Hacker News: npollock

New comment by npollock in "Stack Overflow’s forum is dead but the company’s still kicking"

npollock — Tue, 26 May 2026 18:59:47 +0000

one of the interesting properties of public forums like SO is that the maintainers of software packages get visibility into the problems and frustrations that their users experience

we're losing that signal when the Q&A behavior shifts into language models

New comment by npollock in "Using AI to write better code more slowly"

npollock — Tue, 26 May 2026 00:43:54 +0000

learn by considering critique

New comment by npollock in "Anonymous request-token comparisons from Opus 4.6 and Opus 4.7"

npollock — Sat, 18 Apr 2026 20:03:31 +0000

You can configure the status line to get a feel for token usage:

[Opus 4.6] 3% context | last: 5.2k in / 1.1k out

add this to .claude/settings.json

"statusLine": { "type": "command", "command": "jq -r '\"[\\(.model.display_name)] \\(.context_window.used_percentage // 0)% context | last: \\(((.context_window.current_usage.input_tokens // 0) / 1000 * 10 | floor / 10))k in / \\(((.context_window.current_usage.output_tokens // 0) / 1000 * 10 | floor / 10))k out\"'" }

New comment by npollock in "The Walt Disney Company and OpenAI Partner on Sora"

npollock — Thu, 11 Dec 2025 22:52:17 +0000

Disney said "Our IP is $1B for a 3yr license, or we sue"

Altman said "We can pay with equity, but let's frame it as an investment"

No cash exchanged

New comment by npollock in "Cloudflare: We Will Get Google to Provide a Way to Block AI Overviews"

npollock — Tue, 08 Jul 2025 16:03:40 +0000

it could be the "grounding with google search" feature when using Gemini models

New comment by npollock in "Andrej Karpathy: Software in the era of AI [video]"

npollock — Thu, 19 Jun 2025 19:42:50 +0000

no, but a subset of English could be

New comment by npollock in "Text-to-LoRA: Hypernetwork that generates task-specific LLM adapters (LoRAs)"

npollock — Sun, 15 Jun 2025 21:47:14 +0000

LoRA adapters modify the model's internal weights

New comment by npollock in "Show HN: I built a more productive way to manage AI chats"

npollock — Sat, 24 May 2025 02:45:02 +0000

Here's how I would market this:

Create "packages" of context for popular API & Libraries - make them freely available via public url

Keep the packages up to date. They'll be newer than the cutoff date for many models, and they'll be cleaner then the data a model slurps in using web search.

Voila, you're now the trusted repository of context for the developer community.

You'll get a lot of inbound traffic/leads, and ideas for tangential use cases with commercial potential.

New comment by npollock in "Ask HN: Sharing Evals with OpenAI"

npollock — Thu, 22 May 2025 22:04:42 +0000

Are 7 free weekly eval runs fair compensation for sharing your evaluation and fine-tuning data with OpenAI?

Ask HN: Sharing Evals with OpenAI

npollock — Thu, 22 May 2025 22:04:42 +0000

Article URL: https://help.openai.com/en/articles/10306912-sharing-feedback-evaluation-and-fine-tuning-data-and-api-inputs-and-outputs-with-openai#h_c93188c569

Comments URL: https://news.ycombinator.com/item?id=44067638

Points: 1

# Comments: 1

New comment by npollock in "OpenAI to buy AI startup from Jony Ive"

npollock — Wed, 21 May 2025 18:26:12 +0000

"The takeover of io will provide OpenAI with about 55 hardware engineers, software developers and manufacturing experts"

6.5B / 55 = $118 million per engineer

not a cheap aquihire

New comment by npollock in "OpenAI to buy AI startup from Jony Ive"

npollock — Wed, 21 May 2025 18:24:21 +0000

if you own the device, you control the flow of data and dollars

Ask HN: Most accurate image to HTML model or service?

npollock — Mon, 10 Mar 2025 20:18:39 +0000

I'm looking to convert 30-40 fairly simple images into html and css. I've tried the frontier models, and they do an ok job but struggle with accurate positioning and sizing of elements.

Are there any open source models or paid services that can accurately convert images into html?

Comments URL: https://news.ycombinator.com/item?id=43325578

Points: 1

# Comments: 0

New comment by npollock in "Show HN: Superglue – open source API connector that writes its own code"

npollock — Thu, 27 Feb 2025 18:53:35 +0000

something like this that runs as a browser agent, allowing me to extract structured data from websites (whitelisted) using natural language queries

New comment by npollock in "RLHF Book"

npollock — Sun, 02 Feb 2025 18:46:49 +0000

A quote I found helpful:

"reinforcement learning from human feedback .. is designed to optimize machine learning models in domains where specifically designing a reward function is hard"

https://rlhfbook.com/c/05-preferences.html

Ask HN: How do you distill a frontier model?

npollock — Wed, 29 Jan 2025 21:01:56 +0000

Is it just obtaining a distribution of the next token predictions, or is it more complex?

Comments URL: https://news.ycombinator.com/item?id=42871105

Points: 2

# Comments: 0

New comment by npollock in "Show HN: LLMpeg"

npollock — Sun, 19 Jan 2025 04:29:20 +0000

does the tool snap to I-frames when slicing?

New comment by npollock in "Perplexity got ads"

npollock — Fri, 03 Jan 2025 22:29:58 +0000

how long before the advertisers pressure model providers to "fine-tune" mentions of their competitors?

New comment by npollock in "Launch HN: FlyCode (YC S22) – Stop losing revenue due to failed payments"

npollock — Wed, 30 Oct 2024 17:55:29 +0000

doesn't Stripe have an interest in maximizing this revenue recovery? they earn a percentage of successful charges

New comment by npollock in "The brain simulates actions and their consequences during REM sleep"

npollock — Sun, 18 Aug 2024 23:57:44 +0000

running inference