Articles by dbreunig
304

Why is Claude an Electron app? (dbreunig.com)

1

Analyzing How System Prompts Define Agent Behavior (dbreunig.com)

2

The Potential of RLMs (dbreunig.com)

3

The Rise (and Limits) of Spec Driven Development (dbreunig.com)

2

A OSS Library with No Code, Only Specs (dbreunig.com)

2

2025 in Review: Jagged Intelligence Becomes a Fault Line (dbreunig.com)

1

Why I write (and you should too) (dbreunig.com)

1

Applied AI in 2025: From 'Naked' Model Calls to Tool Use Environment Calls (dbreunig.com)

2

Enterprise Agents Have a Reliability Problem (dbreunig.com)

2

Don't Fight the Weights: Learn to Spot Contexts That Go Against Training (dbreunig.com)

3

About That MIT Report: Enterprise AI Looks Bleak, but Employee AI Looks Bright (dbreunig.com)

3

AI Companies School Like Fish to the New Use Case (dbreunig.com)

1

Monolingual speech recognition models beat multilingual models ~30x bigger (arxiv.org)

1

Are Chatbots Incompatible with Advertising? (dbreunig.com)

44

The AI Job Title Decoder Ring (dbreunig.com)

2

How Will Product Management Keep Up with AI-Accelerated Coding? (dbreunig.com)

1

GPT-OSS 20B delivers GPT-4.1 scores at 5x the speed (dbreunig.com)

10

FLUX.1-Krea and the Rise of Opinionated Models (dbreunig.com)

60

Does the Bitter Lesson Have Limits? (dbreunig.com)

1

Moonshot Used RL for Qualitative Tasks to Write Better (dbreunig.com)

2

Kimi 2 Became One of the Best Tool-Using Models (dbreunig.com)

1

Rewriting Training Data Improved Kimi 2's Performance (dbreunig.com)

4

Task Delegation by Experts Is the AI Metric That Matters (dbreunig.com)

2

Ask HN: How are startups dealing with key talent poaching?

1

Techniques for avoiding common long context fails (dbreunig.com)

2

The Ways Long Contexts Fail (dbreunig.com)

2

An Agentic Case Study: The Messy Work of Building a Gemini Agent to Play Pokémon (dbreunig.com)

37

The drawbridges come up: the dream of a interconnected context ecosystem is over (dbreunig.com)

3

Let the Model Write the Prompt: Using DSPy to Decouple Your Task from the LLM (dbreunig.com)

36

Comparing Claude System Prompts Reveal Anthropic's Priorities (dbreunig.com)

2

Using 'Slop Forensics' to Determine Model Lineage (dbreunig.com)

1

Chatbots Trigger Guardrails Differently Based on Your Age, Gender, and Fandom (dbreunig.com)

4

Breaking Down Claude's 26k+ Token System Prompt (dbreunig.com)

114

DuckDB is probably the most important geospatial software of the last decade (dbreunig.com)

1

Explaining Reasoning Models: How they work, how they're built, & why they matter (dbreunig.com)

1

AI-Powered Apps Are Changing the Dynamic Between Domain Experts and Developers (dbreunig.com)

1

AI Chatbots Are Like Observational Comics (dbreunig.com)

1

ChatGPT fixes your DALL-E prompt if it's bad, but lets it be if it's good (dbreunig.com)

2

Controlling FFmpeg with natural language from the command line (dbreunig.com)

84

A 16TB Mirror of Data.gov on Source.Coop (source.coop)

4

A Gentle Guide to Running a Local LLM, for Complete Beginners (dbreunig.com)

1

DeepSeek Is a Rorschach Test Revealing Each's AI Hopes and Fears (dbreunig.com)

4

The Rise of the Term "Authentic Data" Illustrates AI Anxiety (dbreunig.com)

1

Your Eval Is More Important Than the Model: Building an Example Eval (dbreunig.com)

1

The "Cyren" Song: Meta's AI profiles, Character.ai, and other fake humans (dbreunig.com)

1

A Primer on Synthetic Data: How It's Helping and Shaping LLMs (dbreunig.com)

1

DSPy Does the Prompting for You (dbreunig.com)

2

LLMs Hit the Wall: History Reveals How We Went Fast and Why We'll Go Slow (dbreunig.com)

2

Detecting DuckDB Requests in Your Webapp to Route Them to Remote DBs (dbreunig.com)

1

Surging crypto prices dampened interest in AI in 2020. Will they do it again? (dbreunig.com)

2

Transitioning from the attention era to the automation era (dbreunig.com)

3

It's the UX, Not the LLM: ChatGPT and Perplexity Can Ship a UI Google Can't (dbreunig.com)

1

Generating Descriptive Weather Reports with LLMs (dbreunig.com)

1

AI Use Cases: Gods, Interns, & Cogs (dbreunig.com)

162

AI PCs Aren't Good at AI: The CPU Beats the NPU (github.com/usefulsensors)

1

The Military Implications of Big vs. Small AI (defenseone.com)

13

Wikidata as a Giant Crosswalk File (dbreunig.com)

1

Conflating Overture Places Using DuckDB, Ollama, Embeddings, and More (dbreunig.com)

1

What Building Blocks Remains After an AI Collapse? (dbreunig.com)

2

When is a Minicar as Dangerous as a 3-Ton Truck? (IIHS death-rate data followup) (dbreunig.com)

8

Ask HN: Who Is OpenAI's O1 For?

2

We Need an Ari Emanuel for Small OSS: An Agency for Developers (dbreunig.com)

1

Confronting Your AI Accuser: AI faces the same challenge as breathalyzers (dbreunig.com)

1

Stewart Brand on Finding Where the Future Is Being Made (dbreunig.com)

2

How to Build a Bigger AI Bubble (dbreunig.com)

1

Towards Standardizing Place: Why I'm Excited for Overture's GERS (dbreunig.com)

2

Show HN: Easily map CSV data with lat/lon to H3 for enrichment or aggregation (dbreunig.com)

1

AI Is Smaller, Cheaper, and Faster. (But Not Smarter) (dbreunig.com)

1

Be Better, Not Smaller: Why Most AI Products Are Making a Familiar Mistake (dbreunig.com)

1

Joining Census and Overture IDs with DuckDB; the Importance of Persistent IDs (dbreunig.com)

3

Start-Up Domain Name Trends Over the Years (dbreunig.com)

18

Beneath the AGI and Foundational Model Hype, Sober AI Is the Enterprise Default (dbreunig.com)

3

A Plea for Sober AI (dbreunig.com)

1

Training Models with Customer Data Is the New Network Effect (2016) (dbreunig.com)

110

Show HN: I built a tool for repeatable checklists (steplist.app)

11

A POI database in one line (dbreunig.com)

2

The Rise of Audio AR (dbreunig.com)

1

Paying the AI Premium (dbreunig.com)

3

Potential Alien Signal Was Just a Truck (jhu.edu)

1

In Pursuit of Quiet AI Features and Better UX (dbreunig.com)

1

Déjà Connu: Using Copilot Feedback Quality as a Signal (dbreunig.com)

17

A CCPA Request to OpenAI Took 101 Days and Raises More Questions Than Answers (dbreunig.com)

1

A Foundation Model for Every Culture (dbreunig.com)

2

2024 Will Be a Big Year for Small Models (dbreunig.com)

1

The Tyranny of the Blank Textbox: OpenAI Has Saturated Their Initial Market (dbreunig.com)

3

Extreme Compression with AI: Fitting a 45 Minute Podcast into 40kbs (dbreunig.com)

34

The AI Market Is Firming Up Fast (dbreunig.com)

1

Finding Bathroom Faucets Using Embeddings (dbreunig.com)

1

The 5 Blows to the Human Ego (dbreunig.com)

1

Manipulating AI: Prompt Injection, Trap Tokens, & Post-Training Suggestion (dbreunig.com)

1

Considering 5 AI Privacy Scenarios: From PII Leakage to AI Bombing (dbreunig.com)

2

AI: The Platypus in the Room (dbreunig.com)

2

Bake AI Safewords into Models to Prevent Fraud (dbreunig.com)

1

Practical AI Guidelines and Use Cases for App Developers (dbreunig.com)

2

AI Lies, Privacy, & OpenAI (dbreunig.com)

1

AI Needs Safewords (dbreunig.com)

4

ABS-B Exchange has been suspended from Twitter (twitter.com/rmac18)