This gets a bit weird in the JS optimization world because it is true that each library and parts of the library you don't use are going to be extra useless bytes the user has to download. With tree shaking removing the parts of the library you don't use I think this becomes less of an issue.
I do agree with this specific instance though, that you should drop lodash.
Historically lodash/underscore was written for a time before a lot of syntax/standard libraries were added. The new javascript features obviously had lodash in hindsight so they are pretty good at making common things syntactically easier to do or just straight up taking the good functions and standardizing it (e.g., spread operators, arrow functions, Object.assign, Array.flat, Array.flatMap, etc, etc)
For personal projects, self education, or fun - go ahead and implement the entire lodash lib by hand! But in a business setting a 3.5kb reduction in bundle size and 10 hours spend refactoring the lodash code out of your project is probably a waste of time and money.
I'll agree with the advice that you shouldn't add dependencies until you need them, but saying "you never need lodash" is bad advice IMO. Absolutism is a bad quality to harbor, especially when it's over trivial things.
I know jQuery fell out of favor with the JS community and now is treated like a sign of incompetence but - just as everything in programming holy wars - at the end of the day everything is a tool, and you need to pick the right tool for the job. If the job is "get an MVP up as quickly as possible" and lodash/jQuery/underscore helps you do that, why wouldn't you use it?