Who Is Dario Amodei?

Dario Amodei is an Italian-American artificial intelligence researcher and entrepreneur who has been associated at various points with companies such as OpenAI, Google Brain, and Anthropic. Let’s tell some of his story below.

Category	Details
Full Name	Dario Amodei
Nationality	American
Education	Ph.D. in Physics from Princeton University
Early Career	Research Scientist at Google Brain, Worked on machine learning and artificial intelligence research
Major Companies Founded	Anthropic
Positions	Co-founder and CEO of Anthropic, Former Vice President of Research at OpenAI
Major Projects	Contributions to AI safety and alignment research, Development of large language models and AI systems at OpenAI and Anthropic
Notable Achievements	Key role in advancing AI research, Significant contributions to the field of AI safety and ethics

Table of Contents

Early career

Amodei earned a Ph.D. from Princeton University in 2011 for his work on neural circuits and novel devices for intracellular and extracellular recording.

He then joined Skyline as a part-time software developer and worked on a comprehensive software suite used by researchers in the study of proteins.

Amodei also completed postdoctoral research at the Stanford University of Medicine where he devised new mass spectrometry methods for protein network modeling and biomarker discovery.

On his LinkedIn profile page, Amodei stated that he wrote 14,000 lines of code for the software and over 50 pages of tutorials and documents.

Baidu and Google

Amodei joined Baidu in November 2014.

There, he worked with a small team of AI scientists and systems engineers that included Google Brain co-founder and chief scientist Andrew Ng.

The team worked on difficult problems in AI and deep learning, with Amodei spending most of his time on the Deep Speech 2 series of speech recognition models.

One of the main projects Amodei worked on was Deep Speech 2, an end-to-end deep learning approach that recognized the vastly different languages of English and Mandarin Chinese.

Neural networks were used to handle intricacies of speech such as noisy environments and accents and, in some cases, the output of the resultant system was comparable with transcription provided by human workers.

Amodei also conceived, prototyped, and implemented neural network architectures that achieved significant word error rate (WER) reductions in his lab’s English and Chinese speech systems.

After twelve months or so, he left Baidu and became a deep learning researcher on the Google Brain team. There he worked to extend the capabilities of neural networks and authored several papers on the topic of AI system safety and accident prevention.

OpenAI

Amodei then joined OpenAI in July 2016 where he headed the AI safety team. In an interview with Robert Wilbin of London-based non-profit 80000 Hours, Amodei explained his reasons for moving to the company:

“I thought there were a number of really talented researchers here and it was a good environment in which to think about safety in the context of AI research that’s already been done.’

While he had never worked at a start-up before, Amodei was excited about pushing the boundaries in the AI industry and moving beyond a sole focus on supervised machine learning.

He then became a Research Director in September 2018 and Vice President of Research just over a year later.

As a VP, Amodei was in charge of the team responsible for building GPT-2 and GPT-3.

Collaboration with DeepMind

In 2017, DeepMind announced that it was collaborating with Dario Amodei, Paul Christiano, and Tom Brown at OpenAI.

The collaboration was part of DeepMind’s efforts to assemble diverse voices in the artificial intelligence community and foster positive AI outcomes for people and society.

To that end, Amodei’s efforts were focused on how people could tell a system what to do and, perhaps more importantly, what they didn’t want the system to do.

In essence, Amodei and his staff recognized that AI safety could be increased if the need for humans to write goal functions could be avoided.

Otherwise, simple proxies for complex goals or even complex goals that were slightly incorrect could lead to dangerous or undesirable model behavior.

The end result was a learning algorithm that solves modern RL environments with only small amounts of human feedback. DeepMind and OpenAI’s collaboration also scaled up the approach so that it was useful for more complicated tasks.

Anthropic

Amodei left OpenAI in December 2020 to found the AI start-up Anthropic. The founding team consisted of various senior AI members and also Amodei’s sister Daniela.

Anthropic was started with a mission to ensure that AI does not pose an existential threat to humanity in the future.

The Financial Times also reported that Amodei and 14 other researchers left OpenAI because of concerns about where it was headed after the $1 billion investment from Microsoft in 2019.

Specifically, Amodei believed the investment meant OpenAI’s trajectory had become more corporate and less about democratizing AI. To protect Anthropic from a similar fate, he registered Anthropic as a public benefit corporation – a for-profit entity bound by law to balance profit with social and public good.

Future direction of Anthropic

In April 2023, TechCrunch reported that Anthropic planned to raise up to $5 billion over the next two years. According to internal company documents, the capital would be used to expand into a dozen major industries and better compete with rival OpenAI.

Underpinning the move is a so-called “frontier model” Amodei and his team has called Claude-Next. The model is reputedly 10 times more powerful than its competitors but will nevertheless a $1 billion investment in the next 18 months.

Anthropic describes Claude-Next as a “next-gen algorithm for AI self-teaching” based on its constitutional AI training technique.

This technique seeks to align artificial intelligence and human intentions with systems that respond to questions and perform tasks according to a simple set of guidelines.

Series C pitch deck

In a leaked pitch deck for the Series C funding round, Amodei revealed that despite his earlier concerns about commercialization, this was exactly the path Anthropic would be headed down:

“Anthropic has been heavily focused on research for the first year and a half its existence, but we have been convinced of the necessity of commercialization, which we fully committed to in September [2022].”

The pitch deck also explained that Anthropic had developed an initial product specialization and go-to-market strategy in line with the company’s brand and expertise.

Key takeaways:

Dario Amodei is an Italian-American artificial intelligence researcher and entrepreneur who has been associated at various points with companies such as OpenAI, Google Brain, and Anthropic.
Amodei joined Baidu in 2014 and worked with a small team of AI scientists and systems engineers that included Google Brain co-founder and chief scientist Andrew Ng. He then left Baidu and became a deep learning researcher on the Google Brain team with a particular focus on AI safety.
After almost five years as Vice President of Research at OpenAI, Amodei founded the public benefit corporation Anthropic over concerns that OpenAI had become too commercial. Anthropic was started with a mission to ensure that AI does not pose an existential threat to humanity in the future.

Key highlights from the story of Dario Amodei:

Early Career: Dario Amodei earned his Ph.D. from Princeton University in 2011, focusing on neural circuits and devices for recording. He worked on software for protein study and completed postdoctoral research at Stanford University of Medicine.
Baidu and Google: He joined Baidu in 2014, working on AI and deep learning, including the Deep Speech 2 series for speech recognition. Later, he joined Google Brain, where he contributed to neural network advancements and AI safety.
OpenAI: Amodei joined OpenAI in 2016, initially heading the AI safety team. He later became a Research Director and then Vice President of Research, overseeing projects like GPT-2 and GPT-3.
Collaboration with DeepMind: Amodei collaborated with DeepMind on AI safety, focusing on improving communication with AI systems and avoiding dangerous model behavior due to incorrect goal functions.
Founding Anthropic: Amodei left OpenAI in 2020 to start Anthropic, a public benefit corporation focused on preventing AI from becoming an existential threat. He believed that OpenAI’s commercialization trajectory was problematic.
Anthropic’s Direction: Anthropic planned to raise up to $5 billion in capital to expand its AI technology, especially the “Claude-Next” model, and compete with other AI organizations.
Series C Funding: A leaked pitch deck revealed that Anthropic shifted from a research-focused approach to a more commercial one, aligning its products and strategy accordingly.