Self-updating service

A curriculum that never goes stale.

Once per week, a frontier AI agent, running with maximum thinking, researches what has changed, trended or become obsolete in AI engineering, then evolves this curriculum: adding, updating and retiring tracks, concepts, projects and resources under automated safeguards. This page reports the most recent run, full transparency, real data.

Latest run

Most recent self-update.

Last result: failed
Last update date (IST): 2026-W44
Last run at: 2026-10-26T01:44:14.966Z
Curriculum version: 18
Note: Update error: {"error":{"code":400,"message":"Thinking budget is not supported for this model. Use thinking level instead. See https://ai.google.dev/gemini-api/docs/thinking for details.","status":"INVALID_ARGUMENT"}}

Changelog

Recent curriculum changes.

Every weekly run is logged. The curriculum evolves with the frontier, and you can see exactly when.

Most recent changelog entries, newest first
Date	Version	Summary
2026-10-11	v18	Update RAG to asymmetric late interaction, modernize preference alignment with GRPO, integrate low-latency semantic routing, and strengthen agent runtime security against tool hijacking and prompt injection. (4 change(s) applied).
2026-10-05	v17	Tighten agent sandboxing with hardware isolation and out-of-band watchdogs, adapt Software 3.0 toward intent specification and verification harnesses, expand multimodal world modeling with spatial intelligence, and incorporate statutory non-delegation controls into AI governance. (4 change(s) applied).
2026-09-28	v16	Modernized the curriculum to integrate always-on adaptive reasoning, dynamic inline tooling with KV-cache preservation, self-organizing agent teams, execution/governance separation, dual-use cyber evaluations, and regulatory kill-switch governance. (9 change(s) applied).
2026-09-21	v15	Incorporate DeepSeek-R1 into post-training RLVR and preference optimization, integrate Karpathy's LLM101n into transformer serving internals, and tighten MCP and hybrid RAG concepts against outdated paradigms. (4 change(s) applied).
2026-09-13	v14	Integrate kernel-level microVM/eBPF agent containment, recursive self-improvement loops, and machine-checked Lean 4 verification into agentic systems, applied modeling, and safety evaluations. (4 change(s) applied).
2026-09-06	v13	Update curriculum concepts to address formal EU AI Act systemic-risk enforcement, dynamic multi-tier context engineering with KV-cache budgeting, post-Chain-of-Thought latent safety evaluation, and collusion-resistant multi-agent verification. (4 change(s) applied).
2026-08-30	v12	Updated agentic, modeling, and production evaluation concepts to incorporate Reinforcement Learning from Verifiable Rewards (RLVR), Model Context Protocol (MCP) standards, and Evaluation-Driven Development (EDD). (4 change(s) applied).
2026-08-23	v11	Update curriculum concepts across Agentic Systems, AI Infrastructure, and Applied ML to incorporate Agent-to-Agent (A2A) protocol standardization, Process Reward Models (PRMs), persistent radix prefix caching, and synthetic mid-training capability injection. (4 change(s) applied).