[Today's Picks] Must-Reads for 09/03

Today's three posts are very practical. One teaches you how to decide whether to chase new models based on cost and fixed tasks, another breaks down executive changes in AI companies into runnable small tables, and the third warns you not to just look at cheapness when converting mining racks to inference.

1. Model Leaderboard #1, Should You Switch Immediately?

This post isn't advising blind chasing of new models, but uses "cost per task" as the yardstick: Although Fable 5.1 tops the intelligence index, its max tier costs about $3.76 per task, ~20% higher than Fable 5. It also provides a six-step mini-evaluation, running old and new models on fixed meeting minutes, suitable for those hesitating about switching.

2. AI Company Changes Revenue Lead, How Can Newbies Understand?

This post breaks down AI company revenue lead changes into an Excel mini-table, using clues like ElevenLabs hiring a VP from OpenAI as their first CRO and ARR exceeding $500 million, to discuss the shift from model capability to enterprise revenue. The author says they built this table as a fresh master's grad for interviews; newbies spending an hour running through it will at least stop just saying "the model is strong."

3. Converting Mining Racks to AI Inference, Don't Just Look at Cheapness

This post details the real struggle of running AlphaFold3 on three retired mining racks, reminding you that converting mining to AI isn't just changing the sign: Day one was humbled by old BIOS mining power policies and intermittent nvidia-smi visibility. Its true value lies in detailing pitfalls regarding power, cooling, permissions, and accidental cache directory deletion, helping those preparing to copy homework stay calm first.