subharmonicon

As an Apple user since the launch of the original Mac, I guarantee there have never been any perfect Apple products.

Every Mac OS until OS X crashed continuously on even moderate use. Thankfully Steve did manage to build better systems at NeXT and brought that technology back when Apple acquired NeXT and was able to substantially improve Apple’s Mac product as a result.

It’s still far from “perfect products”, though, even during the time period that Steve was still alive and at Apple. To your point, I do agree that things have gone downhill quality-wise since his death, but I attribute this as much to being completely schedule driven as I do to the personalities involved in managing the projects.

Most execs at most companies really are driven by very short-term planning and thinking, as it benefits their career to do so. They just need enough good quarters/years to justify their promotion from VP to SVP or SVP to CEO. Nobody gets blamed and punished three years after they move on to a new role for the shortsighted choices they made in their previous role.

Having said that, Cutler is not one of those execs, but was certainly subject to pressure from above to get systems out the door and keep the revenue stream coming in. As he mentions early in the interview, he’s really bothered by shipping bugs, really disappointed with the quality of software engineers in the industry, and really bothered by program/project managers treating every bug like a rare corner case. I concur with him on all three. Most of the bugs I file are “corner cases”, and yet I hit those issues every day and my bugs have multiple dups in the database and/or my bug gets duped back to an early report.

Having spent a few decades in this industry, all of it in Big Tech, I have yet to come across a company that doesn’t fold to the same sort of pressure, even if the engineers and first-level managers are pushing to get bugs fixed before shipping, and pushing for more frequent bug fix updates.

Can you cite a reference explaining this ability to re-form new warps from existing threads?

I ask because I’ve seen posts from NVIDIA support saying that divergence is still very expensive and I’ve also seen benchmarks that force divergence in each warp by evenly splitting the warp, and the benchmarks result in 2x runtime when that happens vs. when the control-flow is dynamically uniform.

One thing to keep in mind is that even if you were to dynamically reform-warps, there’s still a potential expense because you then lose the advantage of doing things like accessing adjacent elements of data in adjacent threads. You’re bound to now have more bank conflicts, fewer memory accesses being coalesced, etc. Perhaps they do actually do this warp re-formation, but that itself does have this additional cost.

NVIDIA is intentionally being obtuse and frankly dishonest calling what’s effectively a vector lane a “core” and similarly using “thread” in “SIMT” to mean the execution of one of those vector lanes.

Yes, their architecture is different from many in that they support a separate program counter per lane (which is why they feel justified in calling this a “thread”), but ultimately it’s the rate and throughput of ALUs that matter.

> Or rather what it will become once the VCs who are paying for all this start the squeeze.

How do we know that squeeze isn’t already happening?

The mojo launch was extremely underwhelming, and I wonder if it was forced by early investors to help drive hype and the next round of funding.

> It’s not just weird, it’s an intentionally dishonest evaluation to build hype around their language.

Not too dissimilar to Swift, which was launched as being faster than C/C++.

What if it’s less his choice and more about convincing investors that it makes sense?

Just paid $30k for a variety of tests and a surgery for my dog.

I have insurance, but the yearly coverage limit is $5500.

Some insurance claims to have no limits, but then turns around and denies many claims.

I’m just glad I could afford the surgery and medications to keep my dog out of pain and help him have a few more happy years.

I have no idea what you mean by “how any of this works”.

I didn’t bring up solving the phase ordering problem, you did.

I’m simply pointing out that if you have a compiler where you have multiple IRs or dialects of IR, and you have a pass that is written to work on IR “X”, and then at some point after that pass you translate to IR “Y”, if you want to move your pass after that point of translation, you either need to rewrite your pass so that it operates on “Y”, or you need to translate back to “X” again.

> You realize this is only feasible if you have one team working on a compiler for one domain right?

It sounds like you think I’m advocating for something. I’m not. These are all just engineering trade offs that depend on your goals.

Regarding phase ordering: A single IR allows you to freely reorder passes rather than having to reimplement them if you want to move them earlier or later in the phase order.

Yes, I completely understand the reason for the dialects. I’ve been working on compilers for decades, and what’s interesting about MLIR is that it’s a bit of an anachronistic approach. People used to write optimizers with multiple levels of IR, often based on the same data structures (but sometimes not), and they were effectively different dialects in the same sense as MLIR.

That fell out of favor due to the fact that you end up having to choose and commit to the phase ordering fairly early on and writing separate lowering steps to convert between dialects. So the tide turned toward having a single mid-level IR (and sometimes a single high-level IR for things like specific loop optimizations that was then lowered to that mid-level IR).

It’s not clear that all these dialects are actually helping things, or that the design of something from which it’s easy to make new dialects is that helpful.

Why?

I attended an MLIR online meeting months back and there was someone presenting who had a diagram that literally had a dozen different dialects and converters that were being written between different ones (but not MxN, so it was a strictly relatively arbitrary set of combinations that were being converted between, and mostly one-directional if I recall correctly). This was the “solution” to being able to make various things interoperate, but from a distance it looks a lot like a relatively obvious problem caused by the very nature of the design of MLIR.

I don’t get the sense that this is what the parent comment is talking about at all.

Not to mention that GPUs already execute in-order (at least any that I’m familiar with). They do have multiple execution pipelines, but instruction fetch/decode is in-order unlike something like a typical modern high performance CPU.

It’s not that simple, although it may seem that way for some people.

Twice I have been on very extreme calorie counting regimens for months at a time combined with substantial weight lifting and cardio.

Both times I have observed that if I reduce calories the first thing that happens is I start to feel colder in the same environments. My body adjusts to my new caloric intake by reducing my metabolic rate it seems.

I learned about the idea of a “set point” fifteen years ago and I believe it’s related to what I’ve observed.

I have friends who can do something as simple as cutting out a morning cappuccino and they lose weight as a result. If I do that, there is no measurable difference unless I cut much deeper and substantially increase activity at the same time. Then the weight comes off for a bit and then plateaus again, sometimes for weeks or until I make another adjustment.

One of the times I went through this weight loss regimen I had an air displacement plethysmograph near the end of the weight loss that concluded I had 178lbs lean body mass. I had likely bulked up a little through all the weightlifting I was doing but have always had a fair bit of muscle so you might expect my resting calorie burn to be very stable and for small adjustments in calories to not impact resting metabolic rate, but that’s not what I observed in practice.

It’s hard to fathom how the upper management of these companies couldn’t see this coming when they were binging on hiring.

Did they really believe we were in some new era of pandemic-fueled hyper-growth and that they would be less competitive if they didn’t grow by tens-of-thousands of employees?

I remember in the late 70’s or early 80’s riding my bicycle to the convenience store to buy my mother a carton of cigarettes.

Technically I think you needed to be seventeen to purchase them in Michigan at that time, but I just told the person behind the counter I was buying them for my mother. I was around ten years old.

> Although the findings are deeply fascinating the researchers do indicate they need to be interpreted with caution. The cohort of subjects is incredibly small, and while the researchers did try to match the control group as closely as possible to the monks the findings cannot directly conclude the microbiome differences detected are solely caused by meditation.

Discussed here: https://news.ycombinator.com/item?id=34391043

Also submitted here: https://news.ycombinator.com/item?id=34409204

and here: https://news.ycombinator.com/item?id=34409860

Previously discussed: https://news.ycombinator.com/item?id=34391043

Previous discussion: https://news.ycombinator.com/item?id=34391043

Hadn’t seen that - thanks for the link to the previous discussion!

From the draft recommendation, it looks like this is the eligibility criteria:

VIII. Eligibility for Reparations

Checklist for eligibility:

REQUIRED:

- An individual who has identified as ‘Black/African American’ on public documents for at least 10 years - 18 years or older

You must meet at least TWO (2) criteria from the following list (must have supporting documentation):

- Born in San Francisco between 1940 and 1996 and has proof of residency in San Francisco for at least 13 years

- Migrated to San Francisco between 1940 and 1996 and has proof of residency in San Francisco for at least 13 years

- Personally, or the direct descendant of someone, incarcerated by the failed War on Drugs

- Record of attendance in San Francisco public schools during the time of the consent decree to complete desegregation within the school system

- Descendant of someone enslaved through US chattel slavery before 1865

- Displaced, or the direct descendant of someone displaced, from San Francisco by Urban Renewal between 1954 and 1973

- Listed, or the direct descendant of, a Certificate of Preference holder

- Member of an historically marginalized group that experienced lending discrimination in San Francisco between 1937 and 1968 or, subsequently, experienced lending discrimination in formerly redlined San Francisco communities between 1968 and 2008

https://sf.gov/sites/default/files/2023-01/HRC%20Reparations...

It’s not clear from the article that it’s a requirement. It says “may”:

“They may also have to prove they were born in the city between 1940 and 1996, have resided in San Francisco for at least 13 years, and be someone, or the direct descendant of someone, incarcerated during the war on drugs.”

EDIT: Clarified here: https://news.ycombinator.com/item?id=34391669

Here’s one example: The Unlikely Story Behind Japanese Americans' Campaign For Reparations

https://www.npr.org/sections/codeswitch/2020/03/24/820181127...

“… 40 years after the internment camps closed—President Ronald Reagan signed the Civil Liberties Act, which offered a formal apology and paid $20,000 to each survivor.”

Title modified to fit limit.

Originally: “$5 million for each longtime Black resident? S.F. has a bold reparations plan to consider”

Not disputing this, but my observation is that if that’s true he is hardly alone amongst people working in this space.

I’ve seen multiple people over the last 15 years say that we’re 3-5 years out from a breakthrough that will result in 120-200 year lifespans. Yet those breakthroughs have yet to materialize.

Of course that alone doesn’t mean they are trying to scam people or are perpetrating a fraud, but there seems to be a lot of hype with little to show.

Rather than “well meaning members of the public”, I think the larger problem is with health and science reporting which frequently reports findings of very small studies as if they are definitive fact, with very prominent and provocative headlines.

From what I have read in the past this is just how the funding tends to work. You write a grant proposal for a study with 20 people, get funding, get results, use that to get funding for another study with 200, then 2000, etc.

Frequently they don’t manage to make it past the first couple levels because the results aren’t interesting.

This was just step one: https://www.nytimes.com/2022/11/04/technology/elon-musk-twit...

Many more let go since, and many resignations as well. Perhaps you’re nitpicking layoff vs fire, but the company has far less employees now than before Musk took over.

I’ve always enjoyed “programming in the small” more than “programming in the large”, especially when the latter is in a position where I am part of a huge machine shipping a huge product to a large number of users, which has been most of my career.