antipaul

double check your math

153 * 20.8 million shares is $3.2 bill

Yep. At this point, I essentially don't trust any ML result that shows > 95% accuracy.

So often, those models proved to be over-fitted and not generalizable.

But too many decision makers simply can't properly judge such results.

Or, bring them back to the office where they will still slack off

Are you not turning kings into emperors, by flattening layers and therefore increasing the number of people who report to a director or VP?

I mean in general, not necessarily at Meta.

Is it possible to do both?

I hear senior VPs like Craig Federighi, Jeff Dean, and others, still write code (maybe 1 day a week but still), but also call the big shots

Is this note not entirely, but still a bit of an over-reaction, given that the post-layoff numbers are still much larger than ~2 years ago?

Was it command-F or control-command-f ?

I thought the old one was the latter - and it still seems to work

(Though I admit I am not worry-free when it comes to seeing globe-f)

As long as you do something that was _not_ in the training data, you’ll be able to fool the AI robot, right??

Respect.

Also in article:

“The last big round of layoffs at Apple happened way back in 1997, when co-founder Steve Jobs returned to the company, which then cut costs by firing 4,100 employees.“

Which MBP model? I think it may matter

That's from Manhattan, not Brooklyn, but maybe close enough

“all” = only Mac Pro, which has an Intel processor rather than new Apple Silicon

Options:

* you can have more than one device

* you can have a trusted contact to get you the keys

* you can have recovery codes

I predict this will be possible

They already communicate to you when you use a newer iOS feature that won't work on your other device. But you can still use it. Maybe you won't be able to turn back on those features on the older device, or something

Can someone say more on this point?

If there is a "Revolution Plan (WIP)" document shared amongst a few agitators, and someone in power gets their hand on it (and its "checksum" or whatever), then can they figure out _who else_ has it?

I think this is the part that is not legal, and which they say the don't do, etc

* Amazon uses third-party seller data to copy the site's most popular products, an antitrust report by the House Judiciary Committee alleged on Wednesday.

* Former Amazon sellers told an antitrust subcommittee the company released new products almost identical to their own and "killed" their sales.

* Amazon has denied accusations of this behavior in the past. "We have a policy against using seller-specific data to aid our private-label business," Amazon CEO Jeff Bezos said in July.

https://www.businessinsider.com/amazon-uses-seller-data-copy...

The Missing Readme” book seems right up your alley

https://nostarch.com/missing-readme

“The Missing README fills in that gap—a distillation of workplace lessons, best practices, and engineering fundamentals that the authors have taught rookie developers at top companies for more than a decade”

Get it just for the “Design Doc” template!

I love the “manager as problem solver”.

It’s a somewhat tricky balance between “micromanagement” and problem-solving, but then problem solving is a bit of a discipline after all.

More details, from the people behind Apple’s internal leadership training:

Ever since Steve Jobs implemented the functional organization, Apple’s managers at every level, from senior vice president on down, have been expected to possess three key leadership characteristics:

1. deep expertise that allows them to meaningfully engage in all the work being done within their individual functions

2. immersion in the details of those functions;

3. and a willingness to collaboratively debate other functions during collective decision-making.

When managers have these attributes, decisions are made in a coordinated fashion by the people most qualified to make them.

Article: https://hbr.org/2020/11/how-apple-is-organized-for-innovatio...

If you do it right…

That’s the problem: these metrics often come from overfitted or in-sample data, and are completely unrealistic when it comes to expected generalization performance.

I’m at the point where I never trust performance metrics anymore. Or rather, the worse they are, the more I trust them!

Enlightening sociological reflection!

It seems that individuals often bemoan such a lack of consequences, but for some reason they are still quite prevalent in our “systems.”

I wonder how to harness the good intentions of individuals…

While the tone is a bit too negative (expected of someone leaving one place for “greener grass”), there are terrific points here that completely resonate with my own experience in data science - albeit not at a “median” place, but at a large non-Tech corp.

The biggest point I’d emphasize: “there is a general industry-wide need for people who are good at both data science and coding to oversee firms’ data science practices in a technical capacity.”

Even more, most of the data tech leaders I hear strongly suggest this is not possible without C-suite representation of a data/engineer expert (I’m talking about non-tech companies)

Finally, it’s actually not uncommon to see people shift from data science to data engineering due to similar motivations. It’s actually the technical leadership that is sometimes surprised at the shift. You hear about this in podcasts from DS/DE people.

What's the "explain it like I'm 5 years old" (ELI5) for tree-sitter? Why should I, an emacs user but not lisp hacker, care about it?

Google, Facebook, twitter and the rest

Eg, Google has 80-90% (!) income from ads (incentive)

Yea, wasn’t this experiment ran on jailbroken devices?

Everyone uses data they collect to “improve” their own offerings.

But the others then make money by sharing/selling the data _to third parties_(incentive).

Does Apple use this data only for internal use, or do they also share (sell) it to third parties? With or without privacy?

Not convinced by the example. I don’t see how you can’t use standard scikit-learn for it.

First, the example doesn’t take advantage of sklearn’s built in, super simple parallelization via n_jobs

Then, the entire example could be better wrapped with sklearn’s own cross_validate() which gives you the same functionality: a table of results across experiments.

If you use a different estimator, you can easily concatenate the results into a single df

The rest is the same.

Why you need SQLite for this? (SQLite is great of course for the right use cases)

And if you're doing many orders more experiments (1000s instead of 10s) then that’s probably where MLflow is good (haven’t actually used MLflow)

Thanks

I gave up after 45 minutes... They want so much information/cookies/tracking but even after changing browsers and (temporarily) disabling all security/privacy, I couldn't download.

Strong disagree.

Many small companies or even developers have clear and clean websites.

It's not a question of money. It's a question of taste.

I kind of think this is because Tesla people looked at Apple.com, and (roughly) copied it.

They are smart that way. Other companies would do well to try something like that. But they aren't that smart.