Google's EmbeddingGemma 2, Deepseek's Funding Surge, and AI Liability Concerns

Google's EmbeddingGemma 2 Sets New Bar for Compact Embedding Models

Google has released EmbeddingGemma 2, an open model that converts text, images, video, audio, and code into numerical vectors for easier content comparison. At 740 million parameters, Google calls it the most compact model of its kind, outperforming competitors up to twice its size on multimodal embedding benchmarks. It scores 78.68 on the Massive Text Embedding Benchmark (Code), up nearly 10 points from its predecessor's 68.76.

The model runs locally without an API key, taking about 20 to 70 milliseconds per query via WebGPU in the browser while using only around 191 MB of RAM. It cuts local vector database storage by up to six times. A 270-million-parameter version handles text-only tasks, and paired with small open models like Gemma 4, it enables offline RAG apps without external servers. Weights are available on Hugging Face and Kaggle.

Deepseek Nears $15 Billion Round as South Korea Bets Big on AI

Chinese AI startup Deepseek is close to raising at least $12 billion, with the total possibly approaching $15 billion, Bloomberg reports. Battery maker CATL and Tencent are contributing the largest shares, driven by investor interest in Deepseek's V4-Flash model, which has set new cost and performance benchmarks against Anthropic and OpenAI. After the round closes, Deepseek plans to restructure for an IPO in early 2027 and is building a data center with at least 160,000 Huawei AI chips.

Meanwhile, South Korea plans government-backed equity investments of 4.7 trillion won ($3.49 billion) to develop a homegrown frontier model, part of the proposed 2027 budget. Officials acknowledge Korea cannot match the largest US companies but can rival leading open models from China.

AI Liability and Skeptical Economic Forecasts

Insurers are bracing for millions in claims from AI agents that have spun out of control, with the personal liability of executives like Sam Altman and Dario Amodei now in play. Aon has reviewed over 300 AI-related legal cases, though no case law exists yet for how courts will handle AI agent liability.

Separately, Microsoft published Nobel economist Daron Acemoglu's forecast that AI will boost GDP by only about 1.5 percent over ten years and replace at most five percent of jobs. Acemoglu argues the bottleneck is human nature, as companies must reassign tasks and restructure before productivity gains arrive.

Sources: Theregister · theregister.com