In 2020, OpenAI’s vice president of research Dario Amodei left the artificial intelligence company and cofounded a rival startup called Anthropic. In less than three years, his new company has grown to be worth more than $5 billion and, like OpenAI, earned the backing of Big Tech.
Today (Sept. 25), Anthropic announced Amazon (AMZN) has agreed to invest up to $4 billion in the company with an initial investment of $1.25 billion for a minority stake. Amazon’s move mirrored that of Microsoft in January when it committed to a $10 billion long-term partnership with OpenAI.
Amodei, Anthropic’s CEO, implied he left OpenAI to start his own company because had different ideas from OpenAI CEO Sam Altman. “One thing I’ve learned is that it can be pretty ineffective to argue with your boss and say, ‘Your company shouldn’t do X, it should do Y.’ A much more effective thing to do is, ‘I’m starting a company. We are going to do X and see how it works,” Amodei said during an onstage interview at TechCrunch Disrupt last week.
Anthropic’s main product is Claude, a ChatGPT-like A.I. text generator. Anthropic uses a unique training method called “constitutional A.I.” that aims to build A.I. systems that are not only powerful and intelligent but aligned with human values. Anthropic has been a client of Amazon’s Web Services (AWS) since 2021. As part of the investment agreement, Anthropic will use AWS as a primary cloud provider and use AWS Trainium and Inferentia A.I. chips to build, train and deploy its future foundation models, Amazon said.
Until today’s Amazon deal, Anthropic had raised $1.5 billion in funding—from investors including Google, Salesforce and Zoom—and was most recently valued at $5 billion, making it among the highest-valued A.I. startups in the world. Amazon didn’t disclose how much Anthropic was worth before or after it signed the investment deal.
But Amodei said the funding amount is small relative to how much Anthropic might need for new product development and the potential revenue it could bring in. “It’s all relative,” he said at TechCrunch Disrupt.
Anthropic focuses on enterprise customers, Amodei said, who use Claude to speed up “knowledge work in professional services.” For example, legal data analytics firm LexisNexis is a client. Claude is popular among financial service and legal service providers, Amodei said.
Earlier this month, Anthropic launched its first consumer-facing subscription of Claude 2. The company plans to build a “frontier model” that is 10 times more capable than today’s most powerful A.I. and the development will require $1 billion in spending over the next 18 months, according to TechCrunch.
Amodei worked at OpenAI for five years, leading the development of GPT-2 and GPT-3, the precursors of the large language model powering ChatGPT today. When he left, he brought along a few coworkers, including his sister, Daniela Amodei, who led OpenAI’s recruiting effort and managed its technical safety and policy teams, according to her LinkedIn page.
Amodei said his day-to-day activities now consist of “a mix of things that are very normal and things that are very unusual.” “Half the day it’s fairly normal things any startup would encounter,” such as monitoring operations, hiring for key positions and talking to potential customers, he said at TechCrunch Disrupt, and “the other half can be very unusual because of where we think the technology is going.” That largely involves communicating with parties outside the tech world about the risks of A.I. and how to mitigate them.
In July, Amodei testified before the Senate Judiciary Committee on potential regulation of A.I. Amodei said he fears cutting-edge A.I. could be used to create dangerous viruses and other bioweapons and standard tests must be created to test advanced A.I. models before they can be released to the public.
“There’s something surreal about ping-ponging back between those two. I’ve never seen an industry that works that way,” Amodei said.
