DeepAI CEO Kevin Baragona

AsianFin -- The competition in the field of large models has shifted from computing power to efficiency, said Kevin Beregona, the founder and CEO of DeepAI, at the recently-concluded T-EDGE Conference in Hilton Hotel in Beijing.

California-based generative AI company DeepAI is making waves in the U.S. as a leading provider of AI tools, including chatbots, image and video generators.

Beregona said that 2024 marked a "golden age of AI," with rapid progress happening every month, particularly in the generative AI space. DeepAI, known for being one of the first companies to offer an image generator, has seen dramatic advancements since its launch. When the company first introduced its image generator, the results were often blurry and incoherent. Today, however, the same prompts produce highly detailed and visually captivating images, thanks to deep learning and a more refined understanding of the meaning behind words and concepts.

The company's capabilities have expanded beyond static images, with DeepAI now offering video generators, demonstrating a significant leap in what AI can accomplish in real-time. It's exciting to still be creating and pushing boundaries, Beregona remarked.

Reflecting on the early days of AI, Beregona mentioned the difficulties facing the industry, including the decades-long "AI winter," when there was minimal progress and little interest in neural networks. However, things began to shift around 2020, coinciding with the rise of powerful models like GPT-3, which reshaped the AI landscape and convinced more people of the potential of deep learning.

The biggest catalyst for the current AI boom, according to Beregona, is the drastic reduction in the cost of compute power. He pointed to Nvidia’s success in providing affordable matrix multiplication as a key factor in enabling AI’s rapid expansion. Compute is a commodity now. It’s about finding the cheapest option, Beregona explained.

Despite the impressive advancements in AI, the technology has also sparked political and ethical debates, with increasing concerns over its potential dangers. Beregona pointed out the paradox within the industry: while nations and corporations vie for AI dominance, there’s a simultaneous push to limit its power due to the perceived risks. He pointed out that much of the so-called “secret sauce” behind successful AI models isn’t really a mystery—it's simply about coupling massive computational resources with vast datasets.

As the industry evolves, questions around the future of large language models (LLMs) have arisen. Some industry experts have speculated that progress on LLMs may be slowing. However, Beregona is optimistic, noting that while data might be harder to come by, there's still room for innovation. He explained that researchers are now exploring ways to enhance reasoning skills in AI, which he sees as the next frontier.

Increasing the amount of inference time is one way to make AI smarter, and recent experiments show a promising correlation between reasoning accuracy and extended inference time, Beregona noted.

The path forward could involve specialized reasoning models that perform better than general-purpose LLMs, he added.

Looking ahead, Beregona remains enthusiastic about the potential of AI. He acknowledged that, while the human brain may still outperform the best GPUs in terms of efficiency, ongoing advancements in hardware and data processing mean that the capabilities of AI could continue to expand exponentially.

Compute will keep getting cheaper and more efficient, and the potential is almost unimaginable, he said.

As for the future of DeepAI, Beregona is excited to continue innovating and offering new products to their growing base of users. The company’s ability to keep pushing the envelope in generative AI is a testament to the rapidly evolving capabilities of AI, and DeepAI is positioned to remain at the forefront of the revolution.