Weekly AI Digest – Issue 3: Agentic AI, Riemann Hypothesis Rumors, AI Search Trends, and Real‑time Voice Interaction

Market and Voices

Andrew Ng: The Next Emerging Direction for LLMs Is Agentic AI

On November 14, OpenAI was reported to launch an Agent "Operator" in January, allowing users to issue simple commands for the Agent to code applications, order food, and make plans.

Netizens reacted that the news came too late, noting that competitors such as Anthropic, Google (rumored "Jarvis"), Baidu, and Zhipu have already introduced similar Agent features.

The consensus is that Agents will become a major focus of competition next year.

In a recent analysis on deeplearning.ai, former Stanford AI Lab director Andrew Ng argued that LLM evolution is shifting from optimizing consumer Q&A to supporting agentic workflows, including tool use, computer operations, and multi‑agent collaboration.

He illustrated this with a weather‑forecast example that requires calling an external API, noting that GPT‑4’s native function‑calling makes such tasks more stable and seamless, and that major LLM providers are likely to embed these capabilities directly.

Opinion: LLMs are moving toward supporting agentic workflows, which will mature and improve efficiency in existing processes; timing and business alignment are key to leveraging these advances.

The Riemann Hypothesis Apparently "Proved" by Grok‑3?

The Riemann hypothesis, a famous conjecture about the distribution of prime numbers, was humorously claimed to be proved by xAI engineer Hieu Pham on social media, but he later clarified it was a joke.

Nevertheless, many speculate that AI could eventually solve such millennium‑level problems because mathematical results can be verified cheaply and quickly, and AI research is intensifying in this area, with some predicting AI will become a "super mathematician" by the end of 2026.

Opinion: Although AI has not solved the Riemann hypothesis, its potential in mathematics and broader scientific research is already significant, opening new possibilities for future breakthroughs.

Industry Solutions

When Will AI Search Replace Traditional Search?

Gartner predicts that by 2026 traditional search traffic will drop 26%, while AI‑driven search is gaining strong investment from major players.

Perplexity, an AI‑search unicorn, saw an eight‑fold increase in query volume over the past year, generating roughly $50 million in annual revenue.

AI search leverages LLMs to understand query semantics and retrieve hidden information, offering higher recall than conventional search, though user experience challenges remain, especially when results deviate from expectations.

Opinion: Traditional and AI search each have strengths; while AI search is not yet ready to fully replace classic engines, it can complement them, especially for difficult queries, and enterprises should explore AI‑driven data mining.

Valuable Technologies

Exploring Voice Interaction Solutions

OpenAI recently released a Realtime API that enables developers to build custom conversational agents with sub‑500 ms latency, delivering near‑seamless human‑machine dialogue.

Key advantages of the API include:

Elimination of cumbersome text conversion, drastically reducing latency for smoother voice output.

Natural speech with expressive tones, laughter, and whispers to match conversational context.

Responsive interruption handling for uninterrupted dialogue flow.

Robust noise‑cancellation that maintains clarity even in noisy environments.

In contrast, domestic solutions typically chain separate ASR, LLM, and TTS components, leading to higher latency and less fluid interactions compared to OpenAI’s end‑to‑end streaming architecture.

For further reading, see the linked articles throughout the newsletter.