February 2026

The intersection of AI agents, blockchain, and everyday finance is no longer a distant vision—it’s happening now, and fast. From agentic commerce that hides complex rails behind simple conversations, to trust stacks built on permission, proof, and verifiable receipts, the industry is shifting toward AI as the true interface for financial decisions. Users want outcomes in plain language (“pay this,” “save that,” “grow my assets safely”), not chains or wallets to manage. Fresh off insights from Consensus Hong […]

Ver mais

The era of cloud-tethered computing is officially coming to an end. For the last three years, developers have been held hostage by API rate limits, exorbitant subscription costs, and the looming threat of closed-source data harvesting. Big Tech told us that local AI was a pipe dream. They claimed that running frontier models required server farms the size of small cities. They wanted us dependent on their infrastructure, paying rent for every token generated. Then came the lobster. […]

Ver mais

There was no outage. No alerts. n No failed pipelines. Everything “worked.” But our BI environment was slowing down, getting expensive, impossible to budget for. Not because the warehouse was failing. Because the design was wrong. AI will expose that at scale. The Architecture That Punished Curiosity. For years, the default in modern BI orgs was “DirectQuery”. Users → Reports → Semantic Model → Warehouse → Execution → Results. Every click triggered compute. It felt good. It felt […]

Ver mais

I’m trying to understand the exact logic behind this key step in dynamic programming. We know that V* satisfies the Bellman optimality equation: V*(s) = max_a [ r(s,a) + γ Σ_{s’} P(s’|s,a) V*(s’) ] Now define the greedy policy with respect to V*: a*(s) = argmax_a [ r(s,a) + γ Σ_{s’} P(s’|s,a) V*(s’) ] and define the deterministic policy: π*(a|s) = 1 if a = a*(s) 0 otherwise Step 1: Plug greedy action into Bellman optimality Because π* […]

Ver mais

Every other week, there’s a new wave of viral hype surrounding AI tools. This time around, it’s OpenClaw, a self-hosted autonomous AI agent that can send messages, write emails, browse the web, and perform other tasks. Never mind its potential security vulnerabilities (there are many) and various predecessors that almost worked. The hype is real. Everyone working with Software as a Service (SaaS) products must feel quite uneasy. The wave of “AI wrappers” already taken some casualties, but […]

Ver mais

Something that confused me for a long time: If policies are probability distributions π(a | s) why is the optimal policy in a standard MDP deterministic? Step 1 — Bellman Optimality For any state s: V*(s) = max over π of Σ_a π(a | s) * q*(s, a) where q*(s, a) = r(s, a) + γ * Σ_{s’} P(s’ | s, a) * V*(s’) So at each state, we are solving: max over π E_{a ~ π}[ q*(s, […]

Ver mais

Reinforcement Learning from Human Feedback (RLHF) enables powerful LLM alignment but can introduce reward hacking – models exploit spurious correlations in proxy rewards without genuine alignment. Compounding this, the objectives internalized during RLHF remain opaque, making hacking behaviors difficult to detect or correct. We introduce IR3 (Interpretable Reward Reconstruction and Rectification), a framework that reverse-engineers, interprets, and surgically repairs the implicit objectives driving RLHF-tuned models. We propose Contrastive Inverse Reinforcement Learning (C-IRL), which reconstructs the implicit reward function […]

Ver mais

Networks of interdependent industrial assets (clients) are tightly coupled through physical processes and control inputs, raising a key question: how would the output of one client change if another client were operated differently? This is difficult to answer because client-specific data are high-dimensional and private, making centralization of raw data infeasible. Each client also maintains proprietary local models that cannot be modified. We propose a federated framework for causal representation learning in state-space systems that captures interdependencies among […]

Ver mais

The Claude C Compiler: What It Reveals About the Future of Software On February 5th Anthropic’s Nicholas Carlini wrote about a project to use parallel Claudes to build a C compiler on top of the brand new Opus 4.6 Chris Lattner (Swift, LLVM, Clang, Mojo) knows more about C compilers than most. He just published this review of the code. Some points that stood out to me: Good software depends on judgment, communication, and clear abstraction. AI has […]

Ver mais