วีดีโอ

Xinyi Ling presents #eCeLLM which comes in 3 sizes (S, M, L). As you guessed it can be used to boost Search, Recommendation, and product-based Question-Answering. Moreover, eCeLLM tackles the so-called "Cold-Start Problem" in E-Commerce. Cold-Start Problem is the challenge that there will always be new users and new products where limited data exist. This means we need models that have a very s...

Which is the best #LLM ? let's have them fight it out in the #Chatbot Arena! The one that scores the best in "human satisfaction" wins! But how to measure this subjective metric? Anastasios Angelopoulo presented his work to me. The core idea is simple. 🥊 Humans are presented with the outputs of two different LLMs for the same prompt and have to decide which one performs better. All votes are ag...

70 billion, 1.7 trillion,... AI models have too many parameters. Do you really need all of them for your specific use case? "No" says Avrajit Ghosh from Michigan State #University. Especially neurons in the first model layers can often be removed to drastically shrink the overall #model size without loss of performance. Enjoy the video as much as I did (multiple times by now). Paper: "Optimal E...

We run out of data - maybe already in 2026. But we can do something about it. Pablo V. from Epoch AI shows that the amount of available training data will become a bottleneck in LLM training. This is because even though humans produce more and more data, the demand for data grows much faster. That said, Pablo V. proposes solutions to overcome this challenge. 1. The model architectures should ch...

From: Mila - Quebec Artificial Intelligence Institute, McGill University, Charles University Soon we all talk to AI Agents to achieve any task. But how far away are we actually from this futuristic world? It is good that we have some scientists like Xing Han Lu creating benchmarks for agentic models to make this zoo of agents a bit more transparent. Even though agents are often advertised as ma...

This project describes how you can build a mobile real time GPS tracker with integrated Google Maps. I began this project mainly to see if I can integrate all the different pieces of hardware and software to make a workable solution, and it took some time, but finally when everything was said and done, it looked pretty cool. I tore down everything and rebuilt it from scratch, making detailed no...

Tools and Frameworks Frameworks: Use TensorFlow, Keras, or PyTorch for implementation. Visualization: Consider using libraries like Matplotlib or Seaborn for visualizing training progress and results.

Using Large Language Models (LLMs) with the WhatsApp Cloud API can enhance communication through automated responses, chatbots, and customer support. Here’s a general outline of how to integrate LLMs with the WhatsApp Cloud API..