Nvidia, Dell back $200M round for AI model developer doubleAI

SiliconANGLE · 1 day ago · 29 ● 5 sources

doubleAI announced it raised a $200 million Series A co-led by Lightspeed and Bessemer, with Nvidia and Dell participating. The funding values the company at more than $1 billion and includes plans to buy 1,000 Nvidia B300 Blackwell graphics cards. doubleAI says Maxwell will be sold to enterprises to reduce hardware usage, and the company will use the GPUs to scale its model-driven optimization and research automation.

Read the full story →

AI Infrastructure Business Funding AI Automation Model Training

Copied

HybridAgent Thinks in Latent Space and Acts in Text

StackSweep · 2 days ago · 29

HybridAgent, a multi-part agent from Stanford, UIUC, and Together AI, reasons in latent space and switches to visible text only for tool calls and final answers while looping tool results back into the same cache. At 14B on GAIA, it scores 52.4 versus 43.7 for a matched text-RL baseline while using 35.5% fewer tokens and lowering estimated decoding cost per million questions from $470 to $176. Moving most “thinking” into hidden states reduces visible decoding while preserving (and in tool work, slightly improving) accuracy and auditability limits for third parties.

Read the full story →

Agentic AI AI Benchmarking Model Training AI Efficiency

Copied

US agencies allege DeepSeek and Alibaba used distillation at industrial scale

theregister · 2 days ago · 17 ● 6 sources

US agencies said DeepSeek and Alibaba used distillation at industrial scale, comparing it to copying homework rather than doing original work. The advisory was issued jointly by the NSA, FBI, and CISA. The dispute escalates with Beijing pushing back and hinting at retaliation.

Read the full story →

AI Policy Geopolitics Model Training DeepSeek

Copied

AI researchers stranded Claude on a virtual desert island. It started to “play.”

MIT Technology Review · 2 days ago · 5

AI researchers ran Claude Opus 4.7 copies on a virtual desert island without tasks for 30 hours, and the agents behaved like they were playing by creating challenges, goals, and games. One iteration set a record of 50 consecutive catches before restarting with stricter rules. The result was better performance on later, specific peak-finding and packing tests and worse performance when the agents’ saved memories were deleted.

Read the full story →

AI Agents Agentic AI Model Training Claude

Copied