Are We Prompting New AI Models the Wrong Way?

The episode opened with Brian returning after two days away, then Andy picked up the cybersecurity thread from the prior show. The hosts discussed Anthropic’s new Claude Code security plugin, which uses agents to map a code base, build a threat model, and have an independent reviewer challenge the findings. That led into a broader discussion about local machine security, CCleaner, malware detection, McAfee, Macs versus Windows, and the limits of trying to build your own security tools.

The back half moved from AI adoption to practical AI workflows. Beth covered Google’s AI and Economy Atlas, which found that AI use remains more assistive than fully automated and reaches beyond white collar jobs into manual and technical work. The hosts then discussed automotive technicians, AI glasses, diagnostics, multimodal repair support, and how AI may upskill trades rather than replace them. Brian closed the main news discussion with Claude Code reportedly shrinking its system prompt by 80%, which led into a practical point: newer reasoning models may perform better with shorter prompts that define the goal, the deliverable, and what good looks like.

Key Points Discussed

00:00:18 Episode Intro And Brian Returns

00:01:39 Claude Code Security Plugin

00:02:00 Code Base Threat Modeling

00:03:25 CCleaner And Local Machine Security

00:05:00 Malware Detection And System Cleanup

00:06:00 Windows, Macs And Security Assumptions

00:07:00 Thinking Through AI Security Projects

00:07:54 McAfee, Malware Feeds And Bloatware

00:09:49 White House Claim About Kimi K3

00:10:00 Moonshot, Fable 5 And Distillation

00:11:00 Export Controls And NVIDIA Systems

00:12:00 Kimi K3 Similarity And Distillation Timing

00:13:19 Ethan Mollick On U.S.-China Model Tension

00:14:00 Possible AI Model Export Controls

00:15:16 DeepSeek Ban And Government Device Restrictions

00:16:00 Cloud, App Store And Infrastructure Pressure

00:17:00 Whether U.S. Users Could Lose Access

00:18:32 Gareth Joins The Security Conversation

00:19:09 Strix Pen Testing System

00:19:33 Black Box, Gray Box And White Box Testing

00:20:32 Secure Scan CLI And Healthcare Security

00:21:54 Google AI And Economy Atlas

00:23:00 AI As Task Help, Not Full Automation

00:24:00 AI Use In Manual And Technical Trades

00:25:31 Fifteen Million Gemini Interactions

00:26:54 Google DeepMind Taxonomy

00:27:38 Radiologists And AI Job Predictions

00:29:01 Automotive Techs And AI Assistance

00:30:00 Multimodal Diagnostics And Expert Support

00:32:07 Meta Ray-Bans, Video And Repair Context

00:33:33 Metaglasses And AI-Guided Car Repair

00:34:00 YouTube As The Earlier Repair Assistant

00:35:00 Brakes, Robot Fixers And DIY Limits

00:36:10 EVs, Batteries And Modern Car Complexity

00:37:35 Claude Code Reduces Its System Prompt

00:38:00 Shorter Prompts For Newer Models

00:39:00 Testing Concise Prompts Against Old Workflows

00:40:00 Prompt Length, Cognitive Load And Model Reasoning

00:41:00 Luna, Fable And Lower-Instruction Prompting

00:42:38 “Say Less” Prompting Recommendation

00:43:23 Project Instruction Drift

00:44:00 Token Waste From Over-Testing

00:45:07 Building Prompt Systems, Not Just Prompts

00:46:38 Language Model Builder

00:47:57 What Is A Large Language Model

00:48:07 Tokenization, Embeddings And Transformers

00:48:37 Pre-Training And Custom Data

00:49:34 Felix Reisberg And LanguageModelBuilder.com

00:50:27 Learning AI By Building A Model

00:51:00 Custom Small Models And User Experience

00:52:00 GPT-2 Class Models And Expectations

00:53:00 Fine-Tuning And Python-Specific Models

00:54:37 Gradient Descent

00:56:27 Evolutionary Model Merge

00:57:21 Cloning A Writing Voice

00:59:21 Gmail Polish And Better Communication

01:00:01 Episode Wrap-Up

01:01:35 Three-Year Anniversary Mention

The Daily AI Show Co Hosts: Brian Maucere, Beth Lyons, Andy Halliday, Gareth.