These are days of great excitement for artificial intelligence and for companies that have developed or are developing the most advanced conversational chatbots.
We have regularly written about the ballet of OpenAI’s top management in several articles. And after revolutions and counter-revolutions, it now seems that Sam Altman is firmly back at the helm, with an almost completely renewed board of directors.
Anthropic also somehow entered the OpenAI affair, because according to rumors the old board of the company would have contacted Dario Amodei, founder and CEO of Anthropic, in the hypothesis of a merger between the companies. With Amodei himself who would have had the role of CEO.
The merger, as we know, did not take place. And indeed, the rivalry in the field of AI has been revived by news in the last few hours: Anthropic has announced its very powerful chatbot Claude 2.1.
Anthropic annuncia Claude 2.1
The announcement was made by Anthropic with a long article published on the official website on Tuesday 21 November.
In which we can read about the availability of the new Large Language Model (LLM) Claude 2.1.
And in somewhat cryptic language (but we will explain everything shortly) we discover that the chatbot will have a 200,000 token context window and a “significant reduction in hallucination rate”.
Better understanding the meaning of these sentences will help us discover the potential of Claude 2.1.
Claude 2.1 twice as powerful as GPT-4 Turbo
In this context, the token is the “weight” of a word or a space between two words. It is, in a nutshell, the unit of linguistic measurement of the Large Language Model.
When, at the DevDay on November 6, OpenAI presented its new GPT-4 Turbo with great fanfare, it spoke of a “128K context”, i.e. 128,000 tokens.
Bene: Claude 2.1’s context window is 200,000 tokens, therefore almost double that of the most advanced OpenAI chatbot (and double the size of Claude’s previous version). Translated into even simpler words, a context of 200,000 tokens allows the software to process texts of approximately 150,000 words, or 500 pages, in a single prompt. Concretely, with a single command Claude 2.1 will be able to assimilate texts such as the Odyssey.
The meaning of the phrase hallucination rate is more intuitive. By which we mean a response given by a large language model in an unexpected way and not in accordance with the data assimilated during training. We have read of several occasions in which chatbots sensationally invented answers, which later turned out to be wrong. It is useless to add the danger of such errors. Well: Claude 2.1 halves the hallucination rate.
Company statements
The note reads: “Processing a message 200 thousand tokens in length is a complex undertaking and an absolute novelty in the sector. We’re excited to put this powerful new feature in the hands of our users.
It only takes a few minutes for Claude to complete tasks that would normally require hours of human effort. We expect latency to decrease substantially as the technology advances.”
The launch of Claude 2.1 and more
It has been noted from many quarters that the timing of the announcement of Claude 2.1 by Anthropic may not have been coincidental.
In fact, they coincided with a moment of crisis for OpenAI. As a counter-proof, think about many statements released in recent hours by several other companies involved in the production of AI software. Elon Musk, for example, made it known that his Grok software will be available next week for all Premium+ users of X. And then, Stability AI launched Stable Video Diffusion, to create artificial intelligence videos starting from text prompts. Additionally, Google announced that Google Bard will be able to provide information about YouTube videos.
Investments in Anthropic
We remind you that Anthropic was founded by two brothers of Italian origins, Dario and Daniela Amodei, who had previously worked for OpenAI.
The Amodei brothers’ company has recently attracted very important players in the sector. If Google invested 2 billion euros, Amazon’s economic commitment was double, 4 billion euros.
Over the past year, Anthropic has received $7 billion in funding, but OpenAI has secured $13 billion (as of 2019) from Microsoft alone.
We expect a near future of sparks between the two competitors. It will be a challenge to the last token.