That’s just what they want you to think.

The original case was just bullshit made up by a bored journalist.

https://en.m.wikipedia.org/wiki/Murder_of_Kitty_Genovese

In 2007, an article in the American Psychologist found “no evidence for the presence of 38 witnesses, or that witnesses observed the murder, or that witnesses remained inactive”.

deleted by creator

Although it’s federated nature is kinda dying.

If you’re not on one of the major providers good luck getting people to see your email.

Except that in industry no one gets tenure.

As someone working and publishing in the field this is more a cyber jerk about American exceptionalism than actually true.

Chinese universities and companies publish a shit tonne at pure machine learning conferences. They absolutely do a large amount of research into the fundamentals of machine learning as well as the applied stuff. They’re probably the closest to the US in terms of having large firms that are prepared to bank roll the training of the very large language models.

Alibaba in particular has been constantly doing cutting edge stuff in terms of multimodal language models that are worth paying attention to.

The actual truth is that China does both kinds of work. Broad foundational and applied work lead by independent research groups in companies and universities, and focused application driven stuff for direct application by the state.

Google still stands out in terms of the amount of research it does, but this is because Google is different to everyone -other US research institutes don’t compare to it either.

(Swiss)Germans are completely mad about food.

It’s their culture to complain about everything, except food. All they care about is that it’s as bland as possible and has big portions. If you manage that, they’ll give you five stars every time.

I spent 3 years living in Germany, and not only can you not get anything spicy for love nor money, they also don’t use herbs. It just blows my mind. They’re physically so close to France and Italy, but the food is so far away.

The problem is that Tesla is a meme stock. It’s stupidly overpriced for a car company with such little production.

They could get rid of musk and try to turn it into a normal company with solid fundamentals but this would remove its meme status and crash the share price anyway.

There’s no way to fix Tesla without burning it’s stock price to the ground, so anyone involved is just going to try to ride the crazy train as long as possible.

Come on Elon, make the Cybertruck 2 and really fuck everything up.

Yeah anything responsible that reports and corrects mistakes made in past reporting gets a low score for factual accuracy, when these should be getting the highest scores.

It’s completely backwards.

No it doesn’t work.

But it’s better than not doing it.

People suspect who the author is but maybe you cited those papers because you’re afraid of getting the author to review them, or you’re a fan-boying grad student.

Yeah but they don’t use LLMs for this, they’ll use some other kind of machine learning mixed in a big pipeline of data processing. It makes it really hard to guess how much work it would take to fix. It might require retraining, might just require an easy patch of the rest of the pipeline.

My guess is that they’re just shitty jumpers and there’s nothing to fix anyway.

I’m at this point too. I think the next step is to just declare sock bankruptcy again and throw everything out and start over.

Yeah, but a combination of this approach, and adding all compiled file types including .pyc to .gitignore would fix it.

I’m not describing binary classification, I’m describing multiclass. “Group classification” isn’t really a thing. Yes, your ml system probably guesses what kind of plant it is and then looks up the ediblity of components.

The problem with this is how they will handle rare plants that aren’t in the dataset, or that are in the dataset but with insufficient data to be recognised.

Because multiclass assumes that it’s seen representative data on all possible outputs (e.g. plant types) it will tend to be dangerously confident on plant types it hasn’t seen before.

This is because it can rule out other classes. E.g. if you’re trying to classify as rose, tulip, or daisy and you get a bramble, your classifier is likely to be very certain it’s a rose because tulips and daisies don’t have thorns. So your softmax score is likely to show heavy confidence in rose even though it’s actually none of them.

This is exactly what can go wrong when you try to use the softmax/standard multiclass approach and come across an interesting rare mushroom or wild carrot. You don’t want it to guess which type of plant in the database it’s most like, even if this guess comes with scores, you want it to say that it genuinely doesn’t know and you shouldn’t eat it.

The key issue here is that ‘level of certainty’ doesn’t really mean what you would like it to.

You get back a number yes, but it can change according to what’s visible in the background, the angle that the plants at, how close is it to the camera, and how nice the camera is you’re using (professional photographers use expensive cameras and take shots of different things to everyone else).

Interpreting this score as “how safe is it to eat the plant” is a really bad idea. You will still eat the wrong plant. These scores can lead to very confident random guessing when you show it a plant it’s never seen before.

And no, softmax is a trick for making the scores all sum to one, so you get back a confidence for every possible thing the image could be of.

Teams defaults are pure scummery.

No, don’t alert me on a Sunday night with notifications that I might have missed over the last two days.

It’s worth saying that ml is in a very different position to most of academic publishing.

All of the serious journals are free to publish and fully open access and a significant amount of publication includes enough code that things are mostly replicable. GitHub has done wonders for our field. Also many tech companies use publications as an indication of prestige and go out of their way to publish stuff.

We’re still drowning in too many papers and 95% of everything is shit, but that’s every field really. Talking to musk on twitter is the not right place for a nuanced discussion about publication.

Work socks as well.

They’re socks that go with construction boots. Basically the same as hiking socks but cheaper.

It’s a consequence of parliamentary sovereignty.

Parliament can always dissolve itself and call an election, and it’s an important mechanism for getting rid of the government.

The problem is that the prime minister also has a majority in parliament, and that means he can make parliament dissolve itself when he likes.

This was actually a problem for Johnson. Initially, he didn’t have enough of a majority and it wasn’t clear he could call an election without Corbyn’s support.

Both. It’s like the saying “Governing a big country is like cooking small fish.” (With the explanation that if you keep poking it, it’ll disintegrate) also taught me how to cook fish as well as realpolitik.

The fish advice was most useful.