AI Tech Daily: Key AI Industry Highlights for April 30 2026

Microsoft reported third‑quarter revenue of $82.9 billion, up 18% year‑over‑year, with AI‑related revenue growing 123% to $37 billion, making AI the core growth engine and demonstrating the conversion of AI technology into a strong profit source.

Alec Radford, often called the "father of GPT," tackled the "Hassabis problem" by attempting to train a model that stopped learning in 1911 to derive the 1915 theory of General Relativity, a line of research that could have important implications for AGI development.

DeepSeek announced a multimodal technology that users can experience on both web and app platforms; the company also published the related code and resources on GitHub.

Ant Group's zkDTVM set a new record in the EthProofs benchmark, achieving a block‑proof time of 3.7 seconds and ranking first worldwide, thereby improving the efficiency of blockchain transaction verification.

An Anthropic researcher uncovered a heap‑buffer‑overflow vulnerability in the Linux kernel NFS driver that had persisted for 23 years, confirming five kernel bugs. Linux maintainers noted that AI‑generated vulnerability reports have become effective, with 5–10 valid reports received daily.

OpenAI's new GPT‑5.5 model forbids mentioning "goblin" and other fantasy creatures in system prompts to prevent specific hallucinations. The Codex CLI inadvertently exposed the underlying instruction set, which spans more than 3,500 words.

Stripe introduced 288 new features aimed at building AI economic infrastructure, including an agent‑driven wallet powered by Link, to foster innovation in digital payments.

In an experiment, the "father of GPT" trained an AI model using only pre‑1915 data, and the model surprisingly learned to write Python code.

ChatGPT uninstall rates surged by 413%, while Claude downloads doubled, highlighting growing challenges for leading AI providers.

LLM agents are advancing toward multi‑tool orchestration; the primary challenges involve stable coordination of tools, state management, adapting to changes, and achieving reliable automation of complex tasks.