Read with me
This "read with me" release is a bit techy - just more interesting things happened in the world of tech.
AI and Tech
Ai2 says its new AI model beats one of DeepSeek’s best
Everyone was caught by surprise by DeepSeek performance and how cheap it was a week ago. Now we have Ai2 model (this time from US), that is also fully open source and claims to beat DeepSeek V3 (which is not really open source). That is awesome to see such a rapid development in this space.
Everyone was caught by surprise by DeepSeek performance and how cheap it was a week ago. Now we have Ai2 model (this time from US), that is also fully open source and claims to beat DeepSeek V3 (which is not really open source). That is awesome to see such a rapid development in this space.
The 2025 AI Engineer Reading List
A huge collection of 50 papers that AI Engineer must read. And yeah, Attention is All You Need is not in the list. This tells a lot. Another good source to stay up-to-date with latest AI developments is this git repo with 21k stars: Hannibal046/Awesome-LLM
A huge collection of 50 papers that AI Engineer must read. And yeah, Attention is All You Need is not in the list. This tells a lot. Another good source to stay up-to-date with latest AI developments is this git repo with 21k stars: Hannibal046/Awesome-LLM
Microsoft introduces CoRAG: An AI Framework for Iterative Retrieval and Reasoning in Knowledge-Intensive Tasks
An interesting research from Microsoft Corporation and the Renmin University of China introduce a method for training RAG models to iteratively retrieve and reason before generating answers. Unlike conventional RAG systems, CoRAG dynamically reformulates queries based on the evolving reasoning state. Read full paper here. I would see CoRAG boosting Agentic AI quality dramatically.
An interesting research from Microsoft Corporation and the Renmin University of China introduce a method for training RAG models to iteratively retrieve and reason before generating answers. Unlike conventional RAG systems, CoRAG dynamically reformulates queries based on the evolving reasoning state. Read full paper here. I would see CoRAG boosting Agentic AI quality dramatically.
AI Mistakes vs Human Mistakes
AI agents and large language models (LLMs) indeed introduce a new category of errors that our current systems are not fully equipped to handle. Despite their advanced capabilities, LLMs can sometimes produce surprisingly flawed outputs. This raises significant concerns, especially when such mistakes could potentially lead to harm for humans. Additionally, I am skeptical that reinforcement learning alone offers a sustainable or scalable solution to address these shortcomings in LLMs.
AI agents and large language models (LLMs) indeed introduce a new category of errors that our current systems are not fully equipped to handle. Despite their advanced capabilities, LLMs can sometimes produce surprisingly flawed outputs. This raises significant concerns, especially when such mistakes could potentially lead to harm for humans. Additionally, I am skeptical that reinforcement learning alone offers a sustainable or scalable solution to address these shortcomings in LLMs.
Yandex Perforator: Open Source Performance Tool Can Save $Millions
Yandex Perforator is a continuous profiling tool that helps businesses optimize application performance and reduce costs. By using real-time data, it identifies bottlenecks and improves efficiency, leading to faster development, better resource utilization, and improved user experience. This translates to cost savings through reduced development time, lower infrastructure costs, and increased revenue.
Yandex Perforator is a continuous profiling tool that helps businesses optimize application performance and reduce costs. By using real-time data, it identifies bottlenecks and improves efficiency, leading to faster development, better resource utilization, and improved user experience. This translates to cost savings through reduced development time, lower infrastructure costs, and increased revenue.
Product and business
Prioritize revenue over growth
"First time founders in particular seem to really buy into the winner-take-all marketplace dynamics. But I think that approach is wrong — it was wrong even ten or fifteen years ago, and it's especially wrong now."
"First time founders in particular seem to really buy into the winner-take-all marketplace dynamics. But I think that approach is wrong — it was wrong even ten or fifteen years ago, and it's especially wrong now."
Nobody cares
For me this article is about passion. Engineers, product managers, founders passionate about something can do wonders and change the world. People caring about society beyond the bubble are making the world a better place.
For me this article is about passion. Engineers, product managers, founders passionate about something can do wonders and change the world. People caring about society beyond the bubble are making the world a better place.