anthropic is the new ai research outfit from openai’s dario amodei, and it has $124m to burn

The Rise of Anthropic: A New Approach to AI Safety

The field of Artificial Intelligence is undergoing a significant shift, moving beyond academic research to encompass powerful, industry-leading models such as GPT-3. This evolution necessitates a new focus on responsible development, a belief championed by Dario Amodei.

Founding and Core Principles

Formerly the VP of research at OpenAI, Amodei has established Anthropic alongside his sister, Daniela. The company’s central aim is the creation of large-scale AI systems that are demonstrably steerable, readily interpretable, and inherently robust.

The Challenge of AI Opacity

A key issue the Amodei siblings are addressing is the inherent lack of transparency within these advanced AI models. While models like GPT-3 exhibit remarkable versatility in generating human-quality text across diverse topics, their internal workings remain largely enigmatic.

Consider the task of generating rhyming couplets inspired by Shakespeare and Pope. How does the AI achieve this? What processes are occurring internally? Identifying specific parameters to adjust for desired outcomes – such as mood or vocabulary – proves exceptionally difficult. The precise mechanisms behind this convincing text generation remain obscure.

High-Stakes Applications Demand Explainability

The lack of understanding isn’t merely an academic concern. When AI systems are deployed in critical applications, such as security surveillance or legal decision-making, explainability becomes paramount. A troubling trend exists: as system power increases, the ability to decipher its reasoning diminishes.

Anthropic’s mission statement reflects this concern: “Large, general systems of today can have significant benefits, but can also be unpredictable, unreliable, and opaque: our goal is to make progress on these issues.” The company initially prioritizes research, anticipating future commercial and public benefits.

Integrating Safety from the Ground Up

The company intends to embed safety considerations directly into the core of AI development, rather than attempting to add them as an afterthought. Building models with inherent interpretability may require more effort than constructing powerful, yet opaque, systems.

Funding and Vision

Dario Amodei, now CEO of Anthropic, announced the company’s launch and $124 million in funding. This investment was led by Skype co-founder Jaan Tallinn, with contributions from prominent figures including James McClave, Dustin Moskovitz, and Eric Schmidt.

Anthropic operates as a public benefit corporation, currently focused on fundamental research. The aim is to develop more capable, general, and reliable AI systems, and to deploy them in a manner that demonstrably benefits society.

The Significance of the Name

The company’s name, “Anthropic,” is deliberately linked to the concept of anthropocentrism – the relevance to human experience. It also alludes to the “Anthropic principle,” which suggests that the universe’s conditions allow for intelligent life because, demonstrably, it exists.

This implies that Anthropic’s objective is to cultivate the conditions necessary for the development of beneficial and understandable artificial intelligence.