Humanoid Robots Could Transform Small Manufacturing Companies, & More

Want more research from the ARK Team? Have feedback on our publications? Click here to help inform our content creation.

1. Humanoid Robots Could Transform Small Manufacturing Companies

By: @skorusARK

Why are humans designing humanoid, or generalizable, robots? Versatility is the answer.

In contrast, a wrench excels at one task, tightening nuts and bolts, and not much else. It is not versatile or capable of performing a diverse number of tasks.

Based on data from the US Census Bureau, ARK’s research suggests that most people involved in US manufacturing are employed by small firms, as shown below. Typically, they perform a wide array of tasks, unlike those in large firms who engage in relatively repetitive tasks that have been giving way to automation. Humanoid robots with generalizable capabilities could level the playing field for small companies, boosting productivity and transforming the economy at scale.

2. Amazon’s New Generative AI Suite Is Poised To Compete With Microsoft’s Copilots

By: @downingARK

Last week, Amazon rolled out its generative AI product suite, Amazon Q, to the general public, aiming to increase the productivity of business users and developers.

One product in the suite, Amazon Q Developer, is adding to the number of advanced coding agents—exemplified by Devin and SWE-agent—that use generative AI models to develop software from end-to-end. Using Q Developer internally, a team of five engineers at Amazon modernized 1,000 Java applications in just two days instead of months. Relative to benchmarks, Amazon Q Developer scored 13.4% on SWE-Bench, surpassing the performance of SWE-agent to become the best-performing coding agent to date on that benchmark.[1]?Informed by Wright’s Law,[2] ARK’s research suggests that model capability will continue to increase as costs decline, empowering developers to focus on building high-value products and services while eliminating companies’ technical debt.

Another product, Amazon Q Business, is a chat interface like Microsoft Copilot and ChatGPT Enterprise. The tool aims to increase knowledge-worker productivity across a variety of tasks by answering questions and completing tasks using enterprise data. Users also can create custom user interfaces on demand—“Q Apps”—that are oriented to specific tasks, and then share them throughout the organization. Q Apps could serve as an important milestone in redefining how software is created and used within organizations, potentially lowering costs, increasing the pace of innovation, and displacing incumbent vendors.

3. CRISPR-GPT Is Enabling Scientists To Design Experiments More Efficiently

By: @RongGuoARK, PhD

While CRISPR technology has been used widely in biomedical research to create curative therapeutics, realizing its potential will require deep technical knowledge and lab experience. To increase efficiency in planning experiments, researchers from Stanford and Princeton recently developed

CRISPR-GPT,[3]?a large language model (LLM) agent focused on the automated design of gene-editing experiments.

Unlike general-purpose LLMs that lack the domain knowledge required to design biological experiments, CRISPR-GPT is trained on a large corpus of gene-editing literature to gain domain expertise and reasoning abilities. Instead of spending time searching through literature to design experiments for gene-editing, researchers now can give CRISPR-GPT general experiment goals, such as “I need to knock out EGFR (a gene that causes cancer),” and receive detailed experiment plans, including the best guide RNA choice, the best delivery approach, and recommendations for the best validation experiments.

Drawing on CRISPR-GPT’s ability to convert general experiment goals into detailed experiment plans, more researchers should be able to use gene-editing tools more efficiently, saving time and money while increasing the probability of scientific breakthroughs.

Want more research from the ARK Team? Have feedback on our publications? Click here to help inform our content creation.

[1] Sivasubramanian, S. 2024. “Accelerate software development and leverage your business data with generative AI assistance from Amazon Q.” AWS Machine Learning Blog.

[2] Wright’s Law states that for every cumulative doubling of units produced, costs will fall by a constant percentage. See Winton, B. 2019. “Moore’s Law Isn’t Dead: It’s Wrong—Long Live Wright’s Law.” ARK Investment Management LLC.

[3] Huang, K. et al. 2024. “CRISPR-GPT: An LLM Agent for Automated Design of Gene-Editing Experiments.” arXiv.