Not surprising in the slightest. I work in machine learning and no longer trust most papers I read. A lot of papers have fabricated or dishonest results, and the people who publish these results are often the ones who get recognition, jobs at FB etc. It happens in all fields of so-called "science".

What are some signs an ML paper is untrustworthy?

Being advertised in mainstream media seems like a good initial assessment for shit science :-)

No code

This is an alleged fabrication.

That is malice.

But some races were likely separated for longer than that, so these mutations are likely uneven across populations. Very disturbing if true.

Or they seperated after having this mutation since we've no evidence that any specific ethic group (I assume you used race for this?) has inability of learning a recursive language.

I had the same question while reading TFA. The Khoisan ethnic group split from the main population an estimated 150k years ago, and although their clicking language is peculiar I don't think they have any issue with recursive grammar.

Well I appreciate these interview loops because at least they are still far more objective than big company promo processes when whoever lies the most and forces others to do their dirty work gets promoted. I do wish the interview process was even more objective, just administer and extremely difficult IQ Test instead.

IQ tests are rarely objective and they most definitely don't measure the things most important for a large company.

Doesn't have to be a standard IQ test, I would be fine with the test being about algorithms and systems design type questions. Just please remove the biased human component from interviews and use a more standardized metric across a group of people instead of randomly chosen questions for each candidate with subjective evaluation. The current process is not great even for people with good tech skill due to the randomness and subjective evaluation factors.

Also you are wrong about standard IQ tests not being able to measure relevant skills for tech companies. They not perfect but I bet even those would do a better job than the current process. The psychometric research is clear that they are useful for predicting results across a wide variety of domains.

> standard IQ tests not being able to measure relevant skills for tech companies

I am basing this on what the HR department for the company that I work for tells me. They have been administering them for years and they have never been a reliable indicator of company performance. That doesn't mean they're not used - management loves them.

> The psychometric research is clear

Do you have some sourcing for me? Perhaps I can read up and improve my own professional experience with new colleagues.

Nor for a medium or small company. (I say this as someone who tends to take these tests very well, but they’re not that commercially relevant.)

the vast majority of ML/data scientist positions are actually similar to the hypothetical ML engineer position you described.