#272: Taproot, a Highly Anticipated Bitcoin Software Upgrade, Is Nearing Activation, & More

The Bitcoin network is one month into its signaling period for Taproot, the most highly anticipated and well-supported Bitcoin upgrade since SegWit in 2017. In January 2018, Bitcoin Core developer Greg Maxwell proposed Taproot to increase the privacy and efficiency of Bitcoin transactions. Along with Schnorr signatures, another upgrade that will improve Bitcoin’s digital signature scheme, Taproot should make Bitcoin transactions indistinguishable whether they are on a Lightning Network channel, peer-to-peer, or in complex smart contracts.

To be activated, miners responsible for 90% of the blocks mined during this two-week difficulty epoch must signal support for the Taproot upgrade. This epoch is the third of six possible signaling periods. As of this writing, more than 90% of hash rate has mined one or more blocks signaling support for Taproot but only 77% of blocks have signaled support during this epoch. With only 129 blocks remaining in this epoch, the Taproot activation might have to wait for one of the next three epochs. If it were to fail by the end of the sixth epoch, developers will have to pursue an alternative activation strategy like a user-activated soft fork.

You can follow along Taproot’s path to activation here.

This week JD subsidiary, JD Logistics (JDL), began trading in Hong Kong. According to its founder Richard Liu, JDL provides the mission-critical, technology-enabled logistics infrastructure that supports JD’s demanding e-commerce fulfillment requirements. According to its IPO prospectus, JDL processed and delivered approximately 90% of JD orders within 24 hours.

In 2017, JDL began to offer its integrated logistics services to non-JD customers which now account for almost 50% of its revenues. According to ARK’s research, JDL has built a state-of-the-art integrated logistics company thanks to disciplined investment in physical infrastructure, proprietary automation and predictive technology, as well as 200,000 full-time employees. According to its IPO filing, JDL’s complaint rate is less than 10% of the industry average: 0.002 per million parcels delivered compared to 0.22.

Meanwhile, in other Chinese innovation-related news this week, Meituan reported first quarter results and highlighted its ambition to quadruple daily deliveries to 100 million orders by 2025 with a focus on autonomous technology, particularly drones.

Enabling its success thus far, Meituan has lowered its cost per delivery to $0.70 on an average order value of $7.50. This compares to a leading US food delivery operator with close to $3 cost per delivery on $30 average order value. As a result, food delivery accounts for 20% of total restaurant and dining industry in China, well below 9% in the US.

Typically, primary care is the first point of contact in most healthcare systems, making it vital to the future of genetic medicine. Genetic testing is likely to be relevant to all disease areas and stages of life. The most mature testing categories are hereditary disease screening, non-invasive prenatal testing (NIPT), and carrier screening, all well-suited for primary care. That said, outside of small concierge practices, few individuals have had the opportunity to discuss genetic testing with their primary care providers (PCPs). Despite or because of the obstacles, the coronavirus pandemic seems to have paved a path for integrating genetics into primary care.

Most insurers do not cover proactive genetic testing. Even with coverage, many patients and physicians may worry about denied insurance claims, which could set a patient back by thousands of dollars for a test. Passed in 2008, the Genetic Information Non-Discrimination Act (GINA) ensured that health insurers cannot deny eligibility or coverage based on the results of a genetic test. Unfortunately, these protections do not apply to life or disability insurance. Most PCPs operate under a fee-for-service system, which incentivizes quick patient visits. Genetic testing and interpretation are not suited to this model. A recent review highlighted several operational challenges: (a) difficulties gathering family history, (b) pre- and post-test patient education, (c) difficulties interpreting test results, (d) backend integration with electronic health records (EHRs) and clinical decision support tools (CDSTs), and (e) equitable access to genetic counselors (GCs).

Propelled by the COVID-19 pandemic, we believe innovative virtual healthcare and genetic testing companies are poised to address these obstacles. Perhaps more importantly, these groups are highly incentivized to work together.

Groups like Teladoc (TDOC), One Medical (ONEM), and Accolade (ACCD) have invested aggressively in virtual primary care services. One Medical’s ‘click-and-mortar’ platform spans both digital and in-person primary care and incorporates its own EHR system. It also compensates physicians with salaries instead of fees for service and removes the burden of administrative tasks that devour so much of a physician’s day. In our view, this model could be fertile ground for value-based genetic testing services.

Meanwhile, testing providers like Natera (NTRA), Invitae (NVTA), and Myriad Genetics (MYGN) have launched AI-enabled chatbots to facilitate genetic testing. With acquisitions of Clear Genetics and YouScript, for example, Invitae’s platform offers consent, pre- and post-test education, automated family history collection, results reporting, virtual genetic counseling, integration with EHRs like Epic and Cerner, CDST capabilities, and low-cost patient-pay options.

In our view, these two groups are likely to partner and generate high-quality, longitudinal data sets to improve patient care, train machine learning algorithms, and bolster their value propositions to customers.

Historically, researchers have evaluated AI models primarily on their accuracy: the more accurate the model, the higher it ranked against its peers. Compared to an academic or research setting focused on building the most advanced AI systems, however, accuracy is not the only measure that matters in the real world. A natural language model that responds to queries from a research dataset accurately but does not comprehend spelling mistakes will not perform well as a chatbot. Similarly, a model biased wittingly or unwittingly against a particular race or political opinion should not make credit decisions. 
 
Last week, a new AI startup called Anthropic announced a $124 million Series A funding round. Led by the former Vice President of Research at OpenAI, Anthropic is focused on metrics beyond accuracy as it builds large-scale AI systems with better interpretability and robustness. Facebook AI also announced the launch of Dynaboard by Facebook AI, a tool to evaluate AI models on factors including accuracy, compute, memory robustness, and fairness to help researchers select the correct algorithms for their applications. A product using AI at the edge, for example, might weigh computational efficiency over accuracy in its evaluation process.