KOHSHEEN TIKU
ARTICLE · 28 MIN READ · APRIL 15, 2026
The hardest part of contributing to AI safety isn't picking the right research direction — it's picking a route into the field that matches your background and timeline. This chapter is the field map for that decision.
ARTICLE · 26 MIN READ · MARCH 28, 2026
Alignment tries to make models that won't betray you. Control assumes the model might, and engineers deployment so safety doesn't depend on the answer. This chapter walks through I/O filtering, Constitutional Classifiers, trusted/untrusted model splits, and the broader control paradigm.
ARTICLE · 35 MIN READ · MARCH 18, 2026
Building an agent is the beginning. Knowing whether it's actually working — accurately, efficiently, safely, and reliably — is the ongoing challenge. This chapter covers the complete framework for measuring and maintaining agent performance in production.
ARTICLE · 34 MIN READ · MARCH 14, 2026
Capable agents without guardrails are dangerous agents. This chapter shows how to build the multi-layered defense systems that keep AI behavior safe, predictable, and aligned — from input validation to jailbreak detection to human oversight.
ARTICLE · 33 MIN READ · MARCH 10, 2026
How does an AI agent actually think? This chapter reveals the techniques that transform LLMs from pattern-matchers into deliberate problem-solvers: CoT, ToT, ReAct, self-correction, RLVR, and more.
ARTICLE · 27 MIN READ · MARCH 10, 2026
If governance is the institutional layer (Ch. 5) and security is the adversarial layer (Ch. 4), evaluations are the evidentiary layer underneath both. This chapter walks through the science of evals, the major lab safety frameworks, and the domain-specific benchmarks that decide whether a frontier model ships.
ARTICLE · 33 MIN READ · MARCH 06, 2026
Not every question needs a supercomputer. Resource-aware optimization routes simple queries to cheap, fast models and reserves expensive, powerful ones for genuinely hard problems — saving cost without sacrificing quality.
ARTICLE · 31 MIN READ · MARCH 02, 2026
MCP connects agents to tools. A2A connects agents to agents. The Agent2Agent protocol is the missing standard that lets any AI agent — regardless of framework — collaborate with any other.
ARTICLE · 37 MIN READ · FEBRUARY 26, 2026
LLMs know a lot — but their knowledge stopped the day training ended. RAG is the bridge between static model weights and the live, private, specific knowledge that makes agents actually useful.
ARTICLE · 33 MIN READ · FEBRUARY 22, 2026
Full autonomy sounds ideal — but in high-stakes domains, the cost of a single AI error is too high. Human-in-the-Loop is the pattern that keeps humans in control of the decisions that matter most.