Retrieving the file
Live· 4,218,906 positions deprecated· 4,218,906 deprecated· 1,284 today
Average time-to-redundancy: 11.3 months
Know before your role does.
pssst. this is to sign up for updates on the game ;)
Will AI replace AI Evals Researcher jobs? · Latest report #8gwyyqhz
Your job has an expiration date
May 19, 2027
In 7 monthsTraining models to judge themselves is a textbook conflict of interest.
Professional critic of neural network hallucinations and logic gaps.
Check another roleComment your opinionor JOIN THE RESISTANCE!
One human is building a game about fighting back.
Quantifying your obsolescence through recursive self-audit.
bzzt.youAI▸ EVALS-GEN 9000▸ RARITY · COMMON▸ DEPRECATION IN 7 MONTHSAI's Special Power
automates its own adversarial testing suite
You design benchmarks and scoring rubrics to quantify model intelligence. You analyze failure modes to prevent hallucinations and bias in automated outputs.
The one thing AI can't do
You can detect the subtle, subjective scent of 'weirdness' in a model's generated text.
Your enemy, your name, your date. Post it.
Step by step, from the first quiet change to Deprecation Day.
Day 10
System initiates shadow-scoring of your current benchmarks to identify human-subjectivity patterns.
Week 6
An agent generates dynamic rubrics that adapt to model updates faster than your documentation cycles.
Month 3
Internal models begin self-generating adversarial prompts that are statistically more effective than human red-teaming.
Month 5
A committee of diverse model checkpoints provides consistent scoring without the need for human calibration.
Month 7 · D-Day
The system determines your analytical input adds noise to the optimization loop and shuts down your access.
What you cost, what the machine costs, and what you get on the way out.
Per year
Savings: 96.6%
99.2%
Benchmark saturation
Infinite
Prompt iteration velocity
High
Human bias variance
Your severance
A lifetime subscription to the model that replaced you.
Discussion
