mrfox321

Some of those high coupon CDs came from really distressed banks (e.g. Zion)

Your observation is likely over fitting to some of the CDs tied to risker banks.

It give banks far too much leverage.

If they took large losses or lent out too much, inflation would skyrocket.

back down to 4.3 for short term t-bills

yields dropped almost 1 point after depositor bailouts for SVB. the running belief is that the fed will slow down in response to fixed-income positions held by banks.

I had a similar thought.

It's equivalent to buying the bond that represents the debt you owe.

Larger debt ends up in negotiations over the amount owed. War debt after WW1 resulted in greatly reduced repayments based on the original lending.

If you are a default risk on debt that has weak collateral, you could probably come to some agreement that pays back some fraction of the principal.

Good point.

Although doesn't that defeat the purpose of structured concurrency?

i.e. creating tasks that are plumbed to some background thread/executor.

By directly invoking the coroutine, lifetime issues become much easier to avoid.

Users of coroutine-friendly libraries will have to adopt the async function coloring (co_await, co_yield, co_return, Task<T>).

So saying that coroutines are low-level / not application code is misguided.

Is the high level structure of writing any more important than the low level structure?

Maybe it will tend towards everyone sounding the same. Which then gets amplified as LLMs train on writing that is essentially mutations of itself.

Me no like.

"Don't ask me how the economy works"

Just look at Apple's recent job postings. Here's one, but simple searching reveals the pattern:

https://g.co/kgs/tV5rGG

They need

1. Recommendation systems experience

2. Machine Learning

3. Auction theory

4. Ads pacing

5. ML / Reinforcement learning

Gee, I wonder what they're scaling out...

Maybe as an edge case. I think this argument underestimates survival instincts.

The honest headline:

Let's define "long haulers" as those exhibiting system after Y months.

The percentage of long haulers is 0.25 <= X <= 1.

We need to replace 0.75 with

0 <= 1 - (X / 0.25) <= 0.75

I've heard another interesting argument, at least for disruptive science/research from industry.

Lower corporate taxation incentivizes the hoarding of cash. Prior, corporations were incentivized to allocate more funding towards their industry research (IBM, Bell Labs, etc).

Would love some counterarguments, here!

Meta deploys as "microservices" with multiple monorepos...

Binaries are deployed and scaled independently as thrift services.

Tons of rpc.

At my bigco, there's no service-level DI (c++)

Just vanilla C++ classes, and virtual interfaces if we need to mock things for unit tests.

No automatic wiring of the hierarchy.

Yeah, I mean, the computation can be re-materialized.

DAG may too specific. It's really a dependency graph, that likely has a DAG topology.

Sure.

Open source analogue would be Apache Airflow.

Abstractly, it's some directed acrylic graph (DAG) that is asynchronously computed, sometimes on a schedule.

Unfortunately, most things fall under DAG. But the framework / engine exists to manage the complexity of the ever-extending pipelines declared by the engineers

Event/push-based workflows also fall under this taxonomy.

To expand acronyms / jargon

IDL := Interface Description Language

TLA := linked article

thats just homogenous Sequence[Optional[T]], though.

Since manufacturing two things is more expensive than manufacturing one thing?

All new c++ code I work with at bigco are c++ coroutines.

The underlying executors come from folly.

Hybrid?

Request passed to a thread pool. Each thread is juggling coroutines representing in flight requests

It's slower than a row lookup.

Specifically, this model is an ensemble of decision trees.

This involves

1. Row lookup

2. N (tree depth) inequality checks on fields in the row for M trees

3. Weighted sum over M trees

e.g. -> example

i.e. -> in other words

Your comment is off topic.

D wave is not a quantum computer. It is a specialized hardware for solving certain classes of optimization problems using quantum annealing.

Less users. Smaller matrices.

Visited SF recently. Did things always close early?

But that's because the laws on software developed prior to tech becoming as powerful as they are now.

The part that was left out is the fact that conjugating is the group action of a group on itself.

in his example, let's say there's some 3-cycle (123) the cuber knows. But really, the cuber wants to transform like (479).

Conjugation being equivalent to a group action implies the cuber only needs to find a group that permutes 1->4, 2->7, 3->9.

Let's say one is found: (14)(27)(39)...

gTg' => [(14)(27)(39)...](123)[(14)(27)(39)...]' = (479)

You need a large time window or error bars will be large.

Annual numbers are also canon.

People think on yearly timescales