Artificial intelligence (AI) startup Anthropic said its latest generative AI models outperform those from Google and OpenAI.
The Claude 3 family of models, announced Monday (Mar. 4), include versions known as Haiku, Sonnet and Opus, and are purported to be the company’s most powerful chatbots to date. Haiku will be launched soon, while Sonnet and Opus are available now.
“Opus, our most intelligent model, outperforms its peers on most of the common evaluation benchmarks for AI systems, including undergraduate level expert knowledge, graduate level expert reasoning, basic mathematics, and more,” Anthropic said on a Monday (March 4) blog post.
“It exhibits near-human levels of comprehension and fluency on complex tasks, leading the frontier of general intelligence,” Anthropic added.
Haiku, when released, will be “the fastest and most cost-effective model on the market,” able to digest complex research papers “in less than three seconds.”
Sonnet is also available on Amazon Bedrock, with plans for Opus and Haiku to be available on the platform in a few weeks, per the post.
The new versions are multimodal, meaning they can tackle both text and visual prompts. As noted here last month, this technology marks “perhaps the biggest jump AI has made in the past year … the ability to parse visual, verbal, audio and textual data all together multimodally.”
And while past versions of Claude were prone to making unnecessary refusals that indicated a lack of contextual understanding, Anthropic said it has made progress on correcting this issue.
“Opus, Sonnet, and Haiku are significantly less likely to refuse to answer prompts that border on the system’s guardrails than previous generations of models,” the company said.
Last month saw reports that Anthropic and Google were working to address the limitations of its GenAI systems, with representatives from both firms addressing the limitations of the technology at The Wall Street Journal (WSJ) CIO Network Summit.
A report by the WSJ noted that Anthropic is working to reduce hallucinations and improving accuracy, with Jared Kaplan, the company’s co-founder and chief science officer, mentioning the development of data sets where the AI model responds with “I don’t know” when it has insufficient information.
This approach makes sure that the AI system only offers answers when it can provide citations to back up its responses.
Kaplan stressed the importance of finding the proper balance between caution and usefulness. While reducing hallucinations is critical, too-cautious AI models might respond with “I don’t know” to every query, making them less useful.
Anthropic said Monday that it would soon enable citations in its Claude 3 models “so they can point to precise sentences in reference material to verify their answers.”
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
We’re always on the lookout for opportunities to partner with innovators and disruptors.
Learn More