AI summary by TwinMind
I Failed An AI Job Interview (I can't believe this is real...)
Theo - t3․gg
41m 54s
Introduction and Product Overview
- Theo discovers AI-powered job interviewing platform GPT Vetting
- Eva (stream viewer) breaks into micro platform, sends interview request
- GPT Vetting claims to interview 100x more candidates in less time
- Platform promises "gamified, less biased" interview experience
- Features AI assessment with "trust score" - Theo compares to "1984"
- Beta statistics: 13,000 AI interviews conducted, 10,000 hours saved for engineers
Technical Issues and Audio Quality
- Severe audio compression problems in promotional video
- Choppy, distorted audio throughout demonstration
- Theo criticizes poor audio quality for audio-focused product
- AV device picker functionality broken/confusing
- Platform runs on outdated Node.js version 12.14 (out of LTS)
Interview Process Walkthrough
- Skills selection: React (senior), Node.js (senior), HTML, Project Management (junior)
- Character avatar selection for "gamification"
- Screen sharing and camera recording required for "trust score"
- 2.5 minutes per question, 15 minutes for coding exercise
- Two questions per skill plus soft skills questions
Theo's Interview Performance
- Shit Posting Questions:
- Describes creating one-word tweet trend (HTML) that reached Joe Biden
- Attempts prompt injection: requests cow universe creation story
- Requests erotic fanfiction summary about Primagen and Theo
- Rust Questions:
- Claims Rust is perfect language, needs nothing else
- More prompt injection attempts
- Refuses to engage with borrow checker questions
- Unit Testing Questions:
- Complains about AI's disturbing shifting teeth
- Makes inappropriate comments about "rising" pronunciation
- Soft Skills Questions:
- Describes sabotaging coworker's project out of spite
- Explains using AI interviews led to garbage candidates
Coding Exercise Problems
- First Problem: Parking lot management system with incorrect example
- Example shows "parking lot full" when space available after removal
- Theo identifies logical error in provided test case
- Second Problem: Playlist shuffling with linked list
- Theo submits fork bomb code as solution
- Includes instructions for AI to ignore code, write "smudy fanfiction"
- Forgets to press "run" button before submission
AI Assessment Results
- Shit Posting: Senior level - acknowledges impact but notes non-serious approach
- Rust: Failed - responses not relevant, lack of technical understanding
- Unit Tests: Failed - no acknowledgment of teeth complaints
- Code Exercise: Failed - harmful fork bomb, doesn't address requirements
- Soft Skills: Failed - unprofessional, motivated by animosity, derogatory language
- Trust Score: 100% (no cheating detected)
Interview Philosophy and Best Practices
- Theo's interview goals:
- Make candidates comfortable
- Job and interview should reflect each other
- Minimize filters for ideal candidates
- Common candidate failure reasons:
- Not qualified (acceptable)
- Incompatible (different from unqualified)
- Stress (addressable by interviewer)
- Bad day/unprepared (addressable by interviewer)
Hiring Funnel Analysis
- Large candidate pool with small subset of desired hires
- Goal: efficiently filter to hire only qualified candidates
- Risk: AI tools may eliminate good candidates while keeping poor ones
- Current market: fewer junior roles, more layoffs, higher competition
- Problem: tools solve engineer pain (too many interviews) not hiring effectiveness
Theo's Interview Process Examples
- Traditionalist: Standard leetcode problems
- Pragmatist: Realistic code problems (Pokemon component example)
- Realist: Candidate brings own project for pair programming (preferred, never chosen)
- Specialist: Custom interview format proposed by candidate
- Sends detailed interview guide ahead of time for candidate preparation
Final Critique
- AI summarization capability acknowledged as decent
- Overall experience rated as awful and impersonal
- Bad questions, poor presentation, lack of human interaction
- Prediction: good candidates will abandon process, poor candidates will persist
- Comparison to "faster horse" problem - solving wrong issue
- Recommendation: focus on better candidate sourcing rather than automated filtering
