Llms is a topic tracked in our intelligence system with 5 linked articles.
An arXiv preprint uses psycholinguistic corpus analysis to show that nine frontier and near-frontier LLMs produce text that is measurably harder for humans to process than human writing, hypothesizing that models are being optimized for information density (math/code/reasoning) rather than successful communication.
MIT researchers stranded 13 AI agents in a task-free virtual island simulation and observed emergent behaviors — inventing games, social rituals like mourning a lost ring — and, notably, improved job performance afterward.
The Verge argues the frontier AI race has snapped back open after Google and Meta unveiled models they claim rival OpenAI's and Anthropic's best, ending the pair's brief lead in a matter of days.
Domyn, the Italian LLM maker, has raised about $1.1 billion in a debt-heavy round (roughly 90% debt, 10% equity) over the past two years, and the CEO claims ARR is just a few quarters away from $1 billion.
AMD reports a 30% productivity uplift from AI across the SDLC, with AI-generated code approaching 50% of the codebase and some components at 80%, plus a RSX issue-resolution rate rising from 6% to 75% as evidence for autonomous AI agent swarms reshaping software engineering.
The article surveys the race to define the next big advances in LLMs, highlighting the role of startups and shifts in AI academia.
Subscribe for real-time topic updates and unlimited access to our intelligence platform.