Hacker News: dcastm

New comment by dcastm in "A new era for software testing"

dcastm — Thu, 11 Jun 2026 19:36:04 +0000

Same for me. I actively ask the LLM to write as few tests as possible. Otherwise you end up redundant and low value ttests.

New comment by dcastm in "Postmortem: TanStack NPM supply-chain compromise"

dcastm — Tue, 12 May 2026 14:06:34 +0000

Reminder to use a cooldown period: https://dylancastillo.co/til/securing-package-managers.html

5 years of freelancing: same work, better pay, and a ceiling

dcastm — Sun, 03 May 2026 15:19:33 +0000

Article URL: https://dylancastillo.co/posts/5-years-freelancing-in-europe.html

Comments URL: https://news.ycombinator.com/item?id=47997827

Points: 4

# Comments: 0

LLM research on Hacker News is drying up

dcastm — Fri, 24 Apr 2026 18:12:15 +0000

Article URL: https://dylancastillo.co/til/llm-research-on-hacker-news-is-dying.html

Comments URL: https://news.ycombinator.com/item?id=47893860

Points: 30

# Comments: 11

The last shall be (slightly) safer

dcastm — Wed, 01 Apr 2026 17:22:53 +0000

Article URL: https://dylancastillo.co/til/securing-package-managers.html

Comments URL: https://news.ycombinator.com/item?id=47603806

Points: 2

# Comments: 0

Cover Your Ass

dcastm — Sun, 22 Feb 2026 14:51:21 +0000

Article URL: https://dylancastillo.co/posts/cover-your-ass.html

Comments URL: https://news.ycombinator.com/item?id=47111445

Points: 4

# Comments: 0

I hate AI side projects

dcastm — Fri, 20 Feb 2026 22:03:00 +0000

Article URL: https://dylancastillo.co/posts/ai-side-projects.html

Comments URL: https://news.ycombinator.com/item?id=47094637

Points: 80

# Comments: 120

New comment by dcastm in "Show HN: AI data analyst that runs Python in the browser"

dcastm — Sun, 15 Feb 2026 18:10:34 +0000

Just Pyodide for now!

Show HN: AI data analyst that runs Python in the browser

dcastm — Sun, 15 Feb 2026 17:46:22 +0000

Article URL: https://kasipa.com/

Comments URL: https://news.ycombinator.com/item?id=47025711

Points: 1

# Comments: 3

New comment by dcastm in "Qwen3-Coder-Next"

dcastm — Tue, 03 Feb 2026 18:33:34 +0000

I have the same experience with local models. I really want to use them, but right now, they're not on par with propietary models on capabilities nor speed (at least if you're using a Mac).

New comment by dcastm in "No management needed: anti-patterns in early-stage engineering teams"

dcastm — Wed, 14 Jan 2026 15:54:03 +0000

I’ve worked with great engineers from India/Pakistan. I didn’t hire them, so don’t know too much about the process of how to find them but they were definitely as good as anyone I’ve seen in Europe.

New comment by dcastm in "No management needed: anti-patterns in early-stage engineering teams"

dcastm — Tue, 13 Jan 2026 22:10:05 +0000

I live in Spain. I’ve been in the industry for the last 10 years.

I’ve seen from a very close distance several European companies move a big part of their operations to India. Have had close friends laid off recently and seen them struggle for months to find a new jobs. Plus, I see tighter freelance market these days.

This was unthinkable not long ago.

New comment by dcastm in "No management needed: anti-patterns in early-stage engineering teams"

dcastm — Tue, 13 Jan 2026 21:26:25 +0000

Which is why fewer and fewer companies are hiring in Europe.

New comment by dcastm in "Trump says Venezuela’s Maduro captured after strikes"

dcastm — Sat, 03 Jan 2026 14:03:58 +0000

Except we didn’t and there’s already an ongoing refugee crisis.

[0] https://terrytao.wordpress.com/2024/08/02/what-are-the-odds-...

[1] https://en.wikipedia.org/wiki/Venezuelan_refugee_crisis

New comment by dcastm in "Structured outputs create false confidence"

dcastm — Sun, 21 Dec 2025 16:16:10 +0000

While 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 tokens (e.g., reasoning_effort for OpenAI models). You get the best of both worlds: freeform reasoning and structured output. I tested this, and the results were very similar for both.

2. Let Me Speak Freely? had several methodological issues. I address some of them (and .txt's rebuttal) here: https://dylancastillo.co/posts/say-what-you-mean-sometimes.h...

3. There's no silver bullet. Structured outputs might improve or worsen your results depending on the use case. What you really need to do is run your evals and make a decision based on the data.

New comment by dcastm in "Show HN: Are You a Good Estimator?"

dcastm — Wed, 29 Oct 2025 22:05:52 +0000

Makes sense! I like the slider idea, but not sure if it’d introduce some bias to the results.

Show HN: Are You a Good Estimator?

dcastm — Wed, 29 Oct 2025 19:29:14 +0000

Article URL: https://estimator.dylancastillo.co/

Comments URL: https://news.ycombinator.com/item?id=45751827

Points: 15

# Comments: 9

How to Write Good Prompts

dcastm — Tue, 21 Oct 2025 10:06:43 +0000

Article URL: https://dylancastillo.co/posts/prompt-engineering-101.html

Comments URL: https://news.ycombinator.com/item?id=45654175

Points: 3

# Comments: 0

ClaudeCodeholic

dcastm — Sat, 18 Oct 2025 15:04:00 +0000

Article URL: https://dylancastillo.co/posts/claudecodeholic.html

Comments URL: https://news.ycombinator.com/item?id=45627944

Points: 4

# Comments: 1

New comment by dcastm in "Extract-0: A specialized language model for document information extraction"

dcastm — Tue, 30 Sep 2025 19:18:38 +0000

Hey OP, I found some issues with your code:

During SFT, it uses the full training dataset[1]:

df = pd.read_csv('data/extraction_training_data.csv')

And during the evaluation, it uses the middle part of the same dataset[2]:

df = pd.read_csv('data/extraction_training_data.csv')

df = df[100000:100000+NUM_TEST_SAMPLES]

Also, you split train/test/val by chunk and not by document[3]. Then, the model "has seen" the documents that you're using to evaluate it (even if you're not evaluating it on the same chunks).

[1]: https://github.com/herniqeu/extract0/blob/0f8696a6fb1b620658...

[2]: https://github.com/herniqeu/extract0/blob/0f8696a6fb1b620658...

[3]: https://github.com/herniqeu/extract0/blob/0f8696a6fb1b620658...