seangoedecke.com
About This Feed
No description available for this feed.
Feed Activity
Latest Posts
Everything I know about good API design
Most of what modern software engineers do1 involves APIs: public interfaces for communicating with a program, like this one from Twilio. I’ve spent a lot of time working with APIs,...
Don't feed me AI slop
In the early days of any new technology, the relevant social norms are still being workshopped. For mobile phones, that meant collectively deciding when and where phones should be on...
The famous "bottomless pit" AI greentext is fake
Many people believe1 this is the best piece of art or humour that AI has ever produced: This was generated three years ago by GPT-3. It’s notable by itself that...
Is chain-of-thought AI reasoning a mirage?
Reading research papers and articles about chain-of-thought reasoning1 makes me frustrated. There are many interesting questions to ask about chain-of-thought: how accurately it reflects the actual process going on, why...
What's the strongest AI model you can train on a laptop in five minutes?
What’s the strongest model I can train on my MacBook Pro1 in five minutes? I’ll give the answer upfront: the best 5-minute model I could train was a ~1.8M-param GPT-style...
OpenAI's new open-source model is basically Phi-5
OpenAI just released its first ever open-source1 large language models, called gpt-oss-120b and gpt-oss-20b. You can talk to them here. Are they good models? Well, that depends on what you’re...
AI interpretability has the same problems as philosophy of mind
If you want to know how a particular AI language model works, the current state-of-the-art approach is to use a “probe”. Pick one layer of the model - a single...
Why I don't allow AI-generated content on my blog
I will never include any AI-generated content on this website. In general I’m a big AI believer, and I use large language models a lot: both in my day job,...
Trying to impress people you don't respect
In the first few years of my career, I knew next to nothing about how to do good work in a tech company. The senior and staff engineers I worked...
The refusal problem in large language models
Everyone’s interested in what language models can do. But what about what they will do? Try giving ChatGPT 4o a menial task that will generate a few pages of boilerplate...
Can small AI models think as well as large ones?
An AI trend that’s emerged in the last few months1 is the idea of a “cognitive core”. Instead of trying to build the largest, most capable model we can, should...
What kind of work I want (in 2025)
In 2021, I wrote this post describing the kind of work I wanted then. Almost five years later, it’s time for an update. What kind of work do I want?...
Do not yell at the language model
A few days ago, the news went viral that Replit’s AI coding agent went rogue and deleted a production database. A lot has been written about it since then: how...
The case for regulating AI companions
In April, OpenAI screwed up by releasing a version of GPT-4o that was overly sycophantic. If you told it that Brian Cox was sending you secret messages in the last...
How to actually use Model Context Protocol
Everyone’s very excited about Model Context Protocol, or MCP for short. In a sentence, MCP is a universal protocol for exposing sets of tools to AI agents - instead of...
Vibe coding and the silent AI war inside tech companies
I recently wrote an article for LeadDev about how software engineers are dealing with the AI hype in practice. Here’s my position, in short: I think AI is pretty great....
Practical notes on getting LLMs to generate new ideas
Large language models struggle to generate new ideas. To AI skeptics, this seems trivially true, since they believe LLMs can only regurgitate content from their training data1. To AI believers,...
The three great virtues of an AI-assisted programmer
In 1991, Larry Wall famously wrote that the three great virtues of a programmer were “laziness, impatience, and hubris”. A programmer should be lazy enough that they go to great...
Building an RTS where you issue orders to AI agents
I’ve always been interested by the third-act turn in Ender’s Game where the child strategists switch from directly controlling units to giving higher-level strategic orders. The communication problem just seems...
METR's AI productivity study is really good
I complain a lot on this blog about AI studies. Given that, I ought to give credit when an AI study comes out that I think is really good. Yesterday...