, and complex interactions that traditional weighting methods cannot capture. Ariel Azia showed how to overcome these challenges by creating shared embeddings for users and cohorts and training a neural network to model their nonlinear interactions. This new approach improves the accuracy of cohort analysis and better handles real-world data limitations. #mlprague #mlprague2025 #mlconference #aiconference #machinelearning #AI #conference #prague #10years

⭐ Today at #mlprague - Morning talks ▪️Towards Production-Ready Czech LLMs with Continuous Pretraining Deploying large language models in a specific linguistic context like Czech presents unique challenges, especially when it comes to data quality and model alignment. Ondřej Filip presented how Seznam.cz tackles this by leveraging continuous pretraining strategies to develop Czech-centric LLMs that are ready for real-world applications. ▪️ Data, your worst enemy? In machine learning, we often assume that more data means better models. But data can also be a source of unexpected pitfalls—bias, instability, and even misleading outcomes. Johan Loeckx explored how engineering decisions shape the way we collect and interpret data, and how these choices can undermine the fairness and reliability of AI systems. ▪️ Lies, Damn Lies and Gen AI Generative AI can produce human-like text—but it can also confidently generate falsehoods, leading to serious consequences in trust and decision-making. Jon McLoone showed how to critically assess and detect the distortions created by generative models, arming attendees with tools to evaluate AI-generated content. #mlprague2025 #mlconference #aiconference #machinelearning #AI #conference #prague #10years

The organizer of #MLPrague, @JiriMaterna, opened the conference this morning saying, “We’re celebrating our 10th anniversary with our community of ML and AI enthusiasts from over 30 countries!” #mlprague2025 #mlconference #aiconference #machinelearning #AI #conference #workshops #prague #10years

ML Prague 2025 kicks off today with 11 lectures! We can't wait to celebrate our 10th anniversary together with our international community of machine learning and AI enthusiasts here in Prague. Follow all the main highlights of the conference on LinkedIn, X, and in our official app, @eventeeco #mlprague #mlprague2025 #mlconference #aiconference #machinelearning #AI #conference #workshops #prague #10years

Our workshop day has just finished! Now it's time for sharing the best takeaways from the sessions. ▪️ 3D reconstruction from Images and their application We got hands-on with 3D computer vision! We captured images, built datasets, and turned 2D photos into detailed 3D models using techniques like NeRFs, Gaussian splatting, and multi-view stereo. Really cool to see how modern 3D reconstruction works! ▪️ A practical guide to LLM-based AI agents We learned how to design and build LLM-based AI agents using real frameworks. They explored agent architectures, customized their own agents, and saw how these systems are already transforming industries with the power of generative AI. ▪️ Synthetic Data Generation for Embedding Model Fine-Tuning We learned how to tackle retrieval challenges in non-English and domain-specific languages by generating synthetic data and fine-tuning open embedding models. They gained hands-on experience building stronger, more specialized retrieval systems. ▪️ Parallel Genetic Algorithms in Python We learned how to build and apply parallel genetic algorithms (PGAs) in Python, explored parallel programming basics like locks and shared memory, and tackled real-world case studies from maze solving to quantum cryptography and neural network optimization. ▪️ Real-Time Anomaly Detection and Alerting in Financial Markets Using Stream Processing We learned how to build real-time stream processing systems to calculate stock indicators like RSI, MACD, and Bollinger Bands, detect anomalies, and trigger instant alerts or trading actions — plus how to integrate ML models to enhance decision-making. 👉Last conference tickets at mlprague.com #mlprague #mlprague2025 #mlconference #aiconference #machinelearning #AI #conference #workshops #prague

A morning full of actionable takeaways at ML Prague 2025 workshop day! ▪️ Utilizing Large Language Models for improved anti-tracking in web browsers: We learned how to build an LLM-powered anti-tracking system that analyzes network requests, page content, and user behavior in real time, improving detection accuracy while tackling the challenges of running efficiently in a web browser. ▪️ Beyond Real-World Limitations - Mastering Synthetic Data Generation for Enhanced ML Performance: We dived into how AI agents and LLMs can create high-quality synthetic data, helping ML teams tackle data gaps. Attendees learned hands-on techniques to boost data quality and rethink how they build datasets for better ML results. ▪️ Introduction to Algorithmic Trading: Hands-On Strategy Implementation with Real-World Data: We built our own trading strategies today! Learned how to prep real market data, train deep learning models, backtest them, and deal with real-world stuff like slippage and fees. Super hands-on — felt like working on actual trading systems! ▪️ InstructLab: plug your knowledge into a model easily: We got hands-on with InstructLab! We learned how to extend LLMs with our own data, generate synthetic training data, and even trained and chatted with models right on our laptops. Super cool to see how easy it is to run everything locally! ▪️ Accelerating AI Through Human Knowledge: Teaching to Imitate Experts and Win on the Race Track: We got into Imitation Learning today — teaching AI to learn by copying experts instead of trial and error! It was super cool to see how it speeds up training and makes models way smarter for real-world stuff like robotics, gaming, and self-driving cars. 👉 Last conference tickets at mlprague.com #mlprague #mlprague2025 #mlconference #aiconference #machinelearning #AI #conference #workshops #prague

‘Your Ultimate Guide to the Top 25 Conferences on Artificial Intelligence (AI) in 2025’ includes International Conference on Machine Learning (ICML) See exceptionalagility.com/blog/… for details and the full list #AI #ML #MLConference @icmlconf #MachineLearning #ICML #ICML2025 #Tech

I am excited to hold a talk about In-Context Learning, with inspiration from the “In-Context Learning” or: How I learned to stop worrying and love “Applied Information Retrieval" paper from the @UofGlasgow! This is a promising way to integrate large language models to improve the quality of completing specific tasks without hallucinations. Also, getting started does not take much, particularly when using @vespaengine! Today, at @mlconference in Brooklyn, New York: mlconference.ai/machine-lear…

Cyber Valley is co-hosting the European Robotics Forum 2025! The conference will take place on March 25–27, 2025 in Stuttgart, Germany 👉 cyber-valley.de/en/news/euro… #AI #ML #MachineLearning #Robotics #Conference #MLConference #ERF #ERF2025 #RoboticsConference