Sean Goedecke's personal blog
People are pretty unhappy about Anthropic’s recent announcement that they’re planning to include a hidden watermark in Claude model outputs. Will this lead to a mass exodus from Anthropic models? Will the introduction of watermarking be a meaningful change for users? No. AI text watermarking is not a big deal. It doesn’t make the text worse, it doesn’t make AI outputs more detectable in practice,…
Every time a new open-weight AI model is released, people say that local models are the future. Why spend billions of dollars building out datacenters when everyone will just be able to run AI models on their laptops or phones? I think this idea is doomed. No matter how strong open-weight models get, most inference will always happen in AI datacenters. Local models are too weak to be widely used…
Everyone knows that AI sycophancy is when the model tells you how smart you are. Wow, you’re absolutely right. That’s not just a new idea — it’s genuinely groundbreaking. You’re a very special user. Easy to spot, isn’t it? The discussion around AI sycophancy peaked last year, when the “#keep4o” movement was protesting the removal of OpenAI’s most sycophantic model (GPT-4o), and many people were…
This will be kind of an unusual post. I got a recent email about my writing that I thought was such a good articulation of one common criticism that I’d like to share it (and my response) in full. Here’s the email, from William Murray 1 : Hey Sean, I have enjoyed your writing but your recent essays frustrate me. You say that getting paid for deep thinking in software is coming to an end. You even…
Imagine you’re the guest on some kind of frenetic, software-engineering-themed game show. The host is constantly flipping over new cards with questions that you have to answer as fast as possible: Is this adjustment to the database schema right? Do these bits of data look plausible? Do these five paragraphs of text describe an actual series of manual tests that took place? Does this suggested…
Engineers often complain that visibility should be their manager’s job. In other words, they think engineers should be able to focus on the code, while their manager figures out who’s doing well and rewards them. This attitude is an extension of the “school fantasy”: the idea that your workplace should operate by the same rules as your school or university. After all, you didn’t have to worry…
One recent development in AI is its ability to solve some long-standing problems in mathematics. In 2024 and 2025, this was a trickle: once or twice a year somebody would say that an LLM came up with a proof, and then everyone would argue over whether that counted as “real” mathematical innovation. In 2026, it’s a flood. Almost every day I see some new LLM-produced mathematical result. Prompt…
When you’re on fire, problems are transparent: they’re solved simply by the act of looking at them. Even complicated layers of multiple problems can simply be glanced through like stacked panes of glass. But nobody can work that way all the time. This is a common pitfall for smart engineers. Accustomed to being able to immediately intuit the solution, the first time they run into a problem they…
In the 2010s, if you had technical gaps (say, you couldn’t write CSS), you had to either rely on a skilled colleague or just hope that the answer to your exact problem was out there on the internet. Today, everyone can write sort-of-okay CSS by delegating the task to an LLM. LLMs make everybody into a generalist. Because of this, lots of people don’t think there’s any skill involved in working…
Before large language models, people who worried about AI safety often talked about the “boxing problem”. It goes like this . Suppose some genius figures out artificial intelligence in a late-night coding session on their laptop. Because they’re a genius, they’re smart enough to disable internet access on the laptop before turning it on. In order to escape to the outside world (and begin…
Here’s the big idea in Keith Johnstone’s book Impro : Children are naturally creative, but are violently formed into repressed adults by Western culture and education The process of becoming more creative and expressive is largely a process of unlearning these habits of repression Improv — improvisational comedy — is thus not just the skeleton key for learning to act, but for unlocking a more…
The anonymous blogger Gwern recently completed a thirteen thousand word post called Human-like Neural Nets by Catapulting , in which he offers a theory about why LLMs don’t possess truly flexible human-like intelligence, and how we might train LLMs that do. Theories like this are entirely unremarkable: every crank researcher on the internet has a theory about how to crack AI. But Gwern is…
Software engineers are often told to “start playing politics”, but most engineers have no idea what that means. Their reference point for “playing politics” comes from fiction like Game of Thrones. Are they supposed to raise an army and depose the CEO, or poison each other at team lunch? Should they book Zoom calls with each other and plot schemes? All of that is obviously ridiculous. In terms of…
As a software engineer, how well do you have to understand your own codebase? My guess is that people who work on small codebases with low-turnover teams (say, Redis or games like The Witness ) would say “obviously you have to understand it completely, otherwise you can’t do good work”. I’d also guess that people who work on large codebases with high-turnover teams (say, the Google web search…
Every post I publish represents at least two things I’ve learned: the thing that prompted me to write the post, and the thing I learned in the course of writing it. If I don’t learn anything new while I’m writing, it’s not interesting enough to publish. Typically I learn way more than two things. For instance, in my o3 geoguessr post, I started out with the idea that most AI prompts probably don’t…
The European Union AI Act is Europe’s attempt to comprehensively regulate AI usage. A big part of that is the requirement that AI-generated content be identifiable: either tagged with a watermark or with what the Act calls “digitally signed metadata” 1 . Since all this becomes enforceable in a month, it’s worth figuring out if it makes any sense. I recently discussed AI watermarking at length in…
The European Union AI Act will begin to be enforceable in August 2026, one month from now 1 . One of the biggest new requirements is Article 50 , which requires all AI outputs to be “detectable as artificially generated”. In other words, if LLM providers want to do business in the EU, they will have to apply a watermark to their outputs 2 : some hidden signature that can be used to identify AI…
Stating the obvious is surprisingly useful . Most of your knowledge lives below the threshold of conscious awareness, so it’s possible for a piece of writing to remind you of what you already know. It’s common to know you don’t like something without being quite sure why, and reading an obvious statement (such as “accuracy matters , even when you agree with the broad strokes”) can help clarify why…
Many people claim that AI inference is unprofitable to serve, and thus must be subsidized by an ocean of dumb money from investors who believe that some future AI model will come to dominate the world economy. When that dumb money goes away, so will AI products. According to this view, LLMs are just inherently too expensive (in terms of money, power, and water) to be used in consumer products. In…
People who think current AI use is unsustainable often rely on the claim that inference GPUs only last “three years at the most” under load 1 . The idea here is that once the AI bubble money drains away, current infrastructure will rapidly become obsolete, and there won’t be enough money floating around to buy a whole slate of brand-new GPUs. Inference costs would thus rapidly become way too…
The relationship engineers have with product management is more dysfunctional than with any other part of the company. There’s no shared culture or language like there is with other engineers, and the rules of “who gets to tell who what to do” aren’t as clear-cut as they are with managers. Engineers don’t have a lot in common with legal, or design, or sales, but they also don’t need to interact…
Many engineers should be doing less work. I don’t necessarily mean producing less code or fewer changes, but literally working fewer hours in the day. When they do work, they should be working at a slower pace. I like to aim to be running at 80% utilization by default: unless I have a high-pressure project going on, I spend 20% of my workday away from the computer. High-impact opportunities Why?…
Programmers were better back in the day, weren’t they? Back when we had real programmers. Not just people who got paid to write code, but people who lived it, who were obsessed with their craft, and whose code was a lively expression of themselves. Hackers were hackers in those days before money took over the industry. Don’t even get me started on LLMs. Could there be a better example of today’s…
Where are all the AI-generated projects? This is a common question from AI skeptics: if LLMs are so good at writing code, where is the tsunami of new AI-generated apps, services and games? I personally don’t find this to be much of a paradox. Writing code is only one of the bottlenecks involved in actually shipping a new product, after all. It’s also impossible to talk about the paid work I’ve…
There are only two ways to use LLMs in a computer program: as part of a pipeline, or as an agent. In other words, either you express the control flow of the program in code, or you give an LLM tools and allow it to manage the control flow itself 1 . Here’s how you might structure a trivial “summarize a bunch of information and email it to me” program as a pipeline: context = gather_context (…
In April last year, Kelsey Piper discovered that OpenAI’s o3 model was surprisingly good at figuring out where a photo was taken from. Like human “geoguessr” pros , o3 could sometimes take a nondescript photo of a beach and tell you exactly where it is. Here’s the example Kelsey gave: Several people reproduced this with good results: not a 100% success rate, but clearly far better than you’d do…
It’s common and correct to say that “all code is technical debt”. Adding code is a necessary evil for developing new features: you almost always have to do it, but each line of code adds to the complexity and maintenance burden of the system. All future changes to the system have to work with the existing code, or at least avoid breaking it. Once systems accumulate enough code, they become…
The engineer who says no all the time is a real archetype among senior and staff engineers. Their role is to slow things down, to block the development of features that add complexity, and to ensure that as little code gets written as possible (since code is a liability). We can think of this as the just-say-no engineer 1 , as opposed to the just-say-yes engineer. The just-say-yes engineer is…
A bit over a year ago I wrote How I use LLMs as a staff engineer . Here’s a brief summary of what I used AI for last year: Smart autocomplete with Copilot Short tactical changes in areas I don’t know well (always reviewed by a SME) Writing lots of use-once-and-throwaway research code Asking lots of questions to learn about new topics (e.g. the Unity game engine) Last-resort bugfixes, just in case…
Ever since Golden Gate Claude I’ve been fascinated with “steering”: the idea that you can guide LLM outputs by directly manipulating the activations of the model mid-flight. DeepSeek V4 Flash I was inspired to write this post by antirez’s recent project DwarfStar 4 , which is a version of llama.cpp that’s been stripped down to run only DeepSeek-V4-Flash. What’s so special about this model? It…