Medium

Thinking Loop

Thinking Loop

·

Mar 24

RLAIF’s Hidden Judge Problem

Seven measurable evaluator biases that quietly distort RLAIF prompts — and how to catch them before they harden into your reward signal.

RLAIF’s Hidden Judge Problem

Thinking Loop

·

Mar 24

When Query Expansion Hurts RAG

How “helpful” rewrites quietly poison reranking, bury the right chunks, and make good RAG systems look smart for the wrong reasons.

When Query Expansion Hurts RAG

Thinking Loop

·

Mar 23

RAG Can Make Wrong Answers Sound Right

Why retrieval-augmented generation sometimes boosts confidence, not correctness — and what strong teams do about it.

RAG Can Make Wrong Answers Sound Right

Thinking Loop

·

Mar 21

When One Tool Field Breaks the Agent

Why tiny schema drift causes silent agent failures, bad tool calls, and expensive debugging loops.

When One Tool Field Breaks the Agent

Thinking Loop

·

Mar 21

Reward models: 9 signs you’re training the judge

Learn how reward hacking shows up in modern AI systems, and how to tell when your model is pleasing the evaluator instead of actually…

Reward models: 9 signs you’re training the judge

Thinking Loop

·

Mar 20

Node HTTP Agents: 10 Pooling Knobs That Move p99

The Node.js connection settings that quietly decide whether your API feels fast, flaky, or painfully inconsistent under load.

Node HTTP Agents: 10 Pooling Knobs That Move p99

Thinking Loop

·

Mar 20

RAG Isn’t Accuracy: 8 Confident Failure Modes

Why retrieval-augmented generation can make wrong answers sound more trustworthy — and what smart teams do about it.

RAG Isn’t Accuracy: 8 Confident Failure Modes

Thinking Loop

·

Mar 19

Node backpressure myths that blow up memory

Understand the stream behaviors that quietly inflate buffers, stall consumers, and turn ordinary Node.js workloads into memory incidents.

Node backpressure myths that blow up memory

Thinking Loop

·

Mar 19

Stop Trusting “Stable” RL Runs

Why reinforcement learning experiments that look consistent can still hide brittle variance, false confidence, and broken conclusions.

Stop Trusting “Stable” RL Runs

Thinking Loop

·

Mar 18

Node Concurrency: 9 Queue Starvation Patterns Hidden by Throughput

Why fast aggregate throughput can still hide slow, unfair, and user-visible latency in Node.js systems.

Node Concurrency: 9 Queue Starvation Patterns Hidden by Throughput

Thinking Loop

Thinking Loop

At Thinking Loop, we dive into AI, systems design, productivity, and the curious patterns of how we think and build. New ideas. Practical code. Deep insight.

Following

Help

Status

About

Careers

Press

Blog

Store

Privacy

Rules

Terms

Text to speech