Machine learning

Machine learning is rarely the subject of a ToTheMoon episode; it is the background the hosts fall back on when they want to separate the noisy “AI” of the headlines from what has quietly run inside banks, retail chains and recommendation systems for years. The typical move is a correction: behind a given feature there is machine learning rather than artificial intelligence, and the classical models did not disappear when large language models arrived.

In the early episodes the term appears as a marker of expertise and as a sober footnote. Episode 009 mentioned a professor with a long list of works on machine learning, as well as systems with a huge amount of ML functionality inside. In episode 030 the hosts said outright that one Apple feature was not AI but machine learning. Episode 051 explained reinforcement learning as a kind of machine learning in which an agent is rewarded for good actions, and noted that NLP and classic ML offer plenty of jobs while computer vision remains a niche.

In later episodes the hosts insist that ML models have been used for years in any large system with data — retail, banking, credit — and ask whether LLMs are really replacing them (episode 083). Episode 140 guessed that a particular service ran not on a GPT-5.6-class model but on anomaly-detection models and similar ML systems. Episode 108 listed machine learning alongside regular expressions as a way to work through transcripts, and episode 157 reminded viewers that in data science and machine learning, datasets decide everything — data that shows what went in and what should come out.