1/ More thoughts on impasse over organizing Senate/committees. McConnell demands Democrats commit that they won't "go nuclear" to ban legislative filibusters. Schumer says no. The organizing resolution can be filibustered, hence McC's potential leverage. So what's going on here?

Trump on climate: I don't believe it. Trump on Khashoggi: I don't believe it. Trump on Russia: I don't believe it. Idiocy? A canny game? Doesn't really matter as much as the underlying intellectual autocracy. In this administration only one opinion matters.

It's an assault on facts and reason. It's an assault on good governance. It's an assault on the bureaucracies that are being discounted. It's an assault on science and history. But it is also an assault on one of the underpinnings of democracy.

We live in a system that is grounded on the idea of collaborative government. Leaders may have a final word but even then, there are typically checks and balances. Hopefully, in this case, such checks and balances may work. But they also are being tested.

The president's position that he is smarter than all the world's scientists, than the entire intelligence community, than experts who have studied any issue all their lives, is, of course, asinine. But his belief that government turns solely on his opinion, is monarchic.

Louis XIV reportedly said, "L'etat est moi." That he was the state. Trump asserting that he knows better than all, discounts all advice, that his brain is enough to chart the course for America alone, is essentially saying the same thing.

Important paper from Google on large batch optimization. They do impressively careful experiments measuring # iterations needed to achieve target validation error at various batch sizes. The main "surprise" is the lack of surprises. [thread]

https://t.co/7QIx5CFdfJ

The paper is a good example of lots of elements of good experimental design. They validate their metric by showing lots of variants give consistent results. They tune hyperparamters separately for each condition, check that optimum isn't at the endpoints, and measure sensitivity.

They have separate experiments where the hold fixed # iterations and # epochs, which (as they explain) measure very different things. They avoid confounds, such as batch norm's artificial dependence between batch size and regularization strength.

When the experiments are done carefully enough, the results are remarkably consistent between different datasets and architectures. Qualitatively, MNIST behaves just like ImageNet.

Importantly, they don't find any evidence for a "sharp/flat optima" effect whereby better optimization leads to worse final results. They have a good discussion of experimental artifacts/confounds in past papers where such effects were reported.

1. #Qanon told us that Saudi Arabia is a central piece of the puzzle in taking out the Deep State...we are watching the Deep State attempt to remove Crown Prince MBS and disrupt Pres Trump's administration's ties with MBS and the "new" Saudi/US coalition. Early in his admin,

2....@Potus sent Jared Kushner (JK) to SA on an undisclosed trip and told King Salman that business as usual was over, and if he wanted Pres Trump's support, Prince Alwaleed would have to be divested of $ and power; Alwaleed is one of top Deep State $ men and was tied to HRC.

3. @Potus knew that before he could begin cleansing the US and the rest of the world of the DS, he would have to cut one of the financial head's off the snake (Alwaleed). After JK met w/ King Salman, Alwaleed was arrested and divested of billions; MBS became leader of SA.

4. SA loves Pres Trump because he's a man of his word; SA could not trust Hussein or Clintons because they use SA for oil and leverage; SA has many factions, but essentially you have the new and upcoming SA that is attempting to modernize through Crown Prince MBS, and then you

5...have the old guard represented by Alwaleed who was one of the Deep State financial heads and loyal to the DS more than SA; SA knew Hussein was financing Iran to hold western world hostage, and SA was caught in middle of Deep State black mail; this is why @Potus was celebrated

The best way to know if people like your product is NOT:

- Have a lot of emails in the waiting list
- Become #1 on PH
- Become #1 of HN or Reddit
- Have people that saing "I'll pay"
- Have a lot of free users

The best way to know if they're actually pay for it.

The only thing that matters is your product providing a value, and you can't know this until people will pay money for it.

We can have a first signal of a product/market fit after the launch by check how many sales we've got in the first 24h. Some scale:

200+ This has a high potential.
100+: This has a potential.
50+: Some people need it.
10+: People almost don't need it.
0+: People don't need it.

This is for a single time payment product. For a subscription probably it should be less because people hate subscription.

And obviously, you should have a lot of traffic like 3-5k+.

Here is a stat for my products (macOS apps) for the first 24h after the launch:

Cherthala Karthyayani Kshetram
#Alappuzha, Kerala
#KeralaTemples 🚩

Cherthala Karthyayani Temple Prathishta was done by Vilwamangalam Swamy. Chertala Pooram is very famous. Devi is Mangalya Dayini here removes obstacles relating to marriages. Sthala Puranam says that, once 👇

Vilwamangalam Swamiyar was passing this area after Prathista of Padmananbhaswamy Temple, he felt Devi chaitanyam and recognized as Karthyayani Devi. But Devi seeing Vilwamangalam went inside the pond. Swamy searched for Devi continously for 6 days and on 7th day Vilwamangalam

got hold of Devi's hair. Thus only Devi's head is Prathishta here. Devotees do Kozhi parathal ( offering cock) here as an offering to Karthyayani Devi.

Karthyayaniamme Sharanam
🙏🕉🙏