Top all-time · #6904
Dylan Castillo
Personal blog · dylancastillo.coaiprogrammingcareer
Dylan Castillo's personal blog at dylancastillo.co covers ai, programming and career. Since 2022, 11 different Hacker News readers have linked it 31 times, most often “Are AI Labs Pelicanmaxxing?”. It is still writing: the last post was on 2026-08-01, and you can follow it by feed.
Too few sampled comments (9) to say how often HN recommends it.?
Why HN links it
HN readers link the 'pelicanmaxxing' post, which looks for evidence that AI labs tune models to draw pelicans on bicycles, as support for the claim that there is little such evidence. Also cited for LinkedIn tips.
Summarised by Claude from a sample of 9 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Most-linked posts?
- Are AI Labs Pelicanmaxxing?8 people · 8 mentions · first linked 2026-07-22
- Structured outputs can hurt the performance of LLMs2 people · 3 mentions · first linked 2024-12-09
What Hacker News says?
There was a real probe into this, seems like no: https://dylancastillo.co/posts/pelicanmaxxing.html https://news.ycombinator.com/item?id=49010129
Comment linking /posts/pelicanmaxxing.html
on “Claude Fable 5.1 and Claude Mythos 5.1”Submitted Build an AI Search Engine Using FastAPI, Qdrant, and ChatGPT as a story · 5 points
HN discussionThis "no, Virginia, they aren't pelicanmaxxing" article is what broodbucket is probably thinking of: https://dylancastillo.co/posts/pelicanmaxxing.html In any case, pelicanmaxxing comments without any further proof or evidence probably fall under the "Please don't post shallow dismissals" guideline.
Comment linking /posts/pelicanmaxxing.html
on “MiMo v2.6”Submitted LLM research on Hacker News is drying up as a story · 4 points
HN discussionSomeone tested this, and it doesn't look to be saturated. https://dylancastillo.co/posts/pelicanmaxxing.html Simon made I think a very good argument for why it's still useful, if not the most robust benchmark in the world. https://simonwillison.net/2026/Jul/16/kimi-k3/
Comment linking /posts/pelicanmaxxing.html
on “Muse Spark 1.3”Submitted ClaudeCodeholic as a story · 4 points
HN discussionFor all the comments of "I'm sure they're fine-tuning for pelicans": https://dylancastillo.co/posts/pelicanmaxxing.html "Sorry, HN haters, but there’s little evidence that AI labs are pelicanmaxxing. Or at least they’re not doing it in a plainly obvious manner."
Comment linking /posts/pelicanmaxxing.html
on “Muse Spark 1.3”Submitted Are AI Labs Pelicanmaxxing? as a story · 3 points
HN discussionSomeone studied this (among other thigns): https://dylancastillo.co/posts/pelicanmaxxing.html . Pelicans on bikes always face right in this test, but other animals on other transportation methods sometimes face left.
Comment linking /posts/pelicanmaxxing.html
on “Muse Spark 1.3”Submitted 5 years of freelancing: same work, better pay, and a ceiling as a story · 3 points
HN discussionThat’s a fair question, but it seems that it’s not yet necessary. See here https://dylancastillo.co/posts/pelicanmaxxing.html https://simonwillison.net/2026/Jul/22/
Comment linking /posts/pelicanmaxxing.html
on “Hy4 preview”Submitted How to Write Good Prompts as a story · 3 points
HN discussionWe are also interested in this and may focus on it in a future post. They are two different perspectives: (1) What is the overall quality of LLM+serving stack+harness, and (2) How close does the behavior of our setup match the first-party provider's? In this post, we focused on the latter. Moonshot AI, the creators of Kimi, recommend constrained decoding to be on by default, and also serve Kimi that way. This implie…
Comment linking /posts/say-what-you-mean-sometimes.html
on “Fixing Tool-Call Accuracy in vLLM”Submitted Structured outputs can hurt the performance of LLMs as a story · 3 points
HN discussionWhile I agree that you must be careful when using structured outputs, the article doesn't provide good arguments: 1. In the examples provided, the author compares freeform CoT + JSON output vs. non-CoT structured output. This is unfair and biases the results towards what they wanted to show. These days, you don't need to include a "reasoning" field in the schema as mentioned in the article; you can just use thinking…
Comment linking /posts/say-what-you-mean-sometimes.html
on “Structured outputs create false confidence”Submitted Structured outputs in LLMs: don't put the cart before the horse as a story · 3 points
HN discussionReminder to use a cooldown period: https://dylancastillo.co/til/securing-package-managers.html
Comment linking /til/securing-package-managers.html
on “TanStack NPM Packages Compromised”Submitted The last shall be (slightly) safer as a story · 2 points
HN discussionI wrote some tips on how to use LinkedIn effectively: https://dylancastillo.co/tips-for-standing-out-on-linkedin/ It’s based on published papers about its algorithm + personal experience.
Comment linking /tips-for-standing-out-on-linkedin/
on “Ask HN: How many of you have been laid of twice this year already?”Submitted I hate AI side projects as a story · 2 points
HN discussionAgreed. It's far from solved. Modern LLMs still generate pelican bike SVGs with obvious errors: * some omitted the bottom of the diamond which connects from the pedals to the rear wheel * some added an extra connection from the pedals to the front wheel, making it impossible to steer * none could align the head tube with the fork * none added a correct offset to the fork * none could generate the chain properly in a…
Comment linking /posts/pelicanmaxxing.html
on “Karpathy's Pelican”Submitted I found some issues with Gemini's structured outputs as a story · 2 points
HN discussionSubmitted Deploying a Django App with Kamal, AWS ECR, and GHA as a story · 2 points
HN discussionSubmitted Automatically create a kamal-ready VPS using Terraform and Hetzner as a story · 2 points
HN discussionSubmitted Finetuning won't add new knowledge to your model as a story · 2 points
HN discussionSubmitted Honey, I shrunk the embeddings: Matryoshka vs. PCA as a story · 1 points
HN discussionSubmitted Cover Your Ass as a story · 1 points
HN discussionSubmitted Image classification using In-context Learning and Gemini Flash 1.5 as a story · 1 points
HN discussionSubmitted Create your own code interpreter chatbot with Pyodide as a story · 1 points
HN discussionSubmitted Semantic search with Elasticsearch and LangChain as a story · 1 points
HN discussion
