AI Evaluation
516 episodes · Page 8 of 22
#4696: Why AI Over-Explains Simple Tasks
Why does AI turn a simple request into a 24-page document? We explore the training biases and architectural limits behind AI's tendency to over-del...
#4682: When Context Windows Grow: What Survives?
If context windows jump to 10M tokens, which AI engineering practices survive? We separate load-bearing architecture from temporary scaffolding.
#4670: Why AI Weights Are Indecipherable
Can you actually read an AI model's weights like a book? We explore why the answer is mostly no — and what researchers are doing about it.
#4668: Why Podcast AI Voices Sound Too Perfect
We dig into why AI podcast voices sound too clean—and how TTS is learning to stumble, overlap, and interrupt convincingly.
#4619: The Token Trap: Why AI Context Windows Confuse Everyone
Token counts are everywhere, but they don't tell you what AI can actually do. Here's what should replace them.
#4589: The Answer Reflex: Why AI Models Can't Follow Instructions
DeepSeek passes a simple test that stumps GPT and Claude. Why can't Western models stay in character?
#4570: One Negative Prompt Beat a Dozen Positives — Here's Why
Why one blunt "don't" outperformed a dozen careful "do's" — and how to structure your system prompts for maximum compliance.
#4505: What DeepSeek's Training Data Reveals About Model Voice
Why Chinese AI models trained on 60% English still produce dialogue that feels distinctively different.
#4490: Right to Repair: History, Hype, and Hard Limits
From John Deere tractors to iPhone batteries — the real history, genuine commitments, and where repair advocates overreach.
#4477: Structuring Raw Voice Prompts at Scale
How to decompose messy voice dictation into clean, routable components for AI workflows.
#4444: Testing the Unpredictable: QA for Agentic AI
How QA adapts when your AI system gives different answers to the same question every time.
#4415: Why Your Phone Scan Fails (and How to Fix It)
AI scanners are amazing—until they fail on glare or white paper. Here's how to fix it with stuff you already own.
#4340: Building an AI Tutor That Teaches You to Code Without Writing It For You
How to build an AI tutor that assigns projects, reviews code, and remembers what you've learned — without writing your code for you.
#4338: The Hidden Skill Most AI Models Still Fail At
Why do even top AI models still accidentally paste your questions into the output? A deep dive into a missing benchmark.
#4313: Did You Really Build That App? AI Imposter Syndrome
You built a working app with AI in a weekend. So why does it feel like you don't deserve to charge for it?
#4312: Why Speech-to-Text Still Fails at Its Own Name
When OpenAI's Whisper misheard its own name as "Wispr," it revealed why 95% word accuracy still isn't good enough.
#4304: Rhetoric for Interviews and Pitches
Three persuasion techniques from political speech, applied ethically to job interviews and client pitches.
#4274: Fixing Loop-Damaged Podcasts with AI and FFmpeg
Can an AI agent detect repeated audio, generate FFmpeg commands, and overwrite files in cloud storage without breaking RSS?
#4268: When AI Trains on AI: The Model Collapse Problem
What happens when AI trains only on AI-generated content? The answer is model collapse — and it's already happening.
#4208: The Awkward Honesty of AI Attribution
Why admitting AI helped feels like a confession — and what emerging norms actually work.
#4180: Translation-Safe Writing for Global Supply Chains
How to write English that survives machine translation without losing critical specs.
#4169: Change Detection AI for Construction Sites
How AI catches what experts miss in construction site photos — from Siamese networks to transformer models.
#4157: Prompt Engineering Is a Misnomer — Managing Ambiguity in AI
Why the best AI outputs come not from tighter prompts, but from knowing where to leave room for surprise.
#4122: Why No AR App for Infant CPR?
A parent at 2AM needs visuals, not a PDF from 2014. Why doesn't the app exist?