(Repository Location: https://github.com/jjmccarty/ai-case-study.git)
- The following AI case study outlines Anthropic
- Anthropic was founded and incorporated in 2021 by several former OpenAI executives and employees including [^1]
- Dario Amodei - Co-founder and CEO
- Daniela Amodei - Co-Founder and President
- Jason Clinton - CISO
- Jared Kaplan - Co-Founder and Chief Science Officer
- Ben Mann - Co-Founder, Technical Staff
- Jack Clark - Co-Founder, Head of Policy
- Mike Krieger - Chief Product Office
- Jan Leike - former OpenAI alignment researcher
- Anthropic's current Board of Directors includes:
- Dario Amodei
- Daniela Amodei
- Luke Muehlhauser
- Yasmin Razavi
- Anthropic was founded as a result of directional differences between Anthropics founders (Dario and Daniel Amodei) and OpenAI leadership.
Anthropic provides a very clear mission statement (from the company website) [^2]
Anthropic exists for our mission: to ensure transformative AI helps people and society flourish. Progress this decade may be rapid, and we expect increasingly capable systems to pose novel challenges. We pursue our mission by building frontier systems, studying their behaviors, working to responsibly deploy them, and regularly sharing our safety insights. We collaborate with other projects and stakeholders seeking a similar outcome.
In recent interviews with AI Businesss regarding the startup Dario Amodei stated [^3]
“Anthropic’s goal is to make the fundamental research advances that will let us build more capable, general, and reliable AI systems, then deploy these systems in a way that benefits people,”
Anthropic funding comes from a variety of sources including investment firms and large organizations such as AWS and Google. Over the last year the company has secured over $7.3 billion in funding. With the commitments of companies such as AWS and Google the funding does come with stipulations as part of the partnership including the commitment by Anthropic to utilize assets such as cloud computing resources and chips by the funding entities.
The following companies and firms have publicly invested in Anthropic:
- Amazon Web Services completed a $4 billion investment in Anthropic as of March 2024. The partnership involves Anthropics's use of Amazon's chips to build it's AI software and Amazon has added Anthropic to its cloud services. [^4]
- Google has invested $2+ billion over multiple years in Anthropic [^5]
- Menlo Ventures completed a $750 million funding for Anthropic in 2023 [^6]
- Wisdon Ventures
- Ripple Impact Investments
- Factorial Funds
Anthropic's stated mission is to ensure that AI benefits people and society as a whole. The company isn't attempting to solve a specific AI problem or business problem, but rather identify how we can responsibly advance and adopt AI for the overall benefit of society.
Anthropic was set up by its founders as a public-benefit company in support of its mission. This differentiates the company from its competetitors. While the intent is to be a profitable venture, the public-benefit designation allows Anthropic to act in the benefit for the company versus the benefit of the investors and shareholders. Establishing this type of corporate entity allows Anthropic to meet its goal of having a positive societal impact with AI over share holder profits while still maintaining a viable company.
Anthropic is attempting to solve ethics, safety and research issues within AI as an industry rather than a specific business problem. This further pushes the industry and challenges traditional C corporations.
With some exceptions (such as code suggestions) Anthropic is not providing business specific AI offerings. Rather the company is providing a set of Large Language Models (LLMs) as a foundational platform that businesses can utilize to develop AI solutions to their own unique needs. This allows Anthropic to have a wide variety of customers across industries including (Not comprehensive) [^7]
- GitLab (Technology)
- Steno (Legal Services)
- Gamma (Presentation/Asset Creation Services)
- Jumpcut (Film Industry/Training)
- Asana (Software Services)
- Bridgewater (Investment Services)
- LexisNexis (Legal Services)
- Lonely Planet (Travel services)
- Slack (Communication & Collaboration Services)
Anthropic further serves their customer base through partnership and alliances with services providers such as Amazon Web Services and implementation/consulting partners like Accenture and BCG for customer support.
Anthropic's goal is to build reliable, interpretable, and steerable AI systems. One of their AI systems, Claude, is a group of models that includes Haiku, Sonnet, and Opus. Claude is capable of processing text inputs and multiple images in one request. Claude 3 Opus, a newer version of Claude, has outperformed other competitor chatbots on benchmark tests and is particularly useful for analyzing scientific data, generating computer code, and solving mathematical problems.
Anthropic provides a foundational AI capability called Claude capable of advanced reasoning, vision analysis, code generation and multilingual processing. Claude allows for sub system use designed for functional vs. cost based needs for customers, including [^8]
- Claude - Haiku: A fast low cost model for lightweight workloads
- Claude - Sonnet: Balances cost, efficiency and speed for high throughput tasks
- Claude - Opus: Model handling complex analysis and higher order math and coding add a greater price point.
Anthropic uses a variety of technologies in building their offerings to customers. Many technologies are established as part of investment terms as previously mentioned in the funding analysis. Some of these include:
- AWS Trainium and AWS Inferentia chips utilized to develop and train models.
- Google Kubernetes Engine to run inference for Claude
- Google TPU v5e chips to train generative AI models
Anthropic is most well known for overall AI advancement of Large Language Models (LLMs) with a human like learning process. This work is coupled with a focus on the safety, security and ethics of such models. The company employs a variety of individuals across several disciplines to achieve this goal including experts in machine learning, neuroscience and cognitive sciences.
The AICamp blog provides a comprehensive overview of the foundational concepts behind Claude and a comparison of Claude vs. its OpenAI competitor. [^15]
The following include a summary of Anthropics product from the blog.
Anthropic has a smart computer program called Claude. Here’s what you need to know about it:
- Claude is made to chat with people. It’s really good at having natural conversations.
- It uses a mix of smart tech to remember what you talked about before and keep the chat going.
- The information Claude learned from was carefully picked to be good and fair. This means it tries to avoid saying anything wrong or hurtful.
- Claude is really good at chatting while making sure it stays safe and follows rules set by Anthropic.
- Anthropic made sure Claude can explain why it says certain things, which helps people trust it more.
- Overall, Claude is all about chatting safely and smartly.
Anthropics key competitors include other large players on the Large Language Models (LLMs) field and include primarily:
- OpenAI's ChatGPT
- Google's Gemini
While Anthropic does focus its business on the development and advancement of LLMs like its competitors it attempts to differentiate itself through a purposeful focus on transparency, safety and the ethics of AI. This has begun to push the industry trends in this area across multiple organizations.
Several example areas of responsible AI focus for the company that may be of interest include (but are not limited to):
- This policy provides guidelines for managing and assessing an AI models risks as the capabilities of AI continue to advance.
- The RSP provides a framework of 4 AI Saftey Levels to assessing the potential of risk and is loosely based on governmental biosafety level standards for hazardous biological materials. Essentially, the higher the risk the more stringent the safety standards.
- This RSP is a clear commitment by Anthropic to its alignment with the voluntary commitment made by several US companies to manage the risks associated with AI.
- Anthropic is committed to the implementation of strict child safety measures in all aspects of AI technology. As part of this, Anthropic has aligned with Thorn which is a non-profit dedicated to mitigating the risks of AI to children.
- With its commitment to the safety of AI systems Anthropic has implemented several policies for child protection, extensive red teaming and partnerships with industry experts to ensure the safety of all AI models. Any form of material that exploits children is strictly forbidden and reported to appropriate agencies.
- While not a strict differentiator for Anthropic, the introduction of developer prompts allows for the increased effeciency of development work. This has become a necessary feature of many coding tools and environments and has allowed Anthropic to keep up with industry standards.
- This feature allows for streamlined development, test case and test suite generation.
Anthropics contributions have not however, just been limited to its advancements in safety, security and ethics. The company has introduced through its partnerships additional products and capabilities for individuals and businesses. Such as:
Anthropic Artifacts [^16]
- Artifacts is a feature that integrates with a company's Claude chatbot in a manner allowing users and AI realtime collaboration. This has the potential to allow for fast modification to workflows and processes within an organization.
- This has potential implications to allow for real time assistance such as development assistance, real time writing aids, project management assitance designed to increase human productivity in a collaborative manner with AI.
Claude iPhone App [^17]
- Anthropic in collaboration with Amazon has released its first rendition of Claude as an iPhone app.
- This effort was born out of customer feedback requesting a more streamlined dedicated enterprise app option as many customers employees were already using Claude within their companies.
- The application offers access to all of Anthropic's models with various pricing options for flexibility.
- An Android version is also under development.
The evaluation of Anthropic against its competitors can be viewed across a variety of metrics including financial funding and performance metrics.
When compared to competitor funding Anthropic is not at the head of the pack. While recent funding from investors such as Amazon and Google have been impressive it cannot compare to the funding for OpenAI by Microsoft, who owns approximately 49% of OpenAI equity and has integrated the platform into many Microsoft products [^12]. Through a variety of funding sources OpenAI is working toward funding valued at $100 Billion[^13]. Anthropic falls significantly shorter at it's $7 billion valuation as noted in the funding section of this case study.
However, it does appear that Anthropic has worked toward investment and alignment with Microsoft competitors such as AWS and Google, with AWS offering Anthropic's services as part of its own cloud service offereings.
Despite Anthropics more limited funding, the company has taken steps to learn from potential flaws in the structure of competitor companies. Late in 2023 OpenAI's CEO was fired by the board due to a lack of confidence. While this did not last as a result of wide spread backlash from large investors and the OpenAI employee base the disruption caused ripples across the industry. Anthropics unique corporate setup as a public benefit company would provide an advantage in a similar case as the company's board is bound to benefit trust as well as shareholders. [^14]
From a technical performance perspective it is clear that Anthroptic has dedicated some focus toward ensuring the efficiency and effectiveness of its flagship product, Claude against several of its largest competitors.
In it's most recent performance analysis Anthropic benchmarked it services for the LLM Claude vs. 3.x against several common competitor systems. The image referenced below from Anthropics public website demostrates the performance achieved in comparison to competitors.
With Anthropics focus on providing a societal benefit over a shareholder profit motive it may wish to consider a future focus on models that align with that stated strategy.
Options could include work in
- Health Sciences & Medicine
- Pharmaceutical Advancements
- Environmental and Earth Sciences
Their current alignment and partnership with Amazon Web Services could provide an ideal technical foundation for building services and releasing those options to the wider community of companies. AWS currently maintains over 30% of the cloud infrastructure marketshare [^18] allowing Anthropic to reach a wide audience of potential users.
[1] https://aibusiness.com/verticals/eleven-openai-employees-break-off-to-establish-anthropic-raise-124m#close-modal "Open AI Employees Establish Anthropic"
[2] https://www.anthropic.com/company "Anthropic Mission Statement"
[3] https://aibusiness.com/verticals/eleven-openai-employees-break-off-to-establish-anthropic-raise-124m "Anthropic CEO on purpose of Anthropic"
[4] https://www.aboutamazon.com/news/company-news/amazon-anthropic-ai-investment "Amazon investment in Anthropic"
[5] https://www.wsj.com/tech/ai/google-commits-2-billion-in-funding-to-ai-startup-anthropic-db4d4c50 "Google investment in Anthropic"
[6] https://menlovc.com/perspective/our-investment-in-anthropic-the-foundation-layer-for-generative-ai/ "Menlo investment in Anthropic"
[7] https://www.anthropic.com/customers "Anthropic customers"
[8] https://www.anthropic.com/claude "Anthropic offerings"
[9] https://www.anthropic.com/news/reflections-on-our-responsible-scaling-policy "Anthropic Responsible Scaling Policy (RSP)"
[10] https://www.anthropic.com/news/child-safety-principles "Anthropic Safety By Design Principals"
[11] https://www.anthropic.com/news/evaluate-prompts "Evaluation of Developer Prompts"
[12] https://blogs.microsoft.com/blog/2023/01/23/microsoftandopenaiextendpartnership/ "Microsoft and OpenAI Partnership"
[13] https://www.bloomberg.com/news/articles/2023-12-22/openai-in-talks-to-raise-new-funding-at-100-billion-valuation "OpenAI funding valuation"
[14] https://time.com/6983420/anthropic-structure-openai-incentives/ "How Anthropic is different from OpenAI as a benefit trust"
[15] https://aicamp.so/blog/anthropic-vs-openai-a-comprehensive-comparison "Claude vs OpenAI comparison"
[16] https://www.pymnts.com/artificial-intelligence-2/2024/anthropics-artifacts-looks-to-boost-human-ai-collaboration-in-business/ "Anthropic Artifacts"
[17] https://www.cnbc.com/2024/05/01/anthropic-iphone-ai-app-business-plan-to-compete-with-openai-announced.html "Claude iPhone App"
[18] https://www.statista.com/statistics/967365/worldwide-cloud-infrastructure-services-market-share-vendor/ "AWS Marketshare"