TThe state of the art in AI just got a little bit further along: On Friday, Anthropic, an AI lab founded by a team of disgruntled OpenAI staffers, released the latest version of its Claude LLM. From Bloomberg:
The company announced on Thursday that a new model of the technology behind its popular chatbot, “Claude,” is twice as fast as its most powerful predecessor. In its evaluation, Anthropik said the model outperformed leading competitors such as OpenAI in several key intelligence capabilities, including coding and text-based reasoning.
Anthropik just released the previous version of Claude, 3.0, in March. This latest model is called 3.5, and it’s currently only available on the company’s mid-range model, “Sonnet.” The company says a faster, cheaper, less powerful “Haiku” version is coming soon, as well as a slower, more expensive, but most powerful “Opus.”
But even before Opus arrived, Anthropic claimed to have the best AI on the market. In a series of head-to-head comparisons posted on the company’s blog, 3.5 Sonnet outperformed OpenAI’s latest model, GPT-4o, in tasks like math quizzes, text comprehension, and undergraduate-level knowledge. It wasn’t a clean sweep, with GPT maintaining the lead in several benchmarks, but it was enough to justify the company’s claim that it’s on the cutting edge of what’s possible.
From a more qualitative perspective, AI seems to be a step forward. Anthropic states:
They have a significantly improved ability to understand nuance, humor, and complex instructions, and they excel at writing high-quality content in a natural, relatable tone.
They’re grading their own homework, and their explanation matches the changes I’ve noticed: No matter where the technical benchmarks are, I find talking to the latest version of Claude more enjoyable than any AI system I’ve used before.
But the company isn’t just selling power updates. Instead, in a way favored by smaller competitors around the world, Anthropic is focusing as much on cost as it is on features. The company claims that Claude 3.5 is not only smarter than its predecessor, but also cheaper.
Source: www.theguardian.com
