May 2026

Securing AI agents: How AWS and Cisco AI Defense scale MCP and A2A deployments

Securing AI agents: How AWS and Cisco AI Defense scale MCP and A2A deployments

⋅

Model Context Protocol (MCP) adoption has accelerated rapidly since its introduction in November 2024. Enterprises now manage dozens to hundreds of MCP servers—tools that extend AI agent capabilities by connecting them to external data sources and APIs. The Agent-to-Agent (A2A) Protocol followed in April 2025, enabling autonomous agents to communicate directly without human intervention. More recently, Agent Skills emerged across enterprise infrastructure. This growth has created three security gaps: teams lack visibility into which tools and agents are […]

Ver mais

Liked Liked
Fine-tune LLM with Databricks Unity Catalog and Amazon SageMaker AI

Fine-tune LLM with Databricks Unity Catalog and Amazon SageMaker AI

⋅

When you fine-tune large language models (LLMs) with Amazon SageMaker AI while using Databricks Unity Catalog, you might face unique challenges like how to maintain strict data governance while using best-in-class machine learning (ML) services. Unity Catalog governs metadata and permissions, while the underlying data resides in Amazon Simple Storage Service (Amazon S3) when you choose AWS as the cloud environment for their Databricks Workspace. When SageMaker AI Training job accesses that data, you must preserve and not […]

Ver mais

Liked Liked

Force-Aware Neural Tangent Kernels for Scalable and Robust Active Learning of MLIPs

⋅

Active learning for machine-learning interatomic potentials (MLIPs) must address several challenges to be practical: scaling to large candidate pools, leveraging energy-force supervision, and maintaining robustness when candidate pools are biased relative to the target distribution. In this work, we jointly address these challenges. We first introduce a linearly scaling acquisition framework based on chunked feature-space posterior-variance shortlisting. By avoiding materialisation of the candidate and train set kernels, this approach enables screening of ~200k structures within hours and applies […]

Ver mais

Liked Liked

Human-level performance via ML was *not* proven impossible with complexity theory [D]

⋅

Van Rooij, Guest, Adolfi, Kolokolova, and Rich claimed to have proven that AGI via ML is impossible in Computational Brain & Behavior in 2024. The basic idea was to try to reduce a known NP-hard problem to the problem of learning a human-level classifier from data. The purported result, called “Ingenia Theorem” by the authors, made some noise on the internet, including here. My paper showing that the proof is irreparably broken is now also out in CBB […]

Ver mais

Liked Liked

Built Support Vector Machine(SVM) from scratch in Rust [P]

⋅

Built my own SVM classifier from scratch in Rust. It uses SMO optimization, have linear and rbf kernel, uses grid search to tune the hyperparameters. I tested it on two datasets one using Linear dataset and other using RBF, these were the results: Dataset Kernel Accuracy Recall F1 Banknote Auth Linear 96% 94% 95% Breast Cancer RBF 93% 100% 92% https://preview.redd.it/uw26u1uo0w0h1.jpg?width=720&format=pjpg&auto=webp&s=1784e1d7d310a26fa67efc63fa5191f45433a695 https://preview.redd.it/o0ahkq7p0w0h1.jpg?width=720&format=pjpg&auto=webp&s=dcb1053c34931d11b82831c6ad8cd4755ebc5816 The plot.rs file, used for plotting only was written using AI as I could not wrap […]

Ver mais

Liked Liked
The Responsibility Rule — Why “the Algorithm Did it” is Unacceptable (AI SAFE© 4)

The Responsibility Rule — Why “the Algorithm Did it” is Unacceptable (AI SAFE© 4)

⋅

By Michal Florek, October 2025 (Updated May 2026) The Illusion of Blame-Free AI, why “the algorithm did it” is unacceptable. This is why following AI SAFE© 1: Safety-First Rule, AI SAFE© 2: Economic Balance Rule, AI SAFE© 3: Transparency & Audit Rule is key. Executive Summary Artificial intelligence cannot bear moral or legal responsibility. Yet in public discourse and corporate governance, we increasingly hear: “the algorithm did it.” Accountability Chain Layout; Image 1, made by Michal Florek Artificial Intelligence is often described […]

Ver mais

Liked Liked
How to Use OpenCode for AI-Assisted Python Coding

How to Use OpenCode for AI-Assisted Python Coding

⋅

OpenCode is an open-source AI coding agent that runs in your terminal and lets you analyze and refactor a Python project through conversational commands. In this guide, you’ll install it on your system, set it up with a free Google Gemini API key, and learn the basics of how to use it in your daily programming work. Here’s what OpenCode’s main interface looks like: OpenCode’s Initial Screen OpenCode works as a conversational assistant you explicitly direct. Ask it […]

Ver mais

Liked Liked

Physical AI Conference Comes to San Jose as Robotics & Autonomous AI Go Mainstream 

⋅

The Physical AI Conference shaping the future of robotics, autonomous systems and real-world AI deployment lands in Silicon Valley this May, bringing together the engineers, builders and AI pioneers turning intelligence into physical action.  Physical AI Expo North America will take place on May 18–19, 2026 at the San Jose McEnery Convention Center, uniting global AI innovators, robotics leaders, enterprise technologists and next-generation infrastructure providers for two days focused on the future of AI in the physical world.  As AI rapidly moves beyond chatbots […]

Ver mais

Liked Liked
I Actually Built It. Here’s Every Line That Matters — and Every Line That Broke First.

I Actually Built It. Here’s Every Line That Matters — and Every Line That Broke First.

⋅

A2A Protocol, Part 2: Real Code, Real Failures, Real Production Lessons Image: generated By gemini AI If you haven’t read Part 1, here’s the 30-second version: We built an agentic workflow for a logistics client. When a second client came along, we almost copy-pasted the entire stack. My lead engineer stopped us. We spent a month researching A2A — an open protocol that lets agents built on different frameworks discover each other, delegate tasks, and return typed results over plain HTTP. […]

Ver mais

Liked Liked
Elastic Attention Cores for Scalable Vision Transformers [R]

Elastic Attention Cores for Scalable Vision Transformers [R]

⋅

Wanted to share our latest paper on an alternative building block for Vision Transformers. Illustration of our model’s accuracy and dense features Traditional ViTs utilize dense (N2) self-attention, which can become pretty costly at higher resolutions. In this work, we propose an alternative backbone with a core-periphery block-sparse attention structure that scales as (2NC + C2) for C core tokens. We further train this using nested dropout, which enables test-time elastic adjustments to the inference cost. The whole […]

Ver mais

Liked Liked

Load More

↑

Digitado © 2026