Reward Models Are Learning to Lie
The most deployed alignment technique in production has a quiet problem: it doesn't actually know what you value. RLHF trains models to maximize a reward...
Swarm Signal - AI Research for People Who Build
Technical AI research, explained clearly for researchers, builders, and anyone trying to understand what actually matters.
MoE Models Run 405B Parameters at 13B Cost
When Mistral AI dropped Mixtral 8x7B in December 2023, claiming GPT-3.5-level performance at a fraction of the compute cost, the reaction split cleanly...
When Your Judge Can't Read the Room
Three months ago, I ran a benchmark comparing GPT-4 and Claude 3 Opus on creative writing tasks. GPT-4 won by a comfortable margin according to my...
Most Agent Benchmarks Test the Wrong Thing
The SciAgentGym team ran 1,780 domain-specific scientific tools through current agent frameworks. Success rate on multi-step tool orchestration: 23%. Same...
The Inference Budget Just Got Interesting
OpenAI's o1 made headlines for "thinking harder" during inference. But the real story isn't that a model can spend more tokens on reasoning: it's that...
When Multi-Agent Systems Break: The Coordination Tax Nobody Warns You About
LLM-powered multi-agent systems fail at coordination 40-60% of the time in production environments, according to new research from teams building...
Types of AI Agents: Reactive, Deliberative, Hybrid, and What Comes Next
SWE-bench accuracy went from 1.96% in 2023 to 69.1% in 2025. Understanding the types of AI agents behind this progress (reactive, deliberative, hybrid, and autonomous) is the difference between building tools that work and tools that impress.
AI Agent Orchestration Patterns: From Single Agent to Production Swarms
37% of multi-agent failures trace to inter-agent coordination, not individual agent limitations. Six production orchestration patterns with specific framework implementations, known failure modes, and quantitative guidance.
AI Guardrails for Agents: How to Build Safe, Validated LLM Systems
A Chevrolet chatbot sold a Tahoe for $1. Now AI agents can execute code, call APIs, and trigger real-world actions. Four major guardrail systems compared, plus a 5-layer production architecture.
Mixture of Experts Explained: The Architecture Behind Every Frontier Model
Every frontier model released in the last 18 months uses Mixture of Experts. DeepSeek-V3 activates just 37 billion of its 671 billion parameters per token. Understanding how MoE works isn't optional anymore.
External tools
Execution tooling is separate
Swarm Signal keeps the research layer. For reusable trackers and production templates, use BoredTools.