When OpenAI announced GPT-4, its latest multi-language model, in March last year, it sent shockwaves through the tech world. He was clearly more capable than anything before at chatting, coding, and solving all sorts of thorny problems – including school homework.
Anthropic, an OpenAI rival, announced today that it has developed its own artificial intelligence enhancement that will improve chatbots and other applications. Although the novel model is in some respects the best in the world, it is a step forward rather than a massive leap.
Anthropic’s novel model, called Claude 3.5 Sonnet, is an update to the existing Claude 3 family of artificial intelligence models. It is more adept at solving math, coding and logic problems measured by widely used benchmarks. Anthropic claims that it is also much faster, has a better understanding of language nuances and even has a better sense of humor.
This is undoubtedly useful for those trying to build applications and services based on Anthropic’s AI models. But the company’s news is also a reminder that the world is still waiting for the next step forward in artificial intelligence, similar to the one provided by GPT-4.
OpenAI has been expected to release a sequel called GPT-5 for over a year, with the company’s CEO Sam Altman saying: encouraged speculation that it will provide the next revolution in artificial intelligence capabilities. GPT-4 has cost over $100 million to train, and GPT-5 is widely expected to be much larger and more pricey.
Although OpenAI, Google and other AI developers have released novel models that outperform GPT-4, the world is still waiting for the next massive leap. Progress in AI has recently become more gradual, relying more on innovations in model design and training rather than model size scaling and brute-force calculations as GPT-4 did.
Michael Gerstenhaber, head of product at Anthropic, says the company’s novel Claude 3.5 Sonnet is larger than its predecessor, but draws much of its novel competence from training innovations. For example, the model received feedback to improve its logical reasoning skills.
Anthropic claims Claude 3.5 Sonnet outperforms top OpenAI, Google and Facebook models in popular AI tests, including GPQAa test of specialist knowledge at master’s level in biology, physics and chemistry; MMLU, a test covering computer science, history and other topics; AND HumanEval, a measure of coding skill. However, improvement is only a matter of a few percentage points.
This latest advancement in artificial intelligence may not be revolutionary, but it is happening at a rapid pace: exclusively anthropic announced its previous generation models three months ago. “If you look at the rate of change in intelligence, you appreciate how fast we are moving,” Gerstenhaber says.
More than a year after GPT-4 sparked a frenzy of novel AI investments, massive novel leaps in machine intelligence may prove more complex to achieve. As GPT-4 and similar models are trained on enormous swaths of text, images and video available on the Internet, it is becoming increasingly complex to find novel sources of data to utilize in machine learning algorithms. Much larger models are expected to cost billions of dollars to have greater learning capacity. When OpenAI announced its own update last month, with a voice-and-video model called GPT-4o, the focus was on a more natural and human interface rather than much smarter problem-solving capabilities.