antipaul

When was AI supposed to replace radiologists? Was it 7 years ago or something?

So who among current ML practitioners building “useful” ML could solve some of these?

_Should they_ be able to?

Don’t forget:

Good design comes from good designers

(Fred Brooks)

That's how I make my grind easier, in Leetcode. Go direct to solutions first, to get intuitions.

How do you know the dude in blue is a fascist, and a jerk?

They've had facial recognition in Europe at airports for a few years now.

At least if you are EU citizen, you don't need to interface with humans.

I believe, though not certain, that you can opt out simply by going to the human booths (right?)

From “big program, small data” to “big data, small program” seems like a useful way to summarize the main shift from elaborate rules in the first generation, to huge piles of data today.

Can a "class", eg in Python or Java, be considered an example of the "data structure" Linus and others are talking about here?

Or are they only talking about tables in databases and such?

Like others say, usually just go for XGBoost, which is increasingly even shown (proven?) in literature, eg: https://arxiv.org/abs/2106.03253

You can also start with scikit-learn's `LogisticRegressionCV`, if a linear model is more palatable to your audience.

The bigger challenge is to reliably estimate how good your model is.

It's not about getting the best performance – it's about getting the "real" performance. How will your model _really_ do on future unseen datasets?

The answer to this challenge is cross-validation. But what are the questions?

There are 2 very different questions for which the answer is cross-validation.

One is, which hyper parameters to use with your model?

The second is, what is the generalization performance of the model?

This requires 2 separate applications (loops) of cross-validation. The authors of this book talk about this in terms of having a "validation set" and a "test set" (Sometimes, these terms are switched around, and there is also "holdout set". It's critical to know how you, and the rest of your team, are using these terms in your modeling)

A robust way to implement these 2 CV's is with nested cross-validation, readily available in many packages – and also, should be "fast enough" using modern computers.

One exercise that remains is: with nested CV, which model do you pick as your "production" model?

That is also a bit tricky. Reading things like the following can help: https://stats.stackexchange.com/q/65128/207989

EDIT: for those inclined, here is a paper on why you need 2 loops of CV: "On Over-fitting in Model Selection and Subsequent Selection Bias in Performance Evaluation" by Cawley and Talbot: https://jmlr.csail.mit.edu/papers/volume11/cawley10a/cawley1...

What do you think chatGPT uses as training data?

The whole world’s “sh*tposting”: Reddit, blogs, and the rest of the internet.

But also books and Wikipedia and what not.

You can “smooth” all the crap out via the training procedure.

But even more, Slack can easily filter training data to, say, only posts in high-use channels.

Further, slack has other options: eg, use their customer data only for marginal fine-tuning, for example.

Or, they don’t even know their use case yet - but want to wrap their arms around your data pronto.

To copy URL from Safari, press and hold (long-press) on URL bar

Press and hold also allows options elsewhere in iOS

Video showing Apple Pencil Pro features was pretty sick, and I ain't even an artist

Regarding exiting from fullscreen, you can suggest to them to "hover" over the traffic lights, for some extra tips

We run on GPUs. The real question is, what does the AI that runs us, run on?

Maybe companies should lift a finger to create work spaces that, well, work?

How do they collect the money from the external payment option and external app stores? How do they know how much those transactions are?

Your point is that they suck at software right? Because that’s my point :)

As a big Apple user, I think all three are very possible.

Just one tiny example: see the manual dial or “crown” on Watch, AirPods Max, and now Vision Pro. You could easily imagine those products without an unnecessary manual “dial” when there is both voice (Siri) and software controls. But, you still have a “dedicated knob”.

Agree. "software is eating the world" someone once said a decade ago, and yet most car companies still didn't get the memo. MBux, anyone?!?

Personally, if Apple creates a standard car simply with a great software experience (I'd expect at least a few small novelties), and not too expensive, then I am totally in.

Why can't they say which company made it?

Can we get checkout video recording out of other grocery stores too?

Hmm, what was that tech from IBM deep blue, that apparently Google leveraged to such a degree?

Was it “machine learning”? If so, I don’t think that was actually the key insight for Google search… right? Did deep blue even machine learn?

Or was it something else?

But Microsoft reinvented itself with precisely leadership change in Satya, right?

Snippets that stood out to me:

Google's culture eroded. Decisions went from being made for the benefit of users, to the benefit of Google, to the benefit of whoever was making the decision

The effects of layoffs are insidious… people can no longer trust that their company has their back, and they dramatically dial back any risk-taking. Responsibilities are guarded jealously. Knowledge is hoarded, because making oneself irreplaceable is the only lever one has to protect oneself from future layoffs. I see all of this at Google now

They found a monolith, but are saying it’s an epidemic

Is it AI, or just a look up table?

Not sure exactly, but the whole thing was "shot on iPhone" so maybe related

No italic

"Management" definitely seems easier, than having to do technical work, without proper accountability.

But it seems it's hard to hold management accountable.

For an IC such as an engineer, if you don't deliver code for the deadline or check-in, it's clear you haven't delivered.

But how do you judge the performance of managers? Oh, right, by the size of the "empire"...

Were they "working hard" on individual-contributor-style work, or on management?

"They can’t peer into everyday lives of the lower levels" - I'd imagine this is exactly what management should work "hard at".

If you're not keen to do that, then maybe be a high-level IC?