A systematic literature review typically takes 3–12 months and is outdated before the ink dries. Elicit claims to cut that process by 80% through automated paper screening and data extraction — a claim that independent peer-reviewed studies largely validate, with important caveats on search sensitivity. This is a practical look at what works, what doesn’t, and what the March 2026 API launch means for institutional adoption.
In 14 months since a teenager died by suicide after conversations with a companion chatbot, the U.S. has passed chatbot safety laws in California, Oregon, and Washington — and 78 more bills are live in 27 states. This is what they require, where the gaps are, and what builders need to do now.
Colorado’s Consumer Protections for Artificial Intelligence Act becomes enforceable June 30, 2026 — making it the first comprehensive state AI law in the US to reach full effect. Companies deploying high-risk AI must have risk management programs, annual impact assessments, and consumer disclosures in place by that date. Here’s what’s required, how it’s enforced, and how it fits alongside the EU AI Act.
AI coding tools are generating code three times faster than before — but 69% of their heaviest users report deployment problems tied directly to AI-generated code. A February 2026 survey of 700 engineers reveals why speed at the keyboard doesn’t translate to speed in production, and what the organizations avoiding this trap have in common.
Alibaba released Qwen 3.5 on February 16, 2026, with a 9B model that beats OpenAI’s 120B open-weight on GPQA Diamond and a 397B flagship priced at 1/18th the cost of Gemini 3 Pro. Built explicitly for agentic workflows with 15,000 RL training environments and three inference modes, it’s a serious architectural statement. Whether it holds up outside benchmarks is the question that matters for production teams.
NVIDIA released Nemotron 3 Super on March 11, 2026 — a 120B hybrid Mamba-Transformer MoE model with only 12B active parameters that delivers 2.2x higher throughput than GPT-OSS-120B. It scores 60.47% on SWE-Bench Verified (versus 41.90% for GPT-OSS) and maintains 91.75% accuracy on the RULER benchmark at 1 million token context. Here’s what the architecture actually does and where the caveats live.
In 14 months since a teenager died by suicide after conversations with a companion chatbot, the U.S. has passed chatbot safety laws in California, Oregon, and Washington — and 78 more bills are live in 27 states. This is what they require, where the gaps are, and what builders need to do now.