I think this is akin to x% of the worker ants doing all the work. Once you get to a big enough scale and have to delegate I'm sure every company hits this.

I just wish we didn't have to rely on hiring 100 on paper workers for 5 excellent people committed to the company...

Which when it leads to abuse it's saving face and when it's incompetence it's saving face.

For a competent doctor it's used too let a patient know they're doing their job and an acknowledgement of symptoms.

Unfortunately to a _lot_ of the field "catch-all" "diagnoses" (in intentionally separating these labels). It's the same as diagnosing someone with chronic fatigue. It's diagnosing via exclusion.

The difference between chronic fatigue and brain disorders being that you're more likely to get someone looking to make a "name for themselves" diagnosing or curing the latter vs the former...

This is basically just a rehash of "trained" DNN are a function which is strongly dependent on the initialization parameters. (Easily provable)

It would be awesome to have a way of finding them in advance but this is also just a case of avoid pure DNNs due to their strong reliance on initialization parameters.

Looking at transformers by comparison you see a much much weaker dependence of the model on the input initial parameters. Does this mean the model is better or worse at learning or just more stable?

This is an interesting insight I hadn’t thought much about before. Reminds me a bit of some of the mechanistic interpretability work that looked at branch specialization in CNNs and found that architectures which had built in branches tended to have those branches specialize in a way that was consistent across multiple training runs [1]. Maybe the multi-headed and branching nature of transformers adds and inductive bias that is useful for stable training over larger scales.

[1] https://distill.pub/2020/circuits/branch-specialization/

1) routing (mis-)config problem - key of remote exploit. This should always be something people double check if they don't understand how it works.

2) hard-coded secrets - this is just against best practice. don't do this _ever_ there's a reason secure enclaves exist, not working it into your workflow is only permissible if you're working with black-box proprietary tools.

3) hidden user - this is again against best practice allowing for feature creep via permissions creep. If you need privileged hidden remote accessible accounts at least restrict access and log _everything_.

4) ssrf - bad but should be isolated so is much less of an issue. technically against best practices again, but widely done in production.

5) use of python eval in production - no, no, no, no, never, _ever_ do this. this is just asking for problems for anything tied to remote agents unless the point of the tool is shell replication.

6) static aes keys / blindly relying on encryption to indicate trusted origin - see bug2, also don't use encryption as origin verification if the client may do _bad_ things

parsing that was... well... yeah, I can see why that turned into a mess, the main thing missing is a high-level clear picture of the situation vs a teardown of multiple bugs and a brain dump

>if they don't understand how it works.

The problem quite often is they think they know how it works, Dunning-Kruger effect and all.

Outside the realm of the testable isn't worth discussing to experimentalists so might as well be a non quantifiable field.

Although sociology is perfectly quantifiable and measurable. Even though arguably the underlying relationships between the measurements are extremely difficult to extract.

A better example is pure philosophy and maths rather than sociology to particle theory. But then again, nobody ever accused QFT of being too simple, so maybe I'm arguing against my own point there.

> a philosopher turned science journalist,

Interesting, I read it as the other way round.

I wonder which of the many worlds is correct :p

The moon example is painful, but I was assuming to be a "if the tree falls in the forest... yada yada yada..." Example to justify words on a page. Although at the time my brain was screaming about things like tidal forces and gravitational effects, asif I was about to start discussing the retrograde motion of Venus with a flat earther who doesn't actually want to learn anything with rigour...

Personally I'm more worried by the comparison of Planks constant in the small to c in GR. Yes they represent asymptotic limits in many regards but are certainly not equivalent imho.

>> a philosopher turned science journalist,

> Interesting, I read it as the other way round.

I cheated and looked at the author's bio. :)

> That's a year into its availability in France. Anyone who didn't have their first dose by then probably wasn't getting a dose.

You are aware of the "incentives" offered by the French govt?

Such wonderful options as the ability to go the shops without being arrested that came with, "take the mandated medicine".

That's entirely irrelevant to "what date do we pick as the cutoff for this research?"

From the chart, they picked a very reasonable spot to draw said line.

Explain how? there is a right answer but you'll probably not get it by relying exclusively on the reported data.

NB: most people choosing not to take it in France tend to fall into the medically at risk, stubborn, or, "so far down the rabbit hole that you probably can't trust these people to make sensible life choices" groups. (This alone being a good reason why this 'control' group had a slightly higher all cause mortality at 6months)

Remember, France was one of the wonderful countries where you couldn't legally shop or work if you were deemed to be 'not at risk' && 'unvaccinated' and achieved a very high rate as a result biasing the control group. (This is a purely statistical statement)

And for reference, I do think the vax is dangerous in terms of massive populations and we don't have mass graves due to mRNA problems (although several large cancer blips). In the same way in countries with low vaccination rates we don't have mass graves at 10% population or higher. Cv19 was always going to kill and an untested treatment is likely to kill those who were at risk.

(I'm willing to bet in the case of cv19 the ones who were hit hardest would have been hit badly by either vector, virus or mRNA. But we'll pretty much never be able to prove or disprove that...)

I'm sure both extremes will jump to the rallying cry of "2 more weeks..." So yes of course I'm wrong, I only worked on analysing early 'data' and pulling apart the models so what do I know.

> But we'll pretty much never be able to prove or disprove that...

How convenient for you...