> we find that poverty is still 3 percentage points higher in treated counties 10 years after the Walmart opening.
So... 10 years after Walmart opens, poverty is 3% higher, and annual household earnings decline by $4,230. Is this a huge effect? This is something like a 10 point difference on the SAT (https://www.cs.jhu.edu/~misha/DIReadingSeminar/Papers/DixonR...).
> annual household earnings decline by $4,230. Is this a huge effect?
Yes. Median household income is something around $70K, and rural households are not making anywhere near median household income. If we don't think it's a big effect, then we shouldn't mind sending them $4.5K checks every year to make up for it.
It's an effect large enough to drive 3/100 people affected into poverty.
If you are one of those who remain employed - that is you work a better job than retail in the first place you are much better off. We need better training for those who currently are working retail, but the community is overall better off.
Rural communites like walmart as they know they are something if they are able to support one. plus walmart has stuff at good prices. It is city folks who have never lived in a rural area (a 2 week vacation is not living there) who hate walmart
It's not. That's part of the research. From the article:
> "Indeed, Walmart’s many defenders argue that the company is a boon to poor and middle-class families, who save thousands of dollars every year shopping there. Two new research papers challenge that view. Using creative new methods, they find that the costs Walmart imposes in the form of not only lower earnings but also higher unemployment in the wider community outweigh the savings it provides for shoppers. On net, they conclude, Walmart makes the places it operates in poorer than they would be if it had never shown up at all. Sometimes consumer prices are an incomplete, even misleading, signal of economic well-being."
- can also finetune with RLHF or modify the Q-A pairs directly
- then you have a final model finetune once the Q-A pairs look good
- then you use RAG over the corpus and the Q-A pairs because the model doesn't remember all the facts
- then you have a bullshit detector to avoid hallucinations
So the corpus is very important, and the Q-A pairs are also important. I would say you've got to make the corpus by hand, or by very specific LLM prompts. And meanwhile you should be developing the Q-A pairs with LLMs as the project develops - this gives a good indication of what the LLM knows, what needs work, etc. When you have a good set of Q-A pairs you could probably publish it as a static website, save money on LLM generation costs if people don't need super-specific answers.
To add to the current top-scoring comment, though (https://news.ycombinator.com/item?id=42326324), one advantage of an LLM-based workflow is that the corpus is the single source of truth. It is true that good documentation repeats itself, but from a maintenance standpoint, changing all the occurrences of a fact, idea, etc. is hard work, whereas changing it once in the corpus and then regenerating all the QA pairs is straightforward.
One way to look at it is that most college textbooks are bad. They are written by one person, who has limited time, and limited understanding. And then you have a teacher who knows even less than the textbook, trying to explain the textbook.
In contrast, Wikipedia editors have all the time they care to spend on the subject (which is a lot!). They often reproduce the original discoverer's words, when appropriate. E.g., https://en.wikipedia.org/wiki/Central_dogma_of_molecular_bio... uses Crick's words. Whereas in the case of Maxwell, the formulation "is credited to Oliver Heaviside". Long story short: Wikipedia has the best explanations, and if you don't think so then fix it!
I take away a different lesson: Dart is most likely slow out of the box. The author lists several reasons: int vs. Int64, GC interrupts, load hoist optimization. All of these are issues with Dart. Hence microbenchmarks, even without interpretation or validation, point to issues with the language implementation. They are not "meaningless".
It is true, one microbenchmark only shows that said microbenchmark is slow, not that the language as a whole is slow, but the plural of anecdotes is data. If you systematically evaluate a representative set of microbenchmarks (as in the Computer Language Shootout), then it is proof that the language is slow or fast.
Now of course one can argue about what is "representative", but taking random samples of code from across GitHub seems like a reasonable approach. And of course there is the issue of 1-1 translation but at this point LLM's can do that.