29 August 2023, Bengaluru, India
Anthropic, the AI business co-founded by former OpenAI executives, has unveiled an upgraded version of Claude Instant. This faster, more affordable text-generating model can be accessed through an API.
According to Anthropic, the upgraded Claude Instant, Claude Instant 1.2, combines the advantages of the company’s recently unveiled flagship model, Claude 2, and exhibits “significant” improvements in maths, coding, reasoning, and safety. In internal testing, Claude Instant 1.2 outperformed Claude Instant 1.1, scoring 86.7% on a set of maths questions compared to Claude Instant 1.1’s 80.9% and 58.7% on a coding benchmark.
According to a blog post by Anthropic, “Claude Instant generates longer, more structured responses and follows formatting instructions better.” “Instant 1.2 also shows improvements in quote extraction, multilingual capabilities, and question answering.”
Anthropic believes Claude Instant 1.2 is more immune to jailbreaking attempts and less likely to experience hallucinations. In the context of large language models like Claude, “hallucination” is the phenomenon where a model produces text that is incomprehensible or wrong, while “jailbreaking” is a method that uses deftly worded cues to get over the security measures put in place by the architects of large language models.
Additionally, the 100,000 token context window in Claude Instant 1.2 is the same size as the one in Claude 2. Tokens represent raw text; whereas context window refers to the reader, the model takes into account before generating more text (for example, the word “fantastic” would be broken up into the tokens “fan,” “tags,” and “tic”). Claude Instant 1.2 and Claude 2 can analyze around 75,000 words, or about the length of “The Great Gatsby.”
As we’ve previously covered, Anthropic’s goal is to develop a “next-gen algorithm for AI self-teaching,” as the company puts it in its pitch deck to investors. A virtual assistant that can answer emails, conduct research, create art, books, and other tasks could be made with such an algorithm. We have already seen some of these capabilities in huge language models like GPT-4.
But this algorithm needs to be Claude Instant. Instead, it’s meant to compete with comparable entry-level products like OpenAI and startups like Cohere and AI21 Labs, creating and commercializing their text- and, occasionally, image-generating AI systems.
Anthropic, founded in 2021 and run by Dario Amodei, a former vice president of research at OpenAI, has so far received $1.45 billion at a valuation of a few billion. While that may seem like a lot, the business projects that it will require $5 billion over the next two years to build the chatbot it has in mind.
Anthropic states it has “thousands” of clients and collaborators, including Quora, which offers access to Claude and Claude Instant via its pay-per-use generative AI tool Poe. In conjunction with OpenAI’s ChatGPT, Claude drives DuckDuckGo’s freshly released DuckAssist utility, which immediately responds to users’ basic search requests. Additionally, Claude is a component of the Notion workspace’s technical backend for Notion AI, an AI writing helper.
[Source of Information : Techcrunch.com]
