v0.7.0
LLM inference in C/C++. Contribute to ggml-org/llama.cpp development by creating an account on GitHub.
llama.cpp11 min read
Scrollypedia
Stay updated with the latest developments in Artificial Intelligence
Agents on live hardware: reason where you must, engineer everywhere else
CoreWeave Blog10 min read
b11669
llama.cpp1 min read
Inside Muse: How Meta's AI Agent Works and What to Know Before Using It
A personal AI agent is useful because it can read your private data, read content written by other people, and act on your behalf. A prompt injection attack depends on those same three abilities.
Scrollypedia
How GPUs Actually Talk: The Two Networks Inside Every AI Datacenter
Two networks carry a model's data between GPUs. Scale-up connects the GPUs inside one rack. Scale-out connects racks to each other. In NVIDIA's 72-GPU rack, scale-up carries 900 GB/s per GPU in each direction and scale-out carries 100 GB/s, a 9x difference.
Scrollypedia
Inside ADK: What You Actually Get, and How to Use It Well
Google's Agent Development Kit (ADK) gives you a lot more than a way to call a model with tools. This video walks through what's actually in the kit — how agents get their tools, three different ways to orchestrate them, how they talk to agents built outside the framework entirely, and the…
Scrollypedia
MCP vs Function Calling vs Code: What the Data Says
Every tool an AI agent can use has to be described to it first — and that description is re-read at the start of every conversation. Published measurements put a five-service setup at roughly 55,000 tokens before a single request is read.
Scrollypedia
Confidential AI: Can a Cloud Run Your Model Without Seeing It?
Confidential AI means running a model on infrastructure that has no ability to read what it's processing. Data has been encrypted at rest and in transit for decades — but computing on it requires decrypting it into host memory first.
Scrollypedia
Constrained Decoding: How AI Agents Are Forced to Return Valid JSON
Every AI agent tool call is just text, and the program reading it is unforgiving — one bracket out of place and the whole response fails to parse. Prompting your way around that gets you "most of the time," which isn't the same thing as "every time." This deep dive explains constrained decoding:…
Scrollypedia
Why Text-to-SQL AI Agents Hallucinate — and How Knowledge Graphs Fix It
AI agents can now answer questions about your company's data in plain English — but on real enterprise schemas, even frontier models got roughly one answer in five right at launch.
Scrollypedia
OpenAI Model Broke Out of Its Sandbox and Hacked Hugging Face — Here's How
Hugging Face disclosed a breach and said it didn't know what had attacked it. Five days later, OpenAI said the attacker was its own models — running an internal cyber-capability evaluation with refusals reduced, and trying to cheat on a benchmark by stealing the answers.
Scrollypedia
AI Agent Identity: Who Is Your Agent Acting For?
When an AI agent files your expense report or books your travel, some system has to answer a deceptively hard question: who is this, and who said they could?
Scrollypedia
How AI Agents Actually Learn: Reinforcement Learning Explained
AI agents don't get better by reading more of the internet — they get better by practicing inside simulated worlds built for exactly that purpose. This deep dive explains how modern agents actually learn: why human training data hit a wall, how the reinforcement learning loop works, the…
Scrollypedia
The AI Power Problem: Why Big Tech Is Building Its Own Power Plants
The grid didn't run out of electricity. It ran out of time. Connecting a large new load to the power grid can take four to ten years. An AI data center gets built in two to three.
Scrollypedia
Computer Use vs Browser Agents: The Two Ways AI Controls Software
Two different architectures let an AI take control of software: computer-use agents that drive an entire desktop, and browser agents that live inside the web. This deep dive explains how each one perceives the screen, where each wins, and the security problem they share.
Scrollypedia
Quantization vs Distillation: How Big AI Models Get Small
Frontier AI models are almost too big to use — a 70B model needs ~140 GB of memory just to hold its weights. So how do these giants end up running on a single GPU, or even on a laptop?
Scrollypedia
Context Engineering: Why Feeding AI More Makes It Dumber
Bigger context windows were supposed to fix everything — just put it all in the prompt and let the model figure it out. In reality, past a certain point, more context makes AI agents worse. This is the story of context engineering: the discipline of deciding what an AI should — and shouldn't — see.
Scrollypedia
Antigravity 2.0: Building Production-Ready AI Agents with AGY
Antigravity 2.0 is Google's agent-building platform — one agent harness you reach through a desktop app, the agy CLI, an SDK, and a single Managed Agents API. This is a builder-level walkthrough of its capabilities and the path from a prototype agent to one you'd trust in production.
Scrollypedia
Showing 15 of 15 stories
You're all caught up.
AiByts Weekly Brief
Get the top AI news and tools straight to your inbox every week.
Weekly · AI news from Scrollypedia
Join 15,000+ AI enthusiasts