OpenAI’s announcement of GPT-4, their latest large language model in March, was groundbreaking in the tech world. It displayed superior abilities in chatting, coding, and solving various complex problems, including school assignments.
Anthropic, a competitor to OpenAI, has now announced their own AI advancement that will enhance chatbots and other applications. While their new model is considered the best in some aspects, it represents more of a step forward rather than a significant leap.
Anthropic’s latest model, Claude 3.5 Sonnet, is an upgrade from their existing Claude 3 AI model series. It excels in solving math, coding, and logic problems based on industry benchmarks. Anthropic claims it is faster, better at interpreting language nuances, and even displays a sense of humor.
While this is beneficial for developers utilizing Anthropic’s AI models, it signifies that the world is still awaiting a major advancement in AI similar to that of GPT-4.
The anticipation for OpenAI to release GPT-5 has been growing for over a year. Speculation, encouraged by the company’s CEO Sam Altman, suggests that GPT-5 will introduce another revolution in AI capabilities, surpassing the $100 million cost of GPT-4 with predicted larger size and higher expenses.
Despite newer models being released by OpenAI, Google, and other developers that surpass GPT-4, the world remains in anticipation of the next significant AI leap. Recent progress in AI has been more incremental, focusing on advances in model design and training rather than strictly scaling model size and computation power, which GPT-4 relied on.
Michael Gerstenhaber, Anthropic’s head of product, mentions that their new Claude 3.5 Sonnet model, while larger than its predecessor, gains competence through innovative training methods. The model received targeted feedback to enhance logical reasoning skills.
Anthropic claims that Claude 3.5 Sonnet outperforms leading models from OpenAI, Google, and Facebook in various popular AI benchmarks, including GPQA, MMLU, and HumanEval. While the improvements are modest, they represent a notable progress.
This recent AI advancement may not be groundbreaking, but it shows a rapid pace of development. Anthropic announced their previous model generation just three months ago. Gerstenhaber notes, “If you observe the intelligence rate change, you will realize our swift progress.”
Over a year after GPT-4 triggered new investments in AI, achieving significant leaps in machine intelligence appears challenging. With models like GPT-4 trained on extensive online data, finding new data sources for machine learning algorithms is becoming harder. The cost of significantly increasing model size to boost learning capacity is expected to be in billions. OpenAI’s recent upgrade focus on a more natural and human-like interface with GPT-4o, emphasizing a better user experience over enhanced problem-solving abilities.
