@bernhardsson@can I’m sure that if you try to minimize prediction error while also trying to maximize the initial mental valuation a customer gives to doing the trip with Uber, you get some slightly inaccurate results and are happy about them as an organization
@bernhardsson@can Keep in mind they probably make some money by advertising a lower ETA than realistic because people will compare Uber to other methods, pick Uber, wait, and by the time they realize its late, the true ETA is short enough to justify still waiting for the Uber
@6digitstudio@Griffin69211438@mdt22_@SamRo Selection bias, survivorship bias, correlation, confidence metrics, risk adjusted return, probability distributions of outcomes, kernel density estimations, Monte Carlo methods, overfitting, underfitting, the Kelly Criterion, expected value and game theory, regression
@NVDA_Investor@antibearthesis Google “cyclicality”, there are a lot of good tech companies to buy but Nvidia and the rest of semiconductors at their current valuations are not on the list
@CHORTCHOMP@Cdnwf_@LinkofSunshine@SantaKropotkin They’re about equal because half of data science is communication and creativity about approaching data-based problems, which LLMs are exceptionally horrible at, and right now they produce pretty inefficient code when it comes to big data. And in a lot of orgs, the DS is the DE
@CHORTCHOMP@Cdnwf_@LinkofSunshine@SantaKropotkin I’m choosing to ignore the point about human developers and risk because I fail to see how that doesn’t apply to every other niche within software, including data science
@CHORTCHOMP@Cdnwf_@LinkofSunshine@SantaKropotkin Ok. Data science is not just pattern recognition and data transformation. Is there a reason you think data engineering is uniquely safe from automation, other than “it exists upstream of data science”