Whereas OpenAI is within the middle of an existential crisis, there’s a brand new chatbot replace from Anthropic, the Google-backed AI startup based by former OpenAI engineers who left over disagreements in regards to the firm’s more and more industrial course as its Microsoft partnership went on.
Anthropic has announced that the most recent replace of its chatbot, Claude 2.1, can digest as much as 200,000 tokens directly for Professional tier customers, which it says equals over 500 pages of fabric.
The corporate additionally says Claude will hallucinate — or lie — half as usually as earlier than, and it could actually do issues like search the online or use a calculator by customizable instruments. The chatbot additionally now helps customized, persistent directions and has a brand new check window for making an attempt out prompts.
The 200,000-token function doubles what Claude might deal with earlier than, and it’s a considerably increased restrict than the 32,000-token ceiling of the priciest model of GPT-4. Anthropic calls this an “trade first” and says it permits the chatbot to pore over uploads of whole codebases or works just like the “Iliad.” The upper token depend is unique to the corporate’s Professional plan.
Anthropic additionally introduced Claude a little bit nearer to ChatGPT with the brand new beta software use function. Customers can now join API instruments, and Claude will decide the most effective one for the job relying on the context. That features utilizing a calculator or looking the online, and the corporate says customers can ask for particular API calls utilizing pure language.
Lastly, the corporate mentioned it’s up to date its developer console with a check window for making an attempt out new prompts and has added the power to provide Claude customized persistent directions. Like with GPT-4, this implies you’ll be able to customise the chatbot to reply in particular methods or with particular personalities.