#366: Does An Influencer’s AI Counterpart Signal The Return Of The 1-900 Number?, & More

Snapchat influencer Caryn Marjorie, 23 years old, recently ventured into the AI space with a virtual version of herself.¹ Powered by OpenAI and designed to simulate a virtual girlfriend, CarynAI charges $1 per minute for interactions and, in just one week of beta testing, generated $71,610 from its predominantly male user base. According to Marjorie, CarynAI could reap $5 million per month if 20,000 of its 1.8 million Snapchat followers become paying subscribers.

Does CarynAI signal a return of the popular 1980s 1-900 numbers² that consumers called to access information and entertainment content? In just three years from 1988 to 1991, the market value of companies hosting 1-900 numbers scaled more than 15-fold from $60 million to ~$1 billion as providers monetized content. Like Marjorie’s AI pricing structure, premium phone numbers charged ~$2 for the first minute and $1 per subsequent minute.

The increasing realism of AI companions is sparking debates about their ethical implications as creators rush to implement them. Could users opt for artificial instead of real relationships, transforming human-human interactions? Given appropriate guardrails that ensure the safety associated with these companions, AI could unlock exciting new revenue streams for content creators around the world.

Recently, MosaicML’s release of MPT-7B³ and a boost in Anthropic’s Claude context window to 100,000 tokens⁴ have highlighted the power of expanding context windows, the efficiency of MosaicML’s training platform, and the potential for open-source AI models.

MosaicML’s MPT-7B open-source model stands out for its extensive training on a vast amount of text and code, addressing the limitations of prior open-source models and offering an alternative to the commercial foundation models from OpenAI, Anthropic, and Cohere. Notably, the new model tackles issues related to context length restrictions by accommodating inputs up to 84K tokens, matches the quality of Meta’s LLaMA-7B, and highlights the efficiency of training large models on MosaicML’s platform. Compared to the ~$5 million cost of GPT-3’s final training run in 2020,⁵ Mosaic trained MPT-7B at a cost of only $200k, with no human intervention.⁶

Expanding its context window from 9,000 tokens to an impressive 100,000, or ~75,000 words, Anthropic’s Claude now can analyze and interpret extensive documents more effectively and efficiently with synthesized answers to complex questions. Increased context windows also will enable users to analyze long-form documents like financial statements, research papers, grant proposals, and legal documents.

By reducing failure rates and hastening commercialization, the convergence of next generation sequencing (NGS), artificial intelligence (AI), and CRISPR has the potential to accelerate drug discovery. Currently, drug development costs, including the cost of failures, average ~$2 billion over the ~10 years to commercialization. Pre-clinical costs typically account for ~40% of total clinical development costs.⁷ We believe CRISPR could lower overall drug development costs by accelerating pre-clinical drug development processes in two ways: (1) disease biology and target identification, and (2) discovery and development.

Disease Biology and Target Identification
Scientists analyze gene sequences, or gene expression, in healthy and diseased cells to identify or validate therapeutic targets and understand the mechanisms of drug resistance. A popular screen used in both disease biology and target identification is a functional genomic screen, typically conducted with the aid of RNA interference (RNAi) or small molecule inhibitors.⁸

CRISPR offers a more efficient approach. While RNAi reduces gene expression temporarily, CRISPR disrupts genes permanently, providing a clearer phenotype that elucidates a gene’s functional role. Moreover, CRISPR’s ability to generate multiple knockouts simultaneously enables the exploration of complex genetic interactions, enhancing the understanding of disease pathways. Scientists in the UK and Cambridge, for example, have found⁹ that a CRISPR dropout screen identified genetic vulnerabilities and therapeutic targets in acute myeloid leukemia.

Discovery and Development
CRISPR also can enhance the efficiency and accuracy of drug screening platforms by enabling the high-throughput drug screening so critical to drug discovery. Using CRISPR to create cell lines with specific genetic alterations, researchers can study the impact of potential drugs on various cellular models relevant to disease, allowing for more precise evaluation of drug candidates and reducing significantly the number of false positives in the drug screening process. Scientists from the College of Life Sciences in China, for example, identified¹⁰ potential treatments for acute myeloid leukemia using a CRISPR-Cas9 screen.

As research advances, we expect to see many CRISPR-based therapies developed with pre-clinical CRISPR screening techniques enter the clinic to serve patients more efficiently and effectively.

[1] Sternlicht, A. 2023. “A 23-year-old Snapchat influencer used OpenAI’s technology to create an A.I. version of herself that will be your girlfriend for $1 per minute.” Fortune.
[2] Raviv, S. 2016. “The Rise and Fall of the 1-900 Number.” Priceonomics.
[3] The Mosaic NLP Team. 2023. “Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs.” MosaicML.
[4] Anthrop\c. 2023. “Introducing 100K Context Windows.”
[5] Li, C. 2020. “OpenAI’s GPT-3 Language Model: A Technical Overview.” Lambda.
[6] The Mosaic NLP Team. 2023. “Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs.” MosaicML.
[7] ARK Investment Management LLC. 2023. Big Ideas.
[8] In this process, scientists “knock down” a gene to see the resulting phenotypic change and identify which genes are essential for disease progression and cellular processes.
[9] Tzelepis, K. et al. 2016. “A CRISPR Dropout Screen Identifies Genetic Vulnerabilities and Therapeutic Targets in Acute Myeloid Leukemia.” Cell Rep. doi: 10.1016/j.celrep.2016.09.079.
[10] Su, G. 2020. “CTCF-binding element regulates ESC differentiation via orchestrating long-range chromatin interaction between enhancers and HoxA.” JBC Research Article.