April 27, 2026
This flight had no Wi‑Fi—just spicy takes
Running Local LLMs Offline on a Ten-Hour Flight
Hot laptop, wrong cable, no Wi‑Fi: commenters split between nerd glory and 'just take a nap'
TLDR: One engineer built a billing tool and ran local AI on a 10‑hour flight, then discovered a weak cable was capping power. Comments erupted: some say local AI loops and cramped seats make it pointless, others praise Apple’s efficiency and the lesson of feeling real costs.
No Wi‑Fi, hot laptop, and a power mystery: one engineer spent a ten‑hour flight pushing local AI (think smart text bots you run on your own computer) on a brand‑new MacBook, building a billing dashboard and chewing through roughly 4M tokens. The big twist? The “dead battery” drama turned out to be a bad cable—an iPhone lead capped power at 60W, while the real MacBook cable pulled 94W. Cue the comments section going full airport-security mode.
The crowd split fast. Skeptics like deanc ranted about constant “infinite loops” and called the local‑AI hype overblown, while ddarolfi nit‑picked the model name like it was a boarding pass typo. Pragmatists chimed in: vladgur said the real blocker is the tiny window seat, not chips; bobro asked, “Can’t you just read a book?” Work‑life warriors like j1000 argued that working on planes is a curse, not a flex.
There were also nerd‑belt takes with real stakes: Steve Turner said feeling the heat and drain teaches “mechanical sympathy,” making people less wasteful in the cloud; Jackson Oaks pitched Apple’s battery efficiency over NVIDIA for travel. Meanwhile, the creator logged everything with power tools, proving one point everyone agreed on: instrument before you brag.
Key Points
- •A MacBook Pro M5 Max ran Gemma 4 31B and Qwen 4.6 36B offline via LM Studio during a 10‑hour flight.
- •A DuckDB‑based billing analytics tool for loveholidays was built, revealing patterns and cross‑service correlations.
- •Approximately 4 million tokens were processed for smaller tasks, with local models performing comparably on tight scopes.
- •Limits observed included rapid battery draw, heat at 70–80W, context degradation past 100k tokens, and occasional infinite loops.
- •Instrumentation with powermonitor and lmstats uncovered a 36% power delivery gap caused by using an iPhone cable instead of a MacBook cable.