Home on Hrushikesh Pawar

The Imitation Game Was Never a Benchmark

Sun, 19 Jul 2026 00:00:00 +0530

Abstract. Turing did not give us a universal leaderboard for intelligence. He replaced an underspecified metaphysical question with an operational experiment. Modern debates about whether LLMs think or reason are therefore largely debates about constructs, protocols, judges, and alternative explanations.

This essay is in progress.

1. The question Turing refused to answer

Alan Turing begins Computing Machinery and Intelligence with a question that still frames arguments about AI:

“Can machines think?”

Bun's Zig-to-Rust Rewrite Was Not a Prompt

Sun, 19 Jul 2026 00:00:00 +0530

Abstract. The interesting unit in Bun’s rewrite is not the model call. It is the closed-loop migration system: a porting specification, an externalized lifetime analysis, task sharding, compiler feedback, a language-independent test oracle, adversarial review, and continuous human supervision.

Planned (weeks 6–10). Every statement will be tagged source / observation / inference / speculation; the reported figures (~535K lines of Zig, 1,448 source files, ~50 dynamic workflows, an 11-day path to a passing test suite) are facts to audit, not results to casually reproduce.

Do Coding Agents Need Code Graphs?

Sun, 19 Jul 2026 00:00:00 +0530

Abstract. “Graphs understand repositories” and “grep is enough” are both underspecified claims. A representation is useful only if it helps a system retrieve change-relevant context more accurately or cheaply under a fixed budget.

Planned (weeks 11–18). Conditions will be named honestly (“graph expansion on top of lexical seeds”, not “graph vs grep”), and the lexical bias of patched-file ground truth will be acknowledged up front.

How Neural Networks Learn

Sun, 19 Jul 2026 00:00:00 +0530

Abstract. Backpropagation is not a mysterious brain-inspired learning rule. It is an efficient application of the chain rule to a composed computation. The conceptual path from a perceptron to a multilayer network is a path from linear decision geometry to differentiable representation learning.

Planned (weeks 19–25). Done when every equation maps to a named variable in the implementation.

About

Mon, 01 Jan 0001 00:00:00 +0000

I build production AI systems. This is where I reconstruct the theory behind them, test claims about coding agents and migrations, and publish the derivations, experiments, and failures.

Every essay carries one falsifiable central claim and at least one artifact that required real work — a benchmark with my own numbers, a runnable repo, a derivation, an annotated trace.

Find me on GitHub, or subscribe via RSS.

Now

Mon, 01 Jan 0001 00:00:00 +0000

Current essay: The Imitation Game Was Never a Benchmark — weeks 1–5, ≤8,000 words.

Public research checklist:

Judge recruitment message sent (day 1–2)
Reading pass: Turing 1950 → draft §2 immediately after
experiments/a1-imitation-pilot/protocol.md frozen and committed before any data collection
Pilot run: 24 prompts × 5 categories, 3–5 judges, length-matched answers, randomized order
Analysis + honest labeling as an exploratory pilot

Reading budget: 10–12 h across weeks 1–3. Next in queue: Bun’s Zig-to-Rust Rewrite Was Not a Prompt (the B2 ledger is already accumulating).