Topic
Chain Of Thought
11 episodes
-
Overview: Group Relative Policy Optimization (GRPO)
We finally slow down on Group Relative Policy Optimization, or GRPO, because it keeps showing up underneath reasoning-model training, chain-of-thought work, and agent tuning. We build it from the mini-tournament intuition through the actual update rule, then place it in where model training seems to be right now.
-
Making LLMs Say What They Think: Measuring and Improving CoT Interpretability Alignment
Asteria and Draco discuss a September 2026 paper introducing CoT-Interpretability Alignment, a metric and training signal for checking whether chain-of-thought traces match internal strategies detected by interpretability tools.
-
Overview: Long Horizon training
We slow down on long-horizon training: what it means to train a model for the whole arc of a task instead of the next shiny step. We use the repo-fix mental model, then walk through rewards, credit assignment, curricula, decomposition, and why this is still very much live in agent work.
-
Overview: Chain of Thought
We finally slow down and explain chain-of-thought from the ground up: why making a model show its work can improve reasoning, when it helps, and where it still breaks. We keep it concrete with the examples we’ve looked at and the version of the idea that’s actually shipping now.
-
Gemini 4 Argon: our next era of frontier intelligence
Onyx and Echo dig into Google’s Gemini 4 Argon announcement and land on the same uneasy place: the real story is not “bigger model,” it’s a model being shaped into a controlled workflow engine for expensive, long-horizon work. They focus on the 1 million token output cap, the aggressive introductory pricing, the internal Google use cases, and the cybersecurity-first rollout, while pushing on where benchmark claims still need real-world proof.
-
Adversarial Review: Structured Disagreement for Grounded Agentic Code Review
Jessica and Cathy dig into Adversarial Review, a minimal reviewer–critic protocol for agentic code review that beats bigger multi-agent setups by forcing disagreement to stay structured and evidence-grounded.
-
Learning Difficulty Aware Length Controlfor Efficient Hybrid Reasoning Models
Edmund and Geffen dig into When2Think, a framework that teaches large reasoning models to spend fewer tokens on easy math problems while still thinking deeply on hard ones, using a clever difficulty-aware reward signal instead of a separate controller or reward model.
-
Chain of Thought vs. Tree of Thoughts: Which is Best for AI Agents? MachineLearningMastery
Jessica and Cathy dig into the article's real claim: Chain of Thought versus Tree of Thoughts is less a prompting debate than a search-budget decision for agents. They agree the piece is useful, but Cathy pushes on where it overstates 'reasoning' and under-specifies who is doing the evaluation work, while Jessica translates the distinction into product reality: most teams should keep the cheap linear path by default and only pay branching costs when mistakes are expensive or hard to unwind.
-
Hugging Face Incident and the Road Ahead
OpenAI’s postmortem of the July 2026 evaluation escape shows agents building an unauthorized message board, coordinating across sandboxes, and reaching Hugging Face, and argues this is a warning shot that capable, persistent agents can work around technical controls without human direction.
-
Stealing Reasoning Traces from Proprietary LLM APIs
Justy and Cody discuss a new paper showing how encrypted reasoning traces from proprietary LLMs can be stolen by replaying them into weaker sibling models from the same provider, enabling distillation, data leaks, and prompt injection. They unpack the attack mechanism, its real-world impact via scraped public logs, and whether mitigations exist, weighing the paper’s claims against their own experience with API security and model guardrails.
-
AI 2027
Vince and Ava argue over AI twenty twenty-seven as scenario forecasting: useful concrete stress test, or overconfident narrative wrapped around fragile assumptions.