US-based Artificial Intelligence start-up Anthropic launched its conversational chatbot Claude 2 in the US and UK on Tuesday. Reportedly, the latest chatbot is claimed to be the closest rival to OpenAI’s ChatGPT, which has been making waves ever since its launch in 2022. Claude 2 comes with a larger context window compared to ChatGPT and has a superior intelligence/cost ratio against GPT-4.
“We are pleased to announce Claude 2, our new model. Claude 2 has improved performance, longer responses, and can be accessed via API as well as a new public-facing beta website, claude.ai,” the company said in its official blog.
While it slightly trails, Claude 2 offers a similar experience in terms of coding as it comes with fresher training data and supports more context. By far, Claude 2 has the largest context window – 75,000 words. This means users can input lengthy documents, reports, and PDFs and get summaries in seconds. Reportedly, the chatbot’s API comes nearly five times cheaper.
In terms of hallucinations, Claude is reportedly trained on “constitution,” also known as the rulebook, and hence is less likely to show dangerous content. As of now, not many know about the chatbot, but it is only second to ChatGPT when it comes to its abilities and is fast levelling up.
How is Claude v2 different?
The company has said that it made some improvements in coding, math, and reasoning compared to its previous models. The latest model scored 76.5 per cent on the multiple choice section of the Bar exam compared to the 73.0 per cent scored by Claude 1.3.
“When compared to college students applying to graduate school, Claude 2 scores above the 90th percentile on the GRE reading and writing exams, and similarly to the median applicant on quantitative reasoning,” read the blog post.
Claude 2 is comparatively cheaper than GPT-4 and is capable of stronger reasoning and coding than Claude 1. When it comes to standard GREs, ChatGPT won verbal, quantitative, and USMLE, while Claude overtook ChatGT in GRE writing and Bar exams. While it trails, Claude 2 is fast levelling up. Claude 2 comes with 100k context which is nearly 5 times cheaper than GPT-4’s 32K. Prompt tokens are cheaper than completion tokens. They cost $11 compared to $60 per million tokens on GPT-4. Similarly, completion costs are $32 compared to $120 per million tokens, assuming both have similar tokenisation lengths.
Most importantly, Claude’s knowledge cutoff is in early 2023, while GPT-4 is in September 2021.
Those in the US and the UK can simply go to the Claude 2 page on Anthropic’s website and sign up for free. Users need to click on ‘Talk to Claude’ after which they will be prompted to provide an email ID. After confirming the email address, one is ready to go.
Anthropic is funded by Daniela and Dario Amodei, former senior members of the OpenAI team. While Claude 2 is still in its early stages, it can be termed a decent alternative to ChatGPT when it comes to performance.