- January - 9 entries, 53 links, 29 quotes
- 2nd: Ending a year long posting streak
- 2nd: I still don't think companies serve you ads based on spying through your microphone
- 4th: Weeknotes: Starting 2025 a little slow
- 10th: My AI/LLM predictions for the next 1, 3 and 6 years, for Oxide and Friends
- 20th: DeepSeek-R1 and exploring DeepSeek-R1-Distill-Llama-8B
- 22nd: Six short video demos of LLM and Datasette projects
- 24th: Anthropic's new Citations API
- 24th: A selfish personal argument for releasing code as Open Source
- 31st: OpenAI o3-mini, now available in LLM
- February - 7 entries, 48 links, 16 quotes
- 7th: Using pip to install a Large Language Model that's under 100MB
- 13th: URL-addressable Pyodide Python environments
- 15th: Run LLMs on macOS using llm-mlx and Apple's MLX framework
- 17th: LLM 0.22, the annotated release notes
- 25th: Claude 3.7 Sonnet, extended thinking and long output, llm-anthropic 0.14
- 27th: Initial impressions of GPT-4.5
- 28th: Structured data extraction from unstructured content using LLM schemas
- March - 11 entries, 59 links, 18 quotes, 1 note
- 2nd: Hallucinations in code are the least dangerous form of LLM mistakes
- 2nd: Notes from my Accessibility and Gen AI podcast appearance
- 4th: I built an automaton called Squadron
- 8th: What's new in the world of LLMs, for NICAR 2025
- 11th: Here's how I use LLMs to help me write code
- 12th: Notes on Google's Gemma 3
- 13th: Adding AI-generated descriptions to my tools collection
- 19th: Not all AI-assisted programming is vibe coding (but vibe coding rocks)
- 20th: Calling a wrap on my weeknotes
- 20th: New audio models from OpenAI, but how much can we rely on them?
- 25th: Putting Gemini 2.5 Pro through its paces
- April - 12 entries, 50 links, 26 quotes, 12 notes
- 5th: Initial impressions of Llama 4
- 7th: Long context support in LLM 0.24 using fragments and template plugins
- 9th: Model Context Protocol has prompt injection security problems
- 11th: CaMeL offers a promising new direction for mitigating prompt injection attacks
- 14th: GPT-4.1: Three new million token input models from OpenAI, including their cheapest model yet
- 18th: Image segmentation using Gemini 2.5
- 19th: Maybe Meta's Llama claims to be open source because of the EU AI act
- 21st: AI assisted search-based research actually works now
- 24th: Exploring Promptfoo via Dave Guarino's SNAP evals
- 26th: Watching o3 guess a photo's location is surreal, dystopian and wildly entertaining
- 29th: Qwen 3 offers a case study in how to effectively release a model
- 30th: Understanding the recent criticism of the Chatbot Arena
- May - 11 entries, 59 links, 14 quotes, 21 notes
- 1st: Two publishers and three authors fail to understand what "vibe coding" means
- 5th: Feed a video to a vision LLM as a sequence of JPEG frames on the CLI (also LLM 0.25)
- 6th: Saying "hi" to Microsoft's Phi-4-reasoning
- 10th: Trying out llama.cpp's new vision support
- 15th: Building software on top of Large Language Models
- 21st: I really don't like ChatGPT's new memory dossier
- 22nd: Live blog: Claude 4 launch at Code with Claude
- 25th: Highlights from the Claude 4 system prompt
- 27th: Large Language Models can run tools in your terminal with LLM 0.26
- 30th: Talking AI and jobs with Natasha Zouves for News Nation
- 31st: How often do LLMs snitch? Recreating Theo's SnitchBench with LLM
- June - 8 entries, 39 links, 17 quotes, 12 notes
- 3rd: Tips on prompting ChatGPT for UK technology secretary Peter Kyle
- 6th: The last six months in LLMs, illustrated by pelicans on bicycles
- 7th: Comma v0.1 1T and 2T - 7B LLMs trained on openly licensed text
- 13th: Design Patterns for Securing LLM Agents against Prompt Injections
- 15th: An Introduction to Google’s Approach to AI Agent Security
- 16th: The lethal trifecta for AI agents: private data, untrusted content, and external communication
- 17th: Trying out the new Gemini 2.5 model family
- 23rd: Phoenix.new is Fly's entry into the prompt-driven app development space