The AI terms that
actually matter.
Clear, technical definitions of 74 key concepts — written by AI engineers, not marketers.
Logit Lens
The Logit Lens is a mechanistic interpretability technique that reads a transformer's mind mid-computation. By projecting the hidden state at each intermediate layer through the model's final layer-norm and unembedding matrix, it reveals the model's evolving 'best guess' for the next token — showing how a prediction is refined layer by layer through the residual stream.
Read definitionLoRA Fine-Tuning
LoRA (Low-Rank Adaptation) is a parameter-efficient fine-tuning technique that freezes a pre-trained model's weights and injects small trainable low-rank matrices into key layers, enabling effective specialisation at a fraction of the compute and memory cost of full fine-tuning.
Read definitionMamba Model
The Mamba model is a groundbreaking AI architecture that achieves state-of-the-art performance across various modalities by utilizing selective state spaces for efficient, linear-time sequence modeling
Read definitionMiniMax M2.7
MiniMax M2.7 is an open-source, self-evolving Mixture-of-Experts AI model that autonomously participates in 30–50% of its own training workflow, featuring 230 billion total parameters, a 200K-token context window, and top-tier performance on agentic and software engineering benchmarks.
Read definitionMixture of Experts (MoE)
Mixture of Experts is a neural network architecture where a learned routing mechanism activates only a small subset of specialised sub-networks (experts) for each input token — delivering the capacity of a much larger model at a fraction of the per-token compute cost.
Read definitionModel Context Protocol (MCP)
Model Context Protocol (MCP) is an open standard introduced by Anthropic that defines a universal interface for connecting AI models to external data sources, tools, and services — eliminating the need for bespoke integrations for every AI-to-system connection.
Read definitionMoshi
Moshi is a full-duplex, real-time spoken dialogue AI developed by Kyutai that can simultaneously listen and speak — enabling natural, interruption-aware conversations with theoretical latency as low as 160ms.
Read definitionMulti Agent System
A Multi-Agent System (MAS) is an AI architecture where multiple autonomous, specialized agents interact, collaborate, or compete to solve complex workflows that are too difficult or broad for a single, monolithic AI model.
Read definitionNVIDIA Parakeet TDT
NVIDIA's Parakeet TDT 1.1B is the fastest open-source ASR model available, achieving an RTFx near 2,000× real-time — processing audio 6.5× faster than Canary-Qwen at the cost of some accuracy.
Read definitionOmniFlatten
OmniFlatten is a full-duplex, end-to-end GPT model for seamless voice conversation that converts multi-stream speech and text into a single flattened token sequence, enabling a standard LLM backbone to handle simultaneous listening and speaking.
Read definitionOmniVoice
A massively multilingual open-source TTS model by Xiaomi supporting 646 languages with zero-shot voice cloning and voice design, running 40× faster than real-time.
Read definitionOpenClaw
OpenClaw is an open-source, local-first autonomous AI agent that runs as a persistent daemon on your own hardware, connecting to any LLM backend and operated through messaging platforms like WhatsApp, Telegram, or Discord.
Read definitionStay ahead of the curve
Weekly newsletter on agentic AI, LLMs, and what we're building at Superteams — straight to your inbox.
Ready to ship AI in production?
We deploy fractional AI teams that deliver production-grade systems in 30–90 days. No fluff, no obligation.