275 - Jargon
The time has come for a quick AI/ML Jargon to Normal English Words reference guide:
In-Distribution: Things it trained on.
Out-of-Distribution: Also things it trained on, just combined differently.
Generalization: Combining the things it trained on.
Learning: Memorization
Understanding: Memorization with a Bullshit explanation added
Reasoning: Plausibly combining memorization and bullshit until it gets a new top score on a benchmark.
Benchmark: Things it trained on and happened to not suck at.
State-of-the-Art (SOTA): The wild mating call of gullible investors who lack any scientific background.
Agentic AI: A Magic 8 Ball that mostly just answers its own increasingly stupid questions while constantly running up a steep bill.
World Model: A fuzzy knockoff of Google Maps printed on used toilet paper.
Alignment: Bullshit that sounds like it aligns with the experimenter’s own biases.
There are also “Wild Card” terms, which mean whatever you want them to mean. This gives the following words no translation, as they have no meaning in AI/ML:
- AGI
- ASI
Fortunately for everyone, AI/ML does no actual work on these terms, however much some talk about them. So concludes my only ELI5 post for this year.
I welcome everyone to make their own additions to this list.
Denis O. Chris Surdak Lasse Rindom Kurt Cagle I suspect might have a few of their own.
