scripod.com
Are We Prompting New AI Models the Wrong Way?

Overview

Shownote

Highlights

Transcript

Chapters

Pins

Are We Prompting New AI Models the Wrong Way?

The Daily AI Show

1 DAYS AGO
The Daily AI Show

The Daily AI Show

1 DAYS AGO
OverviewShownoteHighlightsTranscriptChaptersPins

Shownote

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 ...

Highlights

The hosts return to discuss a range of AI topics, starting with a new security plugin for Claude Code that automates threat modeling. This sparks a broader conversation about local machine security, the differences between Mac and Windows, and the challenges of building custom security tools. The discussion then shifts to practical AI adoption, examining a Google report on how AI is used as an assistive tool across various industries, including manual trades like auto repair.
00:00
Anthropic released a security plugin for Claude Code
01:41
Ensure code doesn't introduce security vulnerabilities
02:03
Plugin analyzes code bases for vulnerabilities
03:25
Could Claude Code replace CCleaner?
05:06
Combine deep research with security plugins
06:02
Macs are not inherently safer than PCs
07:00
Self-awareness of what one can't replicate
09:49
Kimi K3 built via distillation of Fable 5
09:58
Something was accomplished in a particular manner.
10:00
An attempt to discredit Moonshot's achievement
11:17
The claim may be exaggerated, but there is evidence of a known pattern.
12:11
Traditional distillation from Fable 5 would have been too quick to be plausible
13:20
Perplexity no longer feels like AI slop
14:01
These policies are not new
15:16
Banning open models for citizens is legally and practically challenging
16:05
Local use is technically possible but resource-prohibitive
17:09
Canada often faces more company-level restrictions
18:40
Strix has 43,000 stars and connects to an AI API
19:10
Highly rated but expensive tool
19:37
AI code must be hardened against health data theft
20:43
Google's new AI and Economy Atlas report
22:03
Only a third of its data comes from English-speaking U.S. users.
23:00
AI serves as a live collaborator for diagnostics.
24:12
AI adoption is broad across industries and occupations
25:33
AI is more a collaborator than an automation tool.
27:04
Demand for radiologists has increased despite warnings
27:43
AI can aid rather than replace skilled workers
29:04
AI augments, not replaces, human expertise
30:06
AI will not replace mechanics but requires upskilling.
33:00
AI could replace the mechanic
33:40
Modern cars are heavily computerized, requiring caution.
34:04
AI-assisted video tutorials make complex tasks accessible
36:10
Proactively replace the 12-volt battery to avoid being stranded.
37:35
Claude Code reduced its system prompt by 80%.
37:36
Newer models prefer fewer examples to avoid constraining behavior.
38:00
Higher-end models are negatively impacted by overly verbose prompting
39:01
Shorter prompts improve AI output quality
40:05
Too many instructions create cognitive load.
41:04
Use fewer, more general instructions for newer models
42:42
Claude Code reduced its system prompt by 80%
43:31
Less specific instructions might have avoided the issue
44:04
Build systems for repeatable prompting, not just prompts.
45:07
Focus on building a structured prompting system
46:38
LanguageModelBuilder.com offers educational content on LLMs.
47:59
Start a project with various base models
48:15
Bring your own data to build your own model
48:38
Build your own language model.
49:35
Learn AI by building it yourself
50:33
Everyone will want to create their own custom models.
51:03
Train a GPT-2 small class model in a day
52:04
Customized data may not match past AI performance
53:01
Fine-tuning AI models on Macs is currently M-series only.
54:40
Gradient descent reduces loss to optimize AI training weights.
56:29
Using language models to clone oneself from emails
57:24
AI helps me be a better version of myself.
59:24
AI models need less verbose communication
1:00:02
Being teased means you're trusted and loved
1:01:35
Goodbye until next day

Chapters

Episode Intro And Brian Returns
00:00
Claude Code Security Plugin
01:39
Code Base Threat Modeling
02:00
CCleaner And Local Machine Security
03:25
Malware Detection And System Cleanup
05:00
Windows, Macs And Security Assumptions
06:00
Thinking Through AI Security Projects
07:00
McAfee, Malware Feeds And Bloatware
07:54
White House Claim About Kimi K3
09:49
Moonshot, Fable 5 And Distillation
10:00
Export Controls And NVIDIA Systems
11:00
Kimi K3 Similarity And Distillation Timing
12:00
Ethan Mollick On U.S.-China Model Tension
13:19
Possible AI Model Export Controls
14:00
DeepSeek Ban And Government Device Restrictions
15:16
Cloud, App Store And Infrastructure Pressure
16:00
Whether U.S. Users Could Lose Access
17:00
Gareth Joins The Security Conversation
18:32
Strix Pen Testing System
19:09
Black Box, Gray Box And White Box Testing
19:33
Secure Scan CLI And Healthcare Security
20:32
Google AI And Economy Atlas
21:54
AI As Task Help, Not Full Automation
23:00
AI Use In Manual And Technical Trades
24:00
Fifteen Million Gemini Interactions
25:31
Google DeepMind Taxonomy
26:54
Radiologists And AI Job Predictions
27:38
Automotive Techs And AI Assistance
29:01
Multimodal Diagnostics And Expert Support
30:00
Meta Ray-Bans, Video And Repair Context
32:07
Metaglasses And AI-Guided Car Repair
33:33
YouTube As The Earlier Repair Assistant
34:00
Brakes, Robot Fixers And DIY Limits
35:00
EVs, Batteries And Modern Car Complexity
36:10
Claude Code Reduces Its System Prompt
37:35
Shorter Prompts For Newer Models
38:00
Testing Concise Prompts Against Old Workflows
39:00
Prompt Length, Cognitive Load And Model Reasoning
40:00
Luna, Fable And Lower-Instruction Prompting
41:00
"Say Less" Prompting Recommendation
42:38
Project Instruction Drift
43:23
Token Waste From Over-Testing
44:00
Building Prompt Systems, Not Just Prompts
45:07
Language Model Builder
46:38
What Is A Large Language Model
47:57
Tokenization, Embeddings And Transformers
48:07
Pre-Training And Custom Data
48:37
Felix Reisberg And LanguageModelBuilder.com
49:34
Learning AI By Building A Model
50:27
Custom Small Models And User Experience
51:00
GPT-2 Class Models And Expectations
52:00
Fine-Tuning And Python-Specific Models
53:00
Gradient Descent
54:37
Evolutionary Model Merge
56:27
Cloning A Writing Voice
57:21
Gmail Polish And Better Communication
59:21
Episode Wrap-Up
1:00:01
Three-Year Anniversary Mention
1:01:35

Transcript

Brian Maucere: Hey, what's going on, everybody? Welcome to the Daily AI Show. Today is July 23rd, 2026. It's episode 774. And it's Thursday. Yeah. So we're live. Gareth: It's 10am on the East Coast. Brian Maucere: With me today are Beth and Andy and I'm ...