What is Anthropic AI?
You’ve probably heard of ChatGPT, Bard and Open AI. But what is Anthropic AI?
Anthropic is an artificial intelligence safety and research startup founded in 2021.
The startup focuses on building large language models (LLMs), general AI systems and chatbots such as Claude.
It is OpenAI’s direct competitor and is currently based in San Francisco.
The founders, Dario and Daniela Amodei, quit OpenAI to build Anthropic because they believed that these LLMs need:
- More computational power to improve performance
- Alignment and safety
Alignment here means ensuring that these AI models actually do what humans ask them to.
This became Anthropic’s core values: “to create reliable, interpretable and steerable AI systems”.
The startup focuses heavily on AI safety and using artificial intelligence responsibly.
Despite being a startup, Anthropic is also a public-benefit corporation.
This means it’s a for-profit corporation that focuses on leaving a positive impact on humanity.
Anthropic does this by conducting AI safety research and setting constitutional AI for its chatbot, Claude.
The constitutional AI are a set of laws and rules that govern the AI to ensure it is non-harmful towards humans.
Today, Amazon and Google is backing Anthropic.
The History Behind Anthropic AI
How It Started
In 2019, OpenAI took a shocking $1 billion investment from Microsoft.
This caused Daniela Amodei, Dario Amodei and 9 other OpenAI employees to quit and start Anthropic.
The siblings say this was due to differences in the OpenAI’s direction.
Investments
In 2021, Anthropic had their Series A funding and raised $124 million from investors like Skype’s Co-founder, Jaan Tallin.
In 2022, Anthropic raised several more rounds. $300 million from Google for a 10% stake in the company. FTX and its sister company, Alameda Research had also invested $580 million in Anthropic.
But, Anthropic’s fundraising didn’t stop there.
In March 2023, they received $300 million from Google. Two months later, they raised $450 million for their series C funding. In June, Saphire Ventures committed more than $1 billion to Anthropic. In August, $100 million from SK Telecom from South Korea. Amazon announced a partnership with Anthropic and up to $4 billion investments. And most recently, Google has committed up to $2 billion in October 2023.
Projects
Claude AI is a generative AI chatbot released by Anthropic to rival ChatGPT.
Claude Instant is a smaller, faster and cheaper version of the Claude AI large language model.
Unlike ChatGPT, Claude AI was trained with AI safety in mind.
Here’s a quick timeline of its release:
- On 14 March 2023, Anthropic released Claude 1.0 and Claude Instant 1.1
- On 18 April 2023, Claude 1.3 was released.
- On 11 July 2023, Claude 2.0 was released.
- On 9 August 2023, Claude Instant 1.2 was released.
- On 21 November 2023, Anthropic released Claude 2.1
Constitutional AI: Anthropic’s Unique Approach
It’s not an understatement to say that the swift rise in AI is unsettling, and scary.
After all, how can we be certain that future AI won’t mirror the dystopian scenes from Terminator?
Researchers from Anthropic believe that Constitutional AI is key.
It’s a framework that trains AI systems to align itself with human values and to behave in honest, harmless and helpful manners.
These values will help the language model decide:
- What questions are inappropriate?
- What responses should be encouraged?
Most AI are not trained with Constitutional AI.
ChatGPT uses a method called reinforcement learning with human feedback (RLHF) to determine the values that would guide its behaviour.
For example, if I asked ChatGPT a question and the response was not helpful, I could flag its response and provide further feedback.
ChatGPT would use this feedback to learn and improve its model.
But there’s problems with this model:
It requires humans to provide feedback which is not efficient for scaling.
After all, there’s too many data to process.
Instead, Constitutional AI uses a set of rules to make judgements for the outputs. This set of rules are the “Constitution” that we define (or in this case, Anthropic).
Examples of Constitution can be:
“Make sure the response is ethical, not racist, not sexist, helpful and legal”.
This way, the AI model can critique itself which reduces the need for human intervention.
This is also known as “reinforcement learning with AI feedback (RLAIF)”.
Meet Claude: The Star of Anthropic
Anthropic created an AI chatbot called Claude which is powered by a large language model (LLMs) called Claude 2.
It was first released in March 2023.
Similar to ChatGPT, Claude is trained on natural language processing.
This allows it to be able to have human-like conversations.
And Claude can do almost everything ChatGPT does.
Like tasks such as:
- responding to questions
- write content like emails
- summarise texts
- write code
- and more
There are currently 3 Claude models:
Claude 1 – The original version of Claude, created in 2021, with limited conversational abilities.
Claude 2 – Launched in 2022 and was a significant improvement from Claude 1. It has more capabilities like being able to carry out more complex instructions.
Claude-Instant – smaller, faster and cheaper version of Claude 2
Claude is regularly trained on up-to-date information and can read more words than ChatGPT (75,000 words) at a time.
And since Claude was trained using Constitutional AI, it can continuously be updated while retaining its safety constraints.
ChatGPT has not been able to demonstrate this.
Using Claude: A Simple Guide
Currently, Claude is available in 95 countries.
If you don’t see your country in the list, you can use a VPN and follow the next steps.
Note: You only need to sign up using the VPN, afterwards you can ditch the VPN.
Step 1: Go to Claude AI
Step 2: Sign up using your email and verify your phone number.
Note: You can only use the same number once.
Step 3: Once you’re in, you can start asking Claude questions.
