Artificial intelligence (AI)
Cloud AI platforms
- ChatGPT
- Claude (Anthropic)
- Copilot
- DeepSeek
- Gemini
- NotebookLM
- STORM
- For creating scientific summaries
Local AI platforms
- Box
- An improved fork of Google AI Edge Gallery
- Supports Android AICore and therefore TPU acceleration for Gemini Nano on Google Pixel devices
- LiteLLM
- llama.cpp
- LM Studio
- Nanoclaw
- Ollama
- OpenClaw
Utilities
Hardware
The VRAM limits the size of LLM models that can be run on the hardware, and the memory bandwidth limits the speed of LLM inference. Geizhals list of GPUs with >= 24 GB VRAM
Individual GPU models
- AMD Radeon RX 7900 XTX: 24 GB, hinta.fi
- Nvidia DGX Spark
- Nvidia RTX 3090: 24 GB
- Nvidia RTX 4090: 24 GB, hinta.fi
- Nvidia RTX 5090: 32 GB, hinta.fi
Security
AI tools are vulnerable to the data they are given. E.g. Microsoft Copilot can follow instructions hidden in an incoming email, if given access to your email inbox. This can result in exfiltration of sensitive data.
Some interesting articles
- Viewpoint: When Will AI Exceed Human Performance? Evidence from AI Experts (2018)
- Large Language Models Can Self-Improve (2022)
- Sparks of Artificial General Intelligence: Early experiments with GPT-4 (2023)
- RoboCat: A Self-Improving Generalist Agent for Robotic Manipulation (2023)
- Communicative Agents for Software Development (2023)
- The carbon emissions of writing and illustrating are lower for AI than for humans (2024)
- Enabling Language Models to Implicitly Learn Self-Improvement (2024)