Capital de Tokens
Ideas, analysis, and independent articles from the channel and its episodes.
Blog RSSPublic leaderboards do not always answer the question you actually care about. We design an architecture for evaluating LLMs with your own benchmarks, starting with fresh competitive-programming problems, reproducible execution, cost metrics, and a clear separation between models and agents.
At Platzi Conf 2026, Freddy Vega presented a deliberately optimistic AI scenario: near-zero-cost expert knowledge, faster science, dramatically cheaper software, and a strategic opening for Latin America. We examine which parts already have evidence and where caution is still warranted.
Anthropic lets users record their screen, clicks, typing, and voice in Cowork so Claude can observe a process and propose a reusable Skill. The feature brings AI agents closer to semantic RPA: teach by doing instead of documenting every step by hand.
Sebastian Thrun and Peter Norvig's historic AI course is still available free on Udacity. It covers search, probability, planning, reinforcement learning, vision, robotics, and NLP: foundations that help explain AI beyond LLMs.
OpenAI began quietly rolling out Luna Reserve in Codex before documenting it publicly. We reconstruct the gpt-reserve timeline, explain its separate allowance, and what changes when it takes over.
Gartner says enterprise AI workloads at scale will not run on quantum hardware before 2028. The key is separating Quantum AI, quantum-inspired methods, hybrid systems, and post-quantum cryptography so companies can invest where measurable value exists today.
An Every team completed Anthropic's official training and found something more useful than a bag of tricks: a shared vocabulary for Skills, MCP, the API, and Claude Code. They also exposed the limits of certifying an industry that changes faster than its courses.
What it means for Bubblewrap to be a low-level unprivileged sandboxing tool, how it uses Linux namespaces and mounts, the role it plays inside Flatpak, and why it should not be confused with a complete security policy.
Cloudflare cut the memory footprint of the DNS cache behind 1.1.1.1 by more than half with five Rust representation changes. The result: roughly 100 TB of RAM freed, higher throughput, and lower latency.
Falling Computer Science enrollment and rising demand for quantitative talent are fueling a new thesis: in the AI era, mathematics can be a major advantage. But the strongest profile is not Math versus CS; it is Math + Programming + AI.
Cursor explains how its Router combines a complexity predictor, a task taxonomy, and a budget-constrained optimizer to decide when to use a cheap model and when to escalate to a frontier model.
`--dangerously-bypass-approvals-and-sandbox` removes two different Codex barriers: the sandbox and approvals. We explain the full security model, read-only/workspace-write/full-access modes, rules, permission profiles, networking, CI, and how to design autonomous agents without handing them the whole machine.
Capital de Tokens
