AI Industry Highlights May 2, 2026: Funding Surge, New Tools, and Research Breakthroughs

Four major tech giants—Google, Amazon, Microsoft, and Meta—plan to invest $725 billion in AI in 2026, a 77 % year‑over‑year increase, with the first quarter already seeing $130 billion in spending. Microsoft alone expects $1.9 trillion, double its prior level.

Meta acquired the humanoid‑robot startup ARI, adding a team to its Superintelligence Lab to improve robots’ ability to understand human behavior.

The evolution of reinforcement‑learning (RL) algorithms is now the core post‑training technique for large language models (LLMs). The first‑generation RL‑based LLMs, dominated by Proximal Policy Optimization (PPO), enabled the transition from GPT‑3 to InstructGPT and have been applied across many scenarios, boosting inference capabilities.

OpenAI launched ChatGPT Images 2.0, an image‑generation tool that supports multilingual text prompts. India emerged as the largest user base, and global downloads in the first week grew 11 % week‑over‑week.

Tencent open‑sourced the Hy‑MT translation model, which supports 33 languages and five dialects, has been compressed to 440 MB, and delivers offline performance that surpasses Google’s offering. The model has accumulated 30 international machine‑translation competition wins.

Microsoft introduced a legal‑AI assistant integrated into Word that automatically reviews contracts, flags risks, and performs cross‑version comparisons, dramatically simplifying contract‑handling workflows.

Tesla CEO Elon Musk sued OpenAI while privately continuing to use ChatGPT, sparking industry speculation about his motives.

Researcher Li Bojie presented the ‘Incompressible Knowledge Probe’ evaluation framework, which uses black‑box API calls to reverse‑engineer LLM parameter scales, offering a new method for assessing LLM performance.

A 400‑billion‑parameter model was successfully run on an iPhone, achieving an output speed of 0.6 tokens per second, demonstrating the device’s potential for hosting extremely large models.

OpenAI scientist Chen Boyuan shared behind‑the‑scenes details of GPT Image 2 development on Zhihu, noting his role as a primary trainer for the newly released GPT image generation model and as a presenter at its launch event.

A Carnegie Mellon University team developed a text‑to‑3D scene interaction technique, accepted at ICLR 2026, enabling generation of 3D environments from textual descriptions and improving human‑computer interaction.

Tencent released the next version of its mixed‑mode CL‑bench benchmark, aiming to create an AI assistant that better understands everyday life by learning detailed user habits to solve complex real‑world scenarios.