Reddit debate: could frontier models counter AI-enabled catastrophes?

Nous Research raised a $90M Series B led by Robot Ventures at a $1.5B valuation, bringing total funding to $158M. Its open-source Hermes agent has been cloned over 24M times and drives roughly 2.5% of global AI token usage.

Essay argues the current approach — use operational excellence to align today's AI, then have that AI automate alignment research — is close to the worst possible plan. Notes labs are "terrified" they cannot execute even step one and are trying to pace the frontier, with government "actively getting in the way."

A three-month randomized trial with practicing patent attorneys found AI raised work quality across the board, but senior lawyers using AI for 90 days showed stronger judgment while juniors showed no average skill gain, their scores splitting into more strong and more low scores.

The tool unlocks 4 additional SMs for free on the CMP170hx, which offers 1.5TB/s memory bandwidth versus 273 GB/s on a 64GB DGX Spark, at less than half the price. The release follows the 128GB DGX Spark's price rise from $4.7k to $7k and a new 64GB version at $5k.

The API serves fresh, Markdown-formatted docs for Google Cloud, Firebase, and Android via semantic and keyword search, chunking, and grounded Q&A. A gcloud CLI surface ships pre-installed in Cloud Shell across Linux, macOS, and Windows.

Talk describes a "software factory": a system of agents that takes input signals like bug reports and user feedback and turns them into production code. Framed as engineers shifting from typing code to governing agents.