thundergolfer

Though significant restrictions on supply do exist, the problem does not seem to be nearly as bad in Australia as in the USA [1]. In Australia there's ~15 medical graduates per 100k vs ~8 in USA.

The doctors in Australia are still definitely overworked though. A decent number of the people I went to school with became doctors in Australia, and though we work roughly the same number of hours (~60-70hrs/week), my work in the software industry is like a stroll in the park.

It was remarkable to hear my male friends who became doctors admitting that they had broken down crying in meetings with their boss because of workplace stress and exhaustion.

1. https://www.ama.com.au/ama-rounds/13-may-2022/articles/more-...

I found Mazzucato's Entrepreneurial State argument highly valuable in this day and age. I'll definitely check out this pod.

Big4 consultancies are decently desirable graduate hirers in my country, despite paying low, demanding long hours, and providing low value training. That's not to mention the malpractice scandals.

The fact that a graduate consultant gig is more respected than teaching roles and government positions shows the rot that's set in over the past 40 years.

EY is banned in Germany from offering accounting services after this scandal: https://www.economist.com/business/2023/04/05/ey-gets-banned....

The others have similar but I think less severe scandals in other countries.

The Acquired podcast did an episode on the history of Visa that covers a lot of how the credit card industry works.

As another commenter noted, this article doesn’t pull out clearly how this whole credit card reward scheme actually works. The Acquired episode does, by the end.

It works like this: the ‘luxury’ credit card providers, partnering with Visa, take money away from merchants in order to extract profit for themselves while keeping the credit card consumers happy. The merchants are pissed about this, and regularly make lawsuits to regulate interchange. The money extracted by the credit card companies and Visa causes merchants to raise prices for everyone regardless of whether they have a rewards card or use a credit card at all.

This creates in effect a massive money transfer from the poor, who do not use rewards cards, to the rich consumers who do. The Acquired podcast provides specific numbers on just how much worse off poor consumers are given this system, and how much the richest consumers benefit.

I come from Australia where interchange fee regulation tamps down on the kind of credit card mania and fetishism seen in the USA.

It says that you are familiar with Anthony Fantano's Youtube channel?

Cool looking website :) Small nit feedback, you say "less servers to operate" when it should be "fewer servers to operate" because servers are countable.

We use shot-scraper internally to automate keeping screenshots in our documentation up-to-date. Thanks for the tool![1]

Agree that Playwright is great. It's super easy to run on Modal.[2]

1. https://modal.com/docs/guide/workspaces#dashboard

2. https://modal.com/docs/examples/web-scraper#a-simple-web-scr...

You don’t sell back to the company. Nasdaq and other exchanges have platforms for facilitating secondary market sales. You can’t trade it for something else, it’s a sale for cash. And no, it doesn’t come into your paycheck you nominate a bank account using the platform.

> Epoch five: from information to insights

Eugh, this article is full of MBA-speak, not something I'd expect from a highly technical Google Fellow.

I sometimes think about the problem of sending a message to someone in the far future such that they'd be guaranteed to receive it, regardless of their location on earth or attentiveness to the arrival time of the message.

It's a fun creative thinking exercise.

You can 'cheat' a bit by thinking of ways of sending a message such that ~everyone on earth would get it at the delivery time, including the intended target. But personalizing the delivery (thus adding privacy) makes the problem far, far more difficult.

Yes, I dm’d Erik, the founder of Modal Labs, asking what I had to do to get a job there, and a few months later I had an offer. I was before that message unknown to Erik, working at Canva in Sydney, but the internet had told me you’ve got to put yourself out there, take your shot :)

A similar mistake was made by Ezra Klein in the NYT at the end of his opinion piece (1).

The 'we can regulate this' argument relies on heavy, heavy discounting of the past, often paired with heavy discounting of the future. We did not successfully regulate tobacco; we failed, millions suffered and died from manufactured ignorance. We did not successfully regulate the fossil fuel industry; we again, massively failed.

But if you, in the present day, sit comfortably, did not personally did not get lung disease, have not endured the past, present, and future harms of fossil fuels — and in fact have benefited — it is easy to be optimistic about regulation.

1. https://www.nytimes.com/2023/11/22/opinion/openai-sam-altman...

It’s 100% false consciousness. The Black Friday consumption is a significant carbon pollution event, but ‘conscious consumers’ (and engineers) appreciate the carbon offset trick.

On the one hand, the article and the talk it is about stresses the severity of the moral transgression: "moral meltdowns", "really crossed very bright lines", "ethical collapse".

But on the other hand, the implications of the moral failure for the moral status of the company and its employees is pushed far away into the corner:

- "These are great companies, great organizations, good people"

- "misguided companies"

- "good people at great companies"

The list that follows is something you can retroactively apply to numerous instances of corporate wrongdoing, but also provides an enormous amount of false positives and a false negatives.

Thousands of businesses don't meet these criteria and yet are morally compromised (e.g. Cargill). Thousands of businesses do meet these criteria and yet aren't going to be called out as "ethically collapsed" by the author before they've been outed and widely accepted as failed.

Googling Will Hurd only shows up a Republican politician with a history at the CIA. Is that the right guy? Can't be.

If the product got good enough you’d have a shot at selling to Canva.

Yes there’s a bug: https://thundergolfer.com/performance/profiling/2022/08/05/s...

This books appears to be using photography from docubyte.com.

I have two prints from the Guide To Computing series and they’re both excellent.

I don’t know the meaning of the detail either, but just because you don’t get why some detail was included does not mean it was unnecessary. It’s possible, even probable, that the detail has meaning to a certain subset of the article’s audience and that the detail aides the story.

Appreciate the references you provide in this comment.

I considered myself a longtermist, but climate change is the #1 risk to me, not AI, and I thought this prioritization was more common amongst the EA and longtermist communities. But my longtermist views come mainly from Derek Parfit, and it's probably the case that the contemporary longtermists are not of Parfit's kind, and unhealthily intertwined with billionaire weirdos (Theil, SBF).

> But longtermists have an answer to this conundrum: the so-called ‘value-neutrality thesis’. This states that technology is a morally neutral object, ie, ‘just a tool’.

This is indeed a stupid idea, but the writer's following examples don't show longtermists endorsing this view. What Toby Ord is arguing is about an incompatibility between technological advancement and public ignorance (invoking Sagan, as he should). As Margaret Mead said, you can have advancement in technology, but you "have to move the whole pattern". You have to change the social, political, and educational realms in concert with technological change. The editor of this piece really should have fixed this part of the essay, because it's obviously weak.

You say it’s more complicated than PG’d claim, but your “just a tool” narrative is a classic oversimplification of technology’s relationship with man.

A tool is an extension of man, and it is not additive as you suggest, even for “talented writers”. A tool interacts with man’s mind and body, and society’s fabric, “adjusting the sense ratios” of man, to use McLuhan’s phrase.

“Just a tool” is lazy thinking that is most effectively used to stifle debate about technology. This has happened with the typewriter, the gun, the computer. For a more sophisticated engagement with technology, see Neil Postman, Ivan Illic, Marshall McLuhan, even Frank Herbert.

Sounds reasonable, and worth nothing that TV's dominant use case is entertainment, whereas it is ChatGPT's minority use case.

ChatGPT is less applicable to _Amusing Ourselves To Death_ and most applicable to his _Technopoly_ book.

Taken straight from Wikipedia:

> This is exemplified, in Postman's view, by the computer, the "quintessential, incomparable, near-perfect" technology for a technopoly. It establishes sovereignty over all areas of human experience based on the claim that it "'thinks' better than we can".

In the computer age Postman experienced, the IBM database system was the dominant mode of computing technolopy, and had mainly the High Modernist faults of a false claim to control and order. ChatGPT and other generative models much more directly attack the supremacy of human thinking, and make it much easier for us to cede control.

Thanks for trying it out!

2-3c per run seems very high. That's probably just the cost if you have to spin up a new container. You can shorten the idle timeout on a container if its going to just serve one request typically. If it's going to serve more requests, then the startup and idle shutdown cost is amortized over more requests :)

You can get it running with one Python script on Modal.com :)

https://github.com/modal-labs/modal-examples/blob/main/06_gp...

> are worth avoiding so you don't get stuck with somebody else's framework.

Modal eng here. Modal is not setup as a framework. Think of more as Python-defined serverless infrastructure that has native support for the Python runtime. This is in some places called "Infrastructure from code", as opposed to "Infrastructure as code" which means just source-controlling K8s YAML and Cloudformation.

A major benefit of this approach is that the cloud becomes part of your dev loop, as opposed to doing `docker build`, `docker push`, `kubectl`, etc just to ship a change to a GPU.

In the script I posted Modal APIs are mixed in with standard Python code for brevity, but many customers just keep their code in their own modules and have a `modal_infra.py` module that defines the serverless infrastructure.

Ah nice. Thank you. I was using backticks

How come you didn't end up using Modal, seeing at it was recommended in the only reply in the thread? [I'm a Modal person looking for insight :)]

On Modal.com these 34 lines of code is all you need to serverlessly run BERT text generation inference on an A10G (which has 24GB of GPU memory). No Dockerfile, no YAML, no Terraform or AWS Cloudformation. Just these 34 lines.

  import modal

  def download_model():
      from transformers import pipeline
      pipeline("fill-mask", model="bert-base-uncased")

  CACHE_PATH = "/root/model_cache"  # model location in image
  ENV = modal.Secret({"TRANSFORMERS_CACHE": CACHE_PATH})

  image = (
      modal.Image.debian_slim()
      .pip_install("torch", "transformers")
      .run_function(download_model, secret=ENV)
  )
  stub = modal.Stub(name="hn-demo", image=image)


  class Model:
      def __enter__(self):
          from transformers import pipeline
          self.model = pipeline("fill-mask", model="bert-base-uncased", device=0)

      @stub.function(
          gpu="a10g",
          secret=ENV,
      )
      def handler(self, prompt: str):
          return self.model(prompt)


  if __name__ == "__main__":
      with stub.run():
          prompt = "Hello World! I am a [MASK] machine learning model."
          print(Model().handler.call(prompt)[0]["sequence"])

Running `python hn_demo.py` prints "Hello World! I am a simple machine learning model."

You can check out available GPUs at https://modal.com/docs/reference/modal.gpu.

There's also a bunch of easy-to-run examples in our docs :) https://modal.com/docs/guide/ex/stable_diffusion_cli

Wow the CTO of OpenAi seems to have ~1 yr of hands on engineering experience, followed by years of product and people management, That’s unexpected. I thought the CTO was Brockman.