📉 Everyone in Crypto is Freaking Out

Your favorite band is an algorithm. Is Google evil? Facebook admits content moderation is hard.

Tether vs. New York State

Last Thursday, NY State Attorney General Letitia James announced a court order against the Bitfinex cryptocurrency exchange and their digital asset “Tether”. For those who have been following the crypto scene, a common reaction is: “Finally? I knew that was going to blow up eventually.” Well, it blew up, but there are some interesting complications.

To summarize, here is the plotline involving Bitfinex to date:

  • Bitfinex is “owned and operated by the same small group of individuals” that also created a stablecoin called Tether. A stable coin is a digital asset that is pegged 1:1 against the US dollar.

  • This group claimed to have an equal representation of dollars in a separate bank account, and anyone could redeem one Tether for a dollar at any time.

  • People have been very skeptical about the legitimacy of this claim, and in 2018 the Tether/Bitfinex team attempted to produce an official audit, which was never fully concluded.

  • Due to regulatory issues, the Bitfinex team couldn’t find a legitimate bank to hold their money, so they resorted to using a less reputable financial institution based in Panama called “Crypto Capital Corp” to process transactions from their platform.

  • The Bitfinex team is now being accused of using funds from the Tether account to float money to their customers, due to issues with their offshore bank account of unknown origins.

If the idea of storing $900M in an unproven offshore bank account with no regulatory oversight makes you feel funny, then you will love what comes next.

Late last year, Crypto Capital stopped responding to Bitfinex requests for payment processing. Understandably, the Bitfinex team hit the big red button and began to funnel money out of the Tether account to facilitate transfers within their platform. The court order asserts that the Bitfinex team has misappropriated $850 million dollars of co-mingled client and corporate funds. It goes on to say that this undisclosed activity has “treat[ed] Tether’s cash reserves as Bitfinex’s corporate slush fund, and are being used to hide Bitfinex’s massive, undisclosed losses and inability to handle customer withdrawals.”

The ironic part is, contrary to common belief, Tether actually had the money in it’s bank accounts all along. The problems arose due to insolvency with their Panamanian payments processor and the need to fulfill activity on the platform.

OpenAI Makes Music

OpenAI released a version of their popular GPT-2 algorithm that is able to produce generated music. They hosted a livestream on Twitch that received tens of thousands of views and combined musical influences from Taylor Swift to Mozart. This is a continuation of the wonderful work started by Christine Payne during her time as an OpenAI scholar.

This approach is an interesting result in proving that large transformer models like GPT-2 are flexible enough to be used for a range of creative tasks, including writing poetry and composing music. Earlier this year, the GPT-2 model received widespread publicity due to it’s ability to generate humanlike essays that were pure fiction. The OpenAI team chose not to release this model, citing concerns over dual-use technology like AI and the ability for bad actors to use the full model to generate fake news.

Since this time, OpenAI has shifted from a non-profit to a Limited LP. It remains to be seen whether they will continue to open source their work.

If you are interested in using MuseNet, you can try it out until May 12th.


📚 Reading

An article on the Facebook team that dealt with the aftermath of the Christchurch shooting. In the first 24 hours of airing, millions of copies of the video were caught by the automated detection system. The system used a hash from an audio snippet to detect copycat uploads. Disturbingly, people were distorting the video dimensions and applying edits to get it through the content filters. Also, Facebook now employs over 7,500 people to watch some of the worst videos on the internet in order to moderate content uploaded to the site. Link.

The head of Tesla’s Autopilot team uploaded a blog post on his methods for training neural nets. The takeaway: keep trying different things and stare at the loss function. Also — spend a lot of time in the data. Link.

How to hide from an image recognition algorithm by wearing a painting around your neck. Link.

A great introduction to Docker. Link.

A popular article on how information is reshaping education, politics, and commerce. Link.


Not a subscriber yet? Sign up here.