We offer our view of Anthropic because it is part of the total ARK Venture Fund portfolio. To see the most updated portfolio, please click here.
Our research suggests that enterprises will adopt two types of AI models: (1) domain-specific models and (2) foundation models. Domain-specific AI designs algorithms and models to resolve specific challenges for a particular industry, function, and use case. Foundation AI models are machine learning models that are trained on large amounts of unlabeled data. They are built on conventional deep learning and transfer learning algorithms.
Directionally, domain-specific models should be smaller and more specialized, trained on platforms like MosaicML, and embedded in vertical software products. Domain models generally require less training data; they can be trained on curated data sets, are easier to align, and provide better inference performance.
By contrast, foundation models are trained on large data sets, capable of a wider range of tasks, and generally are better for open-ended conversational use cases like customer support. Foundation models also are more difficult to align than domain-specific models. We believe generic foundation models such as GPT-4 are at risk of commoditization. On the other hand, we see a compelling opportunity for industry-specific foundation models (GPT for healthcare, for example) and custom enterprise foundation models, whose size and task range fall somewhere between generic foundation models and domain-specific models.
Anthropic leverages a proprietary “Constitutional AI” research process built to align foundation models, prevent hallucination, and curb bias. Their technique combines supervised learning and reinforcement learning to teach “values” to a model. Anthropic’s first product, Claude, is a generic foundation model competitive with GPT-3.5. The company soon will release Claude 2.0, which is competitive with GPT-4. We believe the Claude family of models is an excellent testbed for Constitutional AI, given the wide range of training data and open-endedness of public usage. From a commercial standpoint, Anthropic leverages Constitutional AI to train both industry-specific foundation models and custom enterprise foundation models.
Anthropic recently released Claude 1.0 and should release Claude 2.0 soon. Anthropic’s Founder and CEO, Dario Amodei, previously served as Vice President of Research at OpenAI, where he co-invented reinforcement learning with human feedback (RLHF) and led the GPT- 2 and GPT-3 projects. Anthropic’s broader leadership team also includes Daniela Amodei, OpenAI’s former Vice President of Safety and Policy, as well as other researchers from OpenAI and Google Brain.
Disclosures
We offer our view of Antrhopic because it is part of the total ARK Venture Fund portfolio. To see the most updated portfolio, please click here. Holdings subject to change. Not a recommendation to buy, sell, or hold any specific security.
ARK Venture Fund is a closed-end interval fund. The fund is considered illiquid and there is no secondary market for investors to sell their shares.
BEFORE INVESTING YOU SHOULD CAREFULLY CONSIDER THE FUND’S INVESTMENT OBJECTIVES, RISKS, CHARGES AND EXPENSES. THIS AND OTHER INFORMATION IS IN THE PROSPECTUS, A COPY OF WHICH MAY BE OBTAINED HERE. PLEASE READ THE PROSPECTUS CAREFULLY BEFORE YOU INVEST.
There is no assurance that the Fund will meet its investment objective. The value of your investment in the Fund, as well as the amount of return you receive on your investment in the Fund, may fluctuate significantly. You may lose part or all of your investment in the Fund or your investment may not perform as well as other similar investments. Therefore, you should consider carefully the risks at the bottom of this page.
ARK Investment Management LLC is the investment adviser to the ARK Venture Fund.
ARK’s statements are not an endorsement of any company or a recommendation to buy, sell or hold any security. ARK and its clients as well as its related persons may (but do not necessarily) have financial interests in securities or issuers that are discussed. Certain of the statements contained may be statements of future expectations and other forward-looking statements that are based on ARK’s current views and assumptions and involve known and unknown risks and uncertainties that could cause actual results, performance, or events to differ materially from those expressed or implied in such statements.
