ozb

Personally I'm more interested in technically complex projects, and systems programming more than web apps, vibe coding tools, or MCPs.

What tools have you used? What tips do you have?

For me this would include a couple of fuse filesystems, inter-language bridges, ORM adapter, a good chunk of app/framework code, and a lot of small platforms scripts that I probably wouldn't have bothered with if I had to look up everything instead of using AI. One notable aspect is that I've used a lot of Rust, even though I don't actually know Rust.

I do make sure to understand each block of code if not every line, and more carefully review the tests, which I generally do in a language that I do know. I use primarily aider-chat. Tips would be TDD, systematic design-architect-implement-troubleshoot and propose-criticize-refine loops, and multitasking (tmux) to avoid always waiting on an LLM response.

Please someone create a fuse filesystem that exposes k8s, AWS/GCP resources, Kafka topics, etc.

For example, this illustrates what I mean: https://github.com/ozzieba/k8sfs

can be nice to use regular CLI tools like find/grep/diff/jq rather than a separate workflow for each platform. Something something "everything is a file"

For many people, buying a lottery ticket is rational, in the sense of the Kelly criterion.

The Kelly criterion says that the strategy of maximizing expected value will lose 100% of the time to a different strategy. The most general form that I know of the winning strategy is: minimize your expected time to reach $X, for arbitrarily high values of X, eg $1M or $1B (ie, in the limit to infinity). Following this strategy will eventually beat any other strategy, 100% of the time.

For many people, the expected time to reach say $10 million is infinite: they simply cannot produce more than they consume, and in 10 years (or 100) they expect to be no better off materially than they are today. Any amount they try to save will eventually be used for some emergency (eg car breaks down, or health, or eviction/rent going up, etc).

Even assuming that the expected value of a lottery ticket is deeply negative, it is then Kelly-rational to buy lottery tickets, or more generally take a chance on any possible way of getting out of this equilibrium.

Of course I am handwaving away certain reasonable objections, and the argument can fail if indeed there is a chance of leaving the equilibrium otherwise, and the price of a lottery ticket is too high; but ultimately I believe the argument does apply to a not-insignificant percentage of buyers of lottery tickets.

This does hinge on the fact that in practice you can always "afford" to buy one lottery ticket per $period, yet you may not be able to save $2 per $period without ultimately having to raid the piggy bank for some emergency. So the lottery can be thought of as a creditor-/bankruptcy-/emergency-proof savings account

As a high earner with relatively modest assets so far, I often wonder about how much I should be spending: if my income were reliable then it would make sense to smooth consumption over the long term, ie spend a large portion of my income (and only save for known and unknown future expenses). But in practice it's more prudent to also save for unexpected changes in income, either from being laid off, or economic recessions, or long-term change in demand for my skills; there's disability insurance, but that's just a part of the picture.

So, is there any way for me to say "give me $X00,000 now, for 10% of my income over the next 10 years"? Obviously details would have to be worked out, eg pre- vs post-tax income, potentially adding a large "standard deduction" before earnings are shared, and a minimum yearly income for the sharing to take effect in a given year and count towards the 10. Target yield would presumably have to be around 10-15%. If there were such a deal available, would you as a high earner be interested? I'm wondering if I could make something like this happen for a group of us via private investment (think Yieldstreet and similar platforms).

Looking at https://news.ycombinator.com/item?id=33222687, and inspired by https://www.thirtythreeforty.net/posts/2019/12/my-business-card-runs-linux/ and https://pockit.ai/, I'm wondering what it would take to create actual cards with something like Pi Zero or even Arduino based logic, implementing basic Unix commands, data structures, and I/O (keyboard, lcd, BLE<->smartphone app, etc); the goal would be that each card is sub-$1 and ideally significantly less (for the basic cards, some might be more), and the target market would be some combination of education, fidget/puzzle toy, and perhaps just a fun to interact with these kinds of components in a satisfying tactile way. Could also have a way to convert to a running program. Thoughts welcome.

Given:

- ~Most of the Fed's balance sheet is government debt

- ~Most government debt is held by the Fed

- The Fed executes monetary policy by buying and selling (mostly) government debt; expansionary policy consists of buying Treasury debt, thereby raising its price / lowering yields; this effect seems more direct than the effect on the Federal funds rate

Mathematically, I'd expect:

Nominal Yields on Treasury debt = (future dollar value of debt) / (current dollar value of debt)

= ((future dollar value of debt) / (future value of debt, measured against basket of goods)) * (future value of debt, measured against basket of goods) / (current value of debt, measured against basket of goods )) * ((current value of debt, measured against basket of goods )/(current dollar value of debt))

= (future CPI) * (real yields on Treasury debt) / (current CPI)

Or, rearranging:

Inflation=(future CPI)/(current CPI)=(nominal yield on Treasury debt)/(real yield on Treasury debt)

In particular, I'd expect the "real" yield here to denote the "natural" yield you'd expect if the Fed was not buying debt for much more money.

So, the narrative becomes:

Government spends a lot of money, and market doesn't believe it's a good investment (in terms of effect on future taxable revenue base), therefore Treasury borrowing costs should go up; but Fed forces Treasury yields to stay around 0, so inflation goes up instead.

Does any of this make sense, or am I missing something obvious?

Is there a source that formulates the issue in this way?

The modern scientific publishing industry charges seemingly exorbitant subscription (or author) fees, given its reliance on unpaid academics for not only the research itself, but also editors, board members, and peer reviewers. The Journal of Machine Learning Research seems to provide evidence that this service can be provided for a fraction of the cost (<$10/article) at least in some circumstances; on the other hand, the non-profit PLOS network still charges $1339/article for open-access publication. Could we do better, at least for some larger segment of the market than is currently served by no-fee open-access journals? What services would have to be provided? How much would that actually cost, in money, volunteer hours, or compute resources?