r-zip

How stupid do these CEOs think people are?

They’re not wrong though. The frequency with which these things still just make shit up is astonishingly bad. Very dismissive of a legitimate criticism.

Unwillingness to adapt to the destruction of the middle class and knowledge work is pretty reasonable tbh.

What mathematical guarantees? I don't see any mathematical results in the paper.

They are NOT as unaware of things as we are. That’s like someone seeing a software developer googling stuff and saying “see, they don’t know much more than me”.

An expert refreshing their knowledge on Google is not the same as a layman learning it for the first time. At all.

in no universe is that a legitimate ML interview question

Why not? This seems like the ML equivalent of FizzBuzz. If you don't know how matrix multiplication works well enough to implement it, I would argue that you don't know what you're doing at all.

But they likely did have their wives do a lot of the domestic heavy lifting.

Damn, really?

I think the time estimates depend heavily on the field. A mathematician recently told him that it would take him 1-2 weeks (with nothing else to do in that time) to digest a paper outside his main area.

Did you read the abstract?

What are you talking about? Did we read the same abstract?

Even with an accuracy score, it’s meaningless. Accuracy isn’t the right metric.

This was also a problem a decade ago. I asked a professor for research opportunities and ended up segmenting a bunch of images. That was it. No credit for that work, either (or pay).

Self-supervised training doesn’t involve human labelers, and that’s what’s responsible for most of the recent gains in LLM performance.

Deep nets are notoriously overconfident in their predictions. So why do you expect this approach to succeed?

Would you say the same about golf?

There is an Apple Pay option. I just used it!

Wouldn't ">=" be "not <"?

You’re joking, right?

If you seriously don’t think this will be widely abused, then you’re extremely optimistic.

Spreading propaganda is the most obvious use case, assuming the resource requirement remains high. I don’t think it’s far-fetched either. If these models become less resource-hungry, then we’re in for much worse.

Serious question: do the people who work on this topic think they’re doing a good thing?

I guess that depends on your definition of “world.” For me, that would be a planet, not a moon.

Just ask it the definition of a mathematical term and watch it get it wrong. For example, I asked it to define the Kronecker product and it gave me a lot of correct information, but the actual equations were wrong.

The funny thing is that it shouldn't be. There is plenty of potential, but all sorts of (mostly non-technical) factors are getting in the way.

Can you recommend a few such places? :)

I think a big part of the problem is that most PMs are non-technical, and at some point up the chain so are most managers. Data science, when done thoughtfully, requires you to appreciate minute issues from data collection and management all the way through feature engineering, model selection/design, and validation.

The biggest, hardest bridge to cross was an appreciation of the importance of metrics. For some reason, getting a business person to grok something as simple as precision/recall/F-beta is a near-impossible task. You can do multiple presentations on it (after having honed those presentations over years with multiple manager audiences), and it never sticks. It's always "what's the accuracy?" It's impossible to do good work when your bosses insist on measuring and therefore optimizing for the wrong thing (which in my experience consulting for multiple Fortune 500 businesses, they always do).

Even worse, many organizations have such broken politics/cultures that the managers can't even tell you the big picture of what the project is trying to accomplish. Once you finally piece it together from the people who know their roles in-depth, it becomes clear that what they're trying to do is totally infeasible. At least that was my experience more than half the time.

Computer science...

where's the scientific method? Where's the experiments and rigor?

No time for that, buddy. CXOs need results, ASAP! (This was basically the attitude of my managers at the last place I worked).

Why do math on paper when you can write code and look cool (and not like a useless academic)?

That's all fine, as long as you still understand the underlying assumptions and pitfalls. Many people who skim documentation and throw things together haphazardly do not.