Transhuman enthusiast

[Transhuman enthusiast @EWenxue]

Why does @ArtificialAnlys rate Gemini 3.6 Flash to have a higher intelligence than 3.1 Pro?
Comparing them head-to-head 3.6 Flash is just really stupid, it constantly gets logic wrong that 3.1 Pro aces surprisingly well.

[Transhuman enthusiast @EWenxue]

@tmcoggiola for me no, 3.6 flash sucks and is really stupid, 3.1 pro is much smarter

[Transhuman enthusiast @EWenxue]

@Xander_J_C Good argument, should be extended to neonaticide.

There is no definite difference before and after birth, the baby is still something that has no sense of self or personhood.

Fear and care burden of severe birth defect scenarios would decrease, leading to more babies overall.

[Transhuman enthusiast @EWenxue]

@martytaka777 とんでもない数字です、８０万円を以下を稼ぐ外国人は到底高度とは言えないだろう

[Transhuman enthusiast @EWenxue]

@davidasinclair @cremieuxrecueil https://t.co/uuyxruO870

[Crémieux @cremieuxrecueil]

In July of last year, a paper came out that caused a lot of alarm about GLP-1 drugs.

The paper seemed to show that GLP-1RAs were causing extremely high amounts of blindness through nonarteritic anterior ischemic optic neuropathy, or NAION.

If the results are to be believed, then nearly 8.9% of all people prescribed semaglutide for diabetes develop NAION, compared to about 1.8% of those prescribed non-GLP1RA medication for diabetes. Similarly, they reported that 6.7% of those prescribed GLP-1RAs for weight loss developed NAION compared to 0.8% of those who took other anti-obesity medications. Take a look:

Should we trust this result? No, not at all.

For one, this results comes from data derived from a single-center, an ophthalmology clinic. The dataset is also quite small, maxing out at 17,289 patients with identified data. For two, the results are very surprising given that GLP-1RAs have anti-inflammatory effects.

But these facts didn't stop the media from widely reporting that Ozempic causes blindness.

The other issue? If these effect sizes and base rates are real, then we should have seen millions of people go blind already. Even just using a 5% rate, we should have seen more than two million people going blind from NAION by the beginning of this year. Before that, diabetics who used other GLP-1RAs should have also gone blind at incredible rates.

The plain fact is, this effect size is so large that it cannot have gone unnoticed. This brings me to the big issue:

The study was based on matching, but the data available to the clinic was meager compared to what they should have had. As a result, there was major residual confounding. This study population was also unrepresentative, as we can tell from the base rates of NAION presented for the control groups, or as we could infer from the fact that the data came from an ophthalmology clinic rather than, say, a general hospital.

The frequency of NAION in real, nationally-representative data on tens of millions of Americans is more like 0.16%, not the at-minimum implied amount of more than 1% of the population.

Luckily, we have better data. Firstly, we have a big meta-analysis covering databases containing a total of 37.1 million people. These were either administrative claims databases or electronic health records. Even better, we have a single study covering a national database of 66 million people.

Meta first:

The meta-analysis says that the risk of a NAION-specific diagnosis with semaglutide versus other medications for diabetes was 1.32 (1.14-1.54), far off the mark from the 4.28 in the media-alarming study.

Additionally, semaglutide was not apparently more dangerous than any other GLP-1RA drug when a strict definition of NAION was used (left side, sensitive). It was marginally-significantly worse compared to another anti-diabetic drug, empagliflozin (p = 0.02), using an expansive definition that includes some unconfirmed cases and cases likely to be something else (right side, specific):

So at this point, we should now be thinking two things:

1. The effect size from the popular study is insane on first principles, and it is insane relative to the rest of the literature, which we know to be more reliable given the settings, data quality, etc.

2. Semaglutide is not uniquely bad among GLP-1RA drugs.

Now the big study:

If you've read me for a while, you might know that I dislike meta-analyses. Except in rare occasions, it is better to have a single, very large study, because meta-analyses are based on individual tainted studies that pull it towards the bias of researchers, editors, and reviewers who participate in assembling the relevant literature. I'll link a larger piece on this below.

This big study beats the rest of the literature, and that's going to be super useful for us to make some inferences. This study's database beats the rest of the literature on size, diversity of the population, and on how many variables were available for propensity score matching, and then it throws in results from another 3.9 million person database compiled by Stanford, to show that their results replicate.

I love careful, large-sample research.

Prior to matching and just using all those who got on GLP-1RAs for weight loss versus other non-GLP1 weight loss medications, there was enhanced risk for NAION (HR = 2.95, p < 0.0001). But after matching, this finding was no longer significant, and the risk associated with the drugs was just gone (p = 0.49).

Furthermore, limiting the matched analyses to diabetics only (p = 0.13), to the obese only (p = 0.47), to GLP-1RA users taking semaglutide specifically (p = 0.44), or to patients with confirmed ophthalmological testing (p = 0.66), there was consistently just nothing. The results all went out the window when there was large, multi-center data, matched well.

Then using the Stanford data for a replication, the HR was a piddling 0.94 (p = 0.85).

Where did the risk go? Well, when matching is done well (and this can take a lot of variables), populations are large, and the mechanistic story is nonexistent, you can expect the risk shown in relatively small survival analyses to evaporate.

Or in other words, GLP-1RAs are likely not causing mass blindness, and if there's anything, it's a minor effect on a small base rate, that will take careful, large-sample research to suss out if it exists at all. For the population at large, the risk will clearly be easily counterbalanced by the benefits of these drugs.

Sources:

https://t.co/pEkO4UD87t

https://t.co/5DrJI3CUFY

https://t.co/eypTyLGQLz

https://t.co/Gg1Gl0f1mb

[Transhuman enthusiast @EWenxue]

@purpveluri @nikitabier No, I tried appealing about 10 times, I explained in detail why the bot might have wrongly flagged me (VPN use, traveling etc.).

But I only got automated rejection responses, no sign any human ever looked at it, or even that an automated Grok/LLM checked what I've written.

[Transhuman enthusiast @EWenxue]

"Inauthentic behavior" is ridiculous, my main account is suspended since a few weeks for no reason, any appeals are completely ignored!?
@nikitabier if you are a real human wrongly banned for "Inauthentic Behavior", how can you get your account back?

[Fenix Ammo CEO @FenixAmmoHR]

The Fenix Ammunition main account just got suspended for "inauthentic behavior"?

Is this a fucking joke?

[Transhuman enthusiast @EWenxue]

So many low IQ takes in response to Anatoly's very reasonable and measured post on the whole Malaysia fuckup regarding network school.

Proving his point for him.

Malaysia does have great food, and the view of Merdeka 118 is nice, else it's a very unattractive country.

[Anatoly Karlin 🧲💯 @akarlin]

Didn't comment on Network School's Malaysia issue it until now out of concerns it would come off as counter-signaling, but since it has been forced...

There are always trade-offs in life. ⚖️

You can negotiate economic zones and memorandums of understanding with highly functional liberal First World states. Switzerland would never pull a Honduras (previous government's decision was illegitimate, what's yours is now mine, suck it LOL) or Malaysia (where are you hiding the Zionist spies?) on you. But it's correspondingly expensive, and will not bend over backwards to accommodate your particular preferences.

Or you can negotiate with the less functional polities in the Second and Third Worlds. Leveraging their illiberalism, they can potentially give you much better deals, at least on the surface. But these deals are either less binding (weak rule of law), or are only tolerated in so far as they don't anger local mobs (various populists and Islamists) whom politicians even must appease, to some extent. This is actually more important in states with weaker institutions and less politically sophisticated citizenries.

There's no real solution to this except avoiding putting all one's eggs in one basket. But Balaji knows that.

On Malaysia specifically 🇲🇾

It has an English speaking workforce, a sizable Chinese and Indian "smart fraction" (main reason it's a functional $15k GDPpc country), solid infrastructure, and a commercial law code inherited from the British Empire. This makes it relatively stable and an attractive destination for capital investments that are not "politically loaded."

It's also a country where Muslims are subject to sharia, where 86% (as of 2013) approved of sharia, homosexual activity is criminalized, and elite science production - as proxied by the Nature Index - is on the level of Uganda (i.e. close to zero).

In particular, Muslim Malays have a strange obsession with the welfare of an ethnic group they have no genetic or linguistic ties to. States can legitimately have all kinds of views on the Israeli-Palestine conflict. Spain is notably anti-Israel, but expresses it in civilized ways, such as targeted sanctions and war crimes investigations. Malaysia has a blanket ban on Israeli citizens entering it, which even extends to dual citizens. This is collective punishment which is more characteristic of low IQ mob-like thinking.

I am not saying this makes Malaysia unusually bad. The US under the Trump regime has similar blanket bans on Afghans, Iranians, and several other nationalities whose main crime in the eyes of MAGA ideologues is to have bad/anti-US governments. Nonetheless, from a cynically practical perspective, there is a difference between kicking Afghans and Somalis (already some of the most disprivileged nationalities on the planet), and doing it to Israelis (citizens of a country that is economically First World and a major node in global tech).

This is an OK environment for attracting chip fabs and data centers and the like - it has been perfectly successful at that! - but it's not something that will ever foster a tech cluster with real social, cultural, and creative components. However, I allow that this isn't within the scope of the Malaysian government's ambitions; as a rule, emerging markets prioritize elite enrichment and social stability over grandiose visions of leapfrogging into the future, and Malaysia is no exception to this. But has implications for organizations (such as Network School) trying to brute force this regardless.

[Transhuman enthusiast @EWenxue]

@nikitabier @elonmusk you need to fix this, the current account banning is crazy kafkaesque

[🅱️ete oaks @b0newalljackson]

if anyone cares to read, these are the suggestions i sent @nikitabier regarding the report/suspend problem on @x (before he blocked me)

it would be very easy for @elonmusk to fix if he had any employees acting in good faith

[Transhuman enthusiast @EWenxue]

@mmccorddev @9mmsmg @FenixAmmoHR @nikitabier It's very kafkaesque.
On the appeal form you have to answer "why you think you didn't break X's rules", without them ever telling you what you supposedly did wrong...

[Transhuman enthusiast @EWenxue]

@robj3d3 @levelsio doesn't have to be, I've tested 2 times and came out in the 1900s and 1400s and I also don't have any 'symptoms' of high test except fast recovery, am competitive etc.

for me it turns out my SHBG levels are also really high, so less of it is biologically usable.

[Transhuman enthusiast @EWenxue]

@ArchFalls @TouchOfAWoman @T_Marie60 so how is the "person A gives in and says yes to sex" rape?
person A clearly changed opinion and is consenting, so it's not rape, right?
or are they willfully misleading B, in which case B has no way of knowing that A really doesn't want to.

[Transhuman enthusiast @EWenxue]

X suspensions are similar, my main account got suspended for 'artificial behavior' (I've only ever used it manually, normally), and they don't even tell you what you supposedly did wrong.
I'm already 3 appeals in now and they just answer with "no we think you're artificial".

[molson 🧠⚙️ @Molson_Hart]

They just suspended me.

[Transhuman enthusiast @EWenxue]

@ichiipsy Post Hoc分析で,そして部分的な弱い効果だから、無効である可能性は高いですね

[Transhuman enthusiast @EWenxue]

Also: People that can't like stuff on their main account during working hours, since it gives away that they are wasting time on X.

[Paul Graham @paulg]

Instant 10% increase in likes. Large numbers of people must have different opinions than they dare express, to move the total number of likes by that much.

[Transhuman enthusiast @EWenxue]

@kingkongfalls @gfodor I had the same impression, a lot more snark &amp; commie grouptalk

[SpaceX @SpaceX]

Super Heavy landing burn and soft splashdown in the Gulf of Mexico

[Transhuman enthusiast @EWenxue]

@brianluidog These are great problems, makes you try out a bunch of methods and combinations. And it's easy to teach the method of step by step reasoning with them.

[eugyppius @eugyppius1]

Is there anything more tiresome than morality in politics?

We have had enough of this hypermoralism, on the left and the notional ‚right‘, to last a millennium.

I despise above all those who ask me to care about weakness, whatever its form. I want an amoral politics and an amoral virtue that is finally deaf to disadvantage, death and suffering. If you feel that you have suffered an injustice politically, you must seize what is yours by force or shut up forever.

I am sick to death of the pleading and the whining of the unfortunate, it is a plague on the land.

[Transhuman enthusiast @EWenxue]

@SixBangs @IterIntellectus straight ahead/even is an illusion, in reality it's a tiny bit curved downwards, you just don't realize it.

the only way to walk in a straight line would be to perfectly carve out the part of the circle that you have to downwardly walk through.