Are Decentralized Networks the Antidote to Social Media Hacking?

By Yassine Elmandjra | @yassineARK

This week, hackers targeted several verified Twitter accounts - among them Elon Musk, Joe Biden, and Jeff Bezos - in a "coordinated social engineering attack." They used the compromised accounts in a bitcoin giveaway scam, trying to trick people into sending funds to an address in the hope of receiving more.

The attack appears to have been initiated directly from Twitter's admin panel. Specifically, Twitter's admin tool can disable two-factor authentication and update the email address associated with any Twitter account without notifying the user.

With more than 300 million monthly active users (MAUs), Twitter has been criticized as a massive data honeypot. Because its data is stored on centralized servers, hackers can exploit Twitter accounts fairly easily.

The crypto community has been quick to point out the risks of relying on trusted third parties and are offering perhaps a better solution. Blockchain-based systems, or decentralized information networks, could give users the ability to own their online identities. With a self-sovereign identity system, each user would control his or her own data with a cryptographic private key, turning platforms like Twitter into "dumb-terminal" interfaces to engage with other users.

Twitter recognizes the challenges associated with centralized solutions and is exploring decentralized protocols as solutions. In December 2019, Jack Dorsey announced Project Bluesky, a Twitter-led initiative to fund an independent team of open source architects, engineers, and designers and develop a decentralized standard for social media. If successful, Twitter will become a client of the new standard.

Perhaps these hacks will catalyze Twitter to transition to a decentralized alternative sooner rather than later.

The OCC Is Rolling Out National Payments Charter 1.0 This Fall

By George Whitridge | @GeorgeOfARK

According to Brian Brooks, The Office of the Comptroller of the Currency (OCC) is rolling out a ‘payments charter 1.0’ this fall. Today payment companies such as Square and PayPal must apply for Money Transmission Licenses in each state to move funds to and from sellers and consumers. Companies often spend millions of dollars applying for money transmission licenses, with approval in states like California, Texas, and New York taking up to 12-24 months. A potential game-changer, the OCC’s payments charter will supersede state licenses and offer nonbank payment providers a national platform.

A national payments charter could encourage more innovation and competition in the payments space, especially as the OCC adds more features and functionality. According to Brooks, version 2.0 of the payments charter could include direct access to the Federal Reserve’s real-time payment network and settlement service FedNow. The OCC plans to roll out version 2.0 in 2023 or 2024.

In our view, the OCC’s proposed charters will be positive for the fintech industry, facilitating innovative business models, saving on card processing fees and, ultimately, enabling companies to transform cost centers into revenue drivers.

OpenAI’s GPT Model Needed Only Two Examples to Learn JavaScript

By James Wang | @jwangARK

Barely a month since OpenAI released it to developers, GPT text generator already is performing remarkable tasks. With only two examples in English, Sharif Shameem, founder of Debuild, taught GPT to generate JavaScript (JSX) code. When asked to generate a button that looked like a watermelon, GPT promptly produced code that rendered a circular button with green borders and pink interior. Asked to make a table of the richest countries in the world, it produced a list of developed countries ranked by GDP. Unlike Siri or Alexa which retrieve information from the internet, GPT internalizes knowledge after training on massive datasets and then synthesizes answers in well-written English. Today, Twitter is teeming with examples of GPT feats, from imitating famous philosophers to performing basic accounting tasks.

GPT appears to be the first example of artificial intelligence (AI) that a non-technical person can “train”. While today’s AI tools can perform specific tasks like recognizing images, translating languages, and answering questions, users can’t instruct them to do other tasks. Based on Shameem’s experiment, GPT can be “told” what to do with text examples. With just two examples, it seems to “understand” and execute the task. Though not “general” artificial intelligence by any stretch, GPT could evolve into an “idiot savant”: after ingesting all the information on the internet, GPT should be able to deliver amazing - better than human -results.

In the short term, we believe OpenAI is shaping up to become the Twilio of text generation. Longer term: AI will likely intermediate, disintermediate, and enhance all jobs involving language.

Illumina’s Acquisitions Appear to be Unclogging the “Bioinformatics Bottleneck”

By Simon Barnett | @sbarnettARK

Sequencing applications must be cost-efficient across three workflows: upfront sample preparation, sequencing, and data analysis. While it has reduced sequencing costs dramatically, Illumina (ILMN) has struggled to address the concomitant “bioinformatics bottleneck”. Because of the high compute and storage costs associated with the data explosion, the aggregate cost to sequence and analyze a whole human genome has plateaued recently.

In response, over the past 18 months Illumina has built or bought technologies to reduce the friction associated with genomic data analysis. ARK believes the benefits will lower aggregate sequencing costs immediately for non-enterprise customers such as core labs, R&D groups, and smaller hospitals.

Illumina acquired Edico Genome in 2018 to commercialize DRAGEN—an FPGA-based technology that accelerates genomic data processing and quality control. The NextSeq 1000 and 2000, Illumina’s newest sequencers, come equipped with DRAGEN FPGAs, so customers can purchase both sequencers and on-premises compute infrastructure at the same time. Recently, Illumina acquired BlueBee, a well-regarded cloud-platform with tools for sequencing process analysis and data sharing, which we believe should benefit users of BaseSpace—Illumina’s cloud environment for genomics researchers. This week, Illumina acquired Enancio’s genomic data compression technology, which should lower both the bandwidth necessary for sharing data and the cost of storage, particularly for researchers using cloud or cloud-hybrid compute architectures.

In our view, these investments will cement Illumina’s position as the sequencing platform of choice for researchers and clinicians. A more robust back-end solution will add to the friction that its customers would face if they were to switch to other short-read sequencing platforms.  Moreover, by unclogging the bioinformatics bottleneck, Illumina is enabling capital-constrained hospitals to sequence in-house, hopefully increasing their adherence to precision oncology guidelines.

Tessera Therapeutics Trademarked the Term “Gene Writing”, But They Don’t Own the Space

By Ali Urman | @aurmanARK

Gene editing has been around for a long time but CRISPR technology is the primary reason that excitement around it has exploded in the past five years. Recently both Blue Bird and CRISPR Therapeutics provided evidence of potential cures for sickle cell and beta thalassemia with preliminary data.

Now Tessera Therapeutics, an early-stage life sciences company pioneering gene writing, is claiming to write DNA without breaking it, something impossible to date without off target effects. Tessera uses mobile genetic elements (MGEs) to move or copy DNA into new locations, another technology that is not new. Barbara McClintock won the Nobel Prize in 1983 for her discovery of the first MGE that moved DNA in corn.

Currently the gold standard, CRISPR seems to be the most researched and reproduced gene editing technology. In terms of reach and/or influence, Google Trends suggests that searches for CRISPR are much higher than those for MGEs. Measured by number of publications from 2002-2020, and 1949-2020, respectively, while researchers had published only 19,278 articles on CRISPR, less than half the 41,128 on MGE, year-to-date in 2020 alone, 3,446 CRISPR papers have dwarfed the 422 on MGE.

In our view, CRISPR technology is in the early days of potentially curing disease, especially monogenic diseases. We believe other technologies will have to be supported by strong patents and data before they can compete.