> On a noisy TypeScript repository, it improved F1 by 0.246 and used 12% fewer tokens. The useful predictor was lexical noise, not whether the language had strong static types.
What is "lexical noise", which appears to be a key term in the argument, and why is it not defined? Does it mean that an identifier could show up in, say, a docstring in a non-structured way where a semantic tool would not identify it?
Makes sense to me that LLMs, which are language based, would match better with a tool that searches at a language lever rather than at a deeper structured level.
Yes, but less pithy. Placing the "average person" next to the "average number of legs" makes it clear that the idiomatic and intuitive sense of "average" that most people have is wrong. The more precise "typical person" doesn't make that same point as well, though an intelligent reader would certainly be expected to be able to read the same from it.
This is not general to all subjects in Danish universities. I studied physics in Denmark and while the majority of my exams after the second year match this description (though not all), my impression is that this is not at all the norm. Students in the humanities in my dorm, for example, seemed to have an overwhelming majority of several-day take-home written exams (with no associated oral defense).
How is it even possible that we've reached a point where "yes, this is obvious and pretty unsurprising" is the default response to spying on an industrial scale.
The traceback is actually shown based on the last-thrown exception (that thrown from the finally in this example), but includes the previous "chained exceptions" and prints them first. From CPython docs [1]:
> When raising a new exception while another exception is already being handled, the new exception’s __context__ attribute is automatically set to the handled exception. An exception may be handled when an except or finally clause, or a with statement, is used. [...] The default traceback display code shows these chained exceptions in addition to the traceback for the exception itself. [...] In either case, the exception itself is always shown after any chained exceptions so that the final line of the traceback always shows the last exception that was raised.
So, in practice, you will see both tracebacks. However, if you, say, just catch the exception with a generic "except Exception" or whatever and log it without "__context__", you will miss the firstly thrown exception.
My version of Firefox (146.0 on Debian) has exactly this. If I select a sentence and right-click, I get the menu item "Translate selection to <LANGUAGE>". In the resulting box, I can change the language pair - but the defaults that I have seen were also reasonable.
The tokenization can represent uncommon words with multiple tokens. Inputting your example on https://platform.openai.com/tokenizer (GPT-4o) gives me (tokens separated by "|"):
It's kind of fun to compare this formulation with the seemingly contradictory official arXiv argument for submitting the TeX source [1]:
> 1. TeX has many advantages that make it ideal as a format for the archives: It is plain text, it is compact, it is freely available for all platforms, it produces extremely high-quality output, and it retains contextual information.
> 2. It is thus more likely to be a good source from which to generate newer formats, e.g., HTML, MathML, various ePub formats, etc. [...]
Not that I disagree with the effort and it surely is a unique challenge to, at scale, convert the Turing complete macro language TeX to something other than PDF. And, at the same time, the task would be monumentally more difficult if only the generated PDFs were available. So both are right at the same time.
Working with both at the same time makes their strengths and pitfalls shine. It's like that dual-boot computer where you're constantly in the wrong OS.
HTML has better separation of concerns than latex.
Latex does typesetting a lot better than html.
HTML layout can differ wildly in the same document.
Latex documents are easier to layout in the first place.
What is "lexical noise", which appears to be a key term in the argument, and why is it not defined? Does it mean that an identifier could show up in, say, a docstring in a non-structured way where a semantic tool would not identify it?
Makes sense to me that LLMs, which are language based, would match better with a tool that searches at a language lever rather than at a deeper structured level.