mjohn

That's a bit unfair. The Baumol effect is well known to most people in the author's audience.

If you had clicked on the link to his book you would have found out that it's available for free: https://www.mercatus.org/students/research/books/why-are-pri...

Sure, why not, I'm just thinking out loud to be honest and am probably biased towards feeling quite comfortable using something like ChatGPT. I don't ask ChatGPT to write code for me but regularly use it to explore, a bit like a rubber duck that talks back.

Just curious - why would you not feel comfortable using it in an interview? Interesting to hear a different POV.

I've never tried it, but in the age of AI why not ask candidates to interact with ChatGPT to solve a problem?

In my (limited) interviewing experience the good candidates were the ones who had a good mental model for solving novel problems, were inquisitive and asked good questions. I think these attributes are also a reasonable indicator of general intelligence.

The people who I hired that were less successful were the ones who solved the coding exercises but had a poor mental model of how to solve novel problems and did not ask good questions.

Trying to use ChatGPT could take the pressure off from regurgiating coding minutae. The interviewer would be able to observe how the candidate approaches the problem and whether they can spot where ChatGPT goes wrong. Being less confrontational (after all the "mistakes" are ChatGPT's mistakes) might also help introverted candidates.

No, because it's often very hard to estimate until you're actually working on something. It's common to get a consensus estimate that something is "easy" only for some poor developer to be left with something that is "hard" and takes much longer than estimated.

Also, something might genuinely be easy for someone who has worked on that part of the codebase before. But it could take someone new a long time to get up to speed, again making that developer look bad (this is not to advocate that only devs familiar with the code should work on it - you want to spread the knowledge around).

In the UK there are regulations that require casinos and other gambling venues to block access if a problem gambler doesn't show restraint. It's much like how we would expect a barkeeper to tell a alcoholic who has had too much that they won't be served more and shouldn't drive home.

Similar addictive behaviour occurs with these apps and the games are designed to be as addictive as possible. While users should take personal responsibility, it would be trivial for these app producers to provide users with some protections from themselves.

In economics we would call the latter a Pareto improvement - ie an efficiency improvement without any trade off

That's a bit of a simplification.

From £9.5k to £50k you pay 12% national insurance in addition to income tax, so it's really 32% not 20% tax.

Over £50k you pay 2% national insurance.

Your employer will also be paying 13.8% national insurance.

Once your income exceeds £100k you start losing your personal allowance at a rate of £1 for every £2 over £100k, so you have no personal allowance after earning over £125k. This means that your effective marginal tax rate between £100k & £125k is about 60%

I pretty much agree with you, if the client doesn't have any in-house expertise and stays hands-off then you get the scenario I described but outsourcing can work if it's in addition to in-house expertise.

In this case the client didn't want .Net 3.0 experts to support a legacy app, that's what the consultants built it in and then let it turn into legacy by not performing regular maintenance.

The client didn't know that it was becoming legacy until it was too late because they never developed the in-house expertise to understand what was necessary and the consultants have no incentive to do it by themselves. In fact they can charge more by letting it rot and requiring development of a whole new project to replace the legacy.

I recently joined a company where my main task is to modernise a business critical application built and supported by Wipro over ten years ago. Everything is still on .Net 3.0 (first released in 2007), and despite hundreds of pages of documentation the developers seem to have little understanding of what the application actually does for the business.

The impression I have is that their business model is geared towards building one thing, learning one technology, and then exploiting it for as long as possible with minimal evolution. This is fine if you don't need the software to evolve, but most business applications do need to evolve even if slower than at a start-up or FAANG.

At least where I work there has been the recent realisation that outsourcing all your technology means you will struggle to iterate and improve that technology.

Privacy is not just relevant to advertising. There are a huge number of research opportunities in the social sciences that could benefit and can help make the world better.

Could be differences in risk tolerance. A portfolio of government bonds, for example, would be expected to have a lower return than the S&P.

I'd highly recommend How Behaviour Spreads by David Centola for anyone interested in this topic. His lecture at the Santa Fe Institute is also worth watching as a summary: https://www.youtube.com/watch?v=o0fDcUJMzkI

It's worth bearing in mind that this is not a study on educational outcomes, but is assessing how useful value-added models are for quantifying teacher performance. While the biggest correlative factor may well be socioeconomic status, that does not preclude teacher's having an impact on within-group differences.

From the introduction:

> In this paper, we provide a stark illustration of the limitations to using value-added models to identify high-and low-performing teachers. We do this by applying commonly estimated models to an outcome that teachers cannot plausibly affect: student height. Aside from the implausibility of teacher effects on height, student height is an attractive measure for this exercise since it is symmetrically distributed, interval measured, and arguably less prone to measurement error than achievement. We find that the estimated teacher “effects”on height are nearly as large as the variation in teacher effects on math and reading achievement.

It's also an interesting approach: take a model that is apparently predictive and see if it's also predictive of something that we know is unrelated. By showing that value-added models are displaying spurious correlation maybe policymakers will take note.

Could maybe take a similar approach to validating our ML models?

I don't think pedantry about the name is a useful contribution to the conversation.

While it's true that Economic Sciences prize is not a "real" Nobel prize, it is commonly referred to as a Nobel prize. Interestingly, the Nobel Foundation also lists "Economic Sciences" on their website listing Nobel prizes even though they do not award or fund it: https://www.nobelprize.org/prizes/

It's pretty surprising that a service with access to all of your internal Slack conversations doesn't have a privacy policy on their website. Makes me wonder to what extent they've thought through issues like GDPR.