Anthropic launches Claude Instant 1.2 which hallucinates less than other models

Anthropic has announced the launch of Claude Instant 1.2, a large language model available to businesses through an API. The company described it as a faster, lower-priced yet ‘very capable model’ that can perform a range of tasks such as casual dialogue, text analysis, summarization, and document comprehension.

One of the cool things about this LLM is that it scores the best out of all Claude models in an automated red-teaming evaluation. This means that it offers greater safety, hallucinates less, and is more resistant to jailbreaks.

When it comes to maths and coding, two niches that several AI model producers have been focused on, Claude Instant 1.2 sees some improvement over Claude Instant 1.1.

In the Codex P@1 Python programming benchmark, it scored 58.7% compared to the older version’s 52.8%. In the GSM8k grade-school maths problems benchmark it score 86.7% compared to the prior version"s 80.9%.

While these improvements were pretty big, the new model did fall behind the older version ever so slightly in a few benchmarks but compared to the gains, the declines are negligible.

Unlike Claude 2 which is available to interact with directly on Anthropic’s websites, Claude Instant 1.2 is only available through an API for businesses. Thankfully, you can use it by heading over to Poe.com which lets you use lots of LLMs including the recently released Llama 2 model from Meta.

Poe supports a large range of models including Google’s PaLM and OpenAI’s ChatGPT. In fact, one of the advantages of using Poe to access ChatGPT instead of OpenAI’s own interface is that a phone number isn’t required to make an account.

While most people don’t need to worry about entering their phone number, having Poe available to access ChatGPT is a major boon for anyone without a phone.

Let us know in the comments if you use generative AI. What is your favourite language model?

Source: Anthropic