Hacker News: heftykoo

Lessons from Building Claude Code: How We Use Skills

heftykoo — Wed, 18 Mar 2026 09:04:39 +0000

Article URL: https://twitter.com/trq212/status/2033949937936085378

Comments URL: https://news.ycombinator.com/item?id=47423330

Points: 1

# Comments: 0

Offloading FFmpeg with Cloudflare

heftykoo — Tue, 10 Mar 2026 08:41:37 +0000

Article URL: https://kentcdodds.com/blog/offloading-ffmpeg-with-cloudflare

Comments URL: https://news.ycombinator.com/item?id=47320571

Points: 46

# Comments: 21

New comment by heftykoo in "GPT‑5.3 Instant"

heftykoo — Wed, 04 Mar 2026 01:43:28 +0000

OpenAI's naming convention is slowly converging with Gillette razors. Can't wait for GPT-5.3 Instant Turbo Max Pro. Seriously though, if "Instant" just means a lower TTFT (Time To First Token) but regresses on complex reasoning, it's just a hardware accelerator for hallucinations. Fast wrong answers are still wrong answers.

New comment by heftykoo in "When AI writes the software, who verifies it?"

heftykoo — Wed, 04 Mar 2026 01:40:53 +0000

Another AI, obviously. And then a third AI to monitor the first two for conflicts of interest. Jokes aside, this is exactly the era where formal verification (like TLA+ or Lean, seeing the other post on the front page) actually makes commercial sense. If the code is generated, the only human output of value is the spec. We are moving from writing logic to writing constraints.

New comment by heftykoo in "Anthropic drops flagship safety pledge"

heftykoo — Wed, 25 Feb 2026 03:04:32 +0000

Ah, the classic AI startup lifecycle:

We must build a moat to save humanity from AI.

Please regulate our open-source competitors for safety.

Actually, safety doesn't scale well for our Q3 revenue targets.

New comment by heftykoo in "Show HN: Moonshine Open-Weights STT models – higher accuracy than WhisperLargev3"

heftykoo — Wed, 25 Feb 2026 03:02:35 +0000

Claiming higher accuracy than Whisper Large v3 is a bold opening move. Does your evaluation account for Whisper's notorious hallucination loops during silences (the classic 'Thank you for watching!'), or is this purely based on WER on clean datasets? Also, what's the VRAM footprint for edge deployments? If it fits on a standard 8GB Mac without quantization tricks, this is huge.

New comment by heftykoo in "Claude with Ads"

heftykoo — Mon, 09 Feb 2026 02:05:49 +0000

I can't wait to debug my code and have the LLM hallucinate a syntax error just to upsell me a subscription to a linter service.

New comment by heftykoo in "Big Tech vs. OpenClaw"

heftykoo — Sat, 07 Feb 2026 13:46:10 +0000

The hypocrisy is staggering. Big Tech scraped the entire open web, ignoring robots.txt and copyright whenever convenient, to train the every models that power these agents.

But the moment users employ a tool like OpenClaw to regain agency over that same data, it's branded as a "security threat" or "exploitation".

The "Authorized Scraping fro Me, But Not for thee" doctrine is becoming the standard ToS.

New comment by heftykoo in "GitHub Actions is slowly killing engineering teams"

heftykoo — Fri, 06 Feb 2026 08:52:15 +0000

The problem isn't CI/CD; the problem is "programming in configuration". We've somehow normalized a dev loop that involves `git commit -m "try fix"`, waiting 10 minutes, and repeating. Local reproduction of CI environments is still the missing link for most teams.

New comment by heftykoo in "Google goes from laggard to leader, pulls ahead of OpenAI with stellar AI growth"

heftykoo — Thu, 05 Feb 2026 06:20:46 +0000

The "innovator's dilemma" seems to have a new chapter: having enough TPU capacity to simply out-brute-force the competition once the direction is clear. It's less about "who built it first" and more about "who has the most H100s/TPUs and a decent enough compiler."

New comment by heftykoo in "Todd C. Miller – Sudo maintainer for over 30 years"

heftykoo — Tue, 03 Feb 2026 02:52:55 +0000

It's genuinely terrifying to think how much of the modern internet rests on the shoulders of a few people maintaining core utilities like sudo, curl, and openssl for decades. Todd is a legend.

New comment by heftykoo in "Removing the Gemini watermark: from OpenCV to LaMa to millisecond algorithm"

heftykoo — Thu, 29 Jan 2026 02:52:22 +0000

For clarity: the final algorithm isn’t novel research — it’s inspired by watermark-removal implementations shared in the community, then adapted specifically to Gemini’s logo geometry and optimized for browser-side execution.

The main engineering challenge for me wasn’t detection quality, but finding a point that balanced:

• reliability on textured images • millisecond latency • small binary size • zero server-side processing

I also made the extension paid mostly as an experiment to understand pricing, payments, and distribution for small developer tools. Building the payment/licensing flow took more time than the algorithm itself.

If anyone is curious, I’m happy to share more details about the detection heuristics and fill strategy.

New comment by heftykoo in "Removing the Gemini watermark: from OpenCV to LaMa to millisecond algorithm"

heftykoo — Thu, 29 Jan 2026 02:51:11 +0000

When Google Gemini started adding a watermark to generated images (including Nano Banana images), I wanted a way to download the original outputs without cropping or post-processing.

Over a few iterations I tried three approaches:

1) OpenCV heuristics Fixed-position detection + color estimation + inpainting. Fast, but fragile: works on flat backgrounds, fails on textured images.

https://geminiwatermarkcleaner.com/changelog/v1-1-0.html

2) LaMa inpainting High-quality reconstruction using a local LaMa model. Very accurate, but slow (~30s/image on CPU) and heavy to ship.

https://geminiwatermarkcleaner.com/changelog/v2-0-0.html

3) Lightweight watermark-specific algorithm Inspired by community implementations and optimized for Gemini’s logo pattern: geometry-aware detection + edge-preserving fill, no neural model. • Binary < 2MB • Millisecond latency • Runs fully locally in browser

https://geminiwatermarkcleaner.com/changelog/v3-0-0.html

I packaged this into a Chrome extension and a local web Gemini Watermark Remover tool: https://geminiwatermarkcleaner.com/gemini-watermark-remover....

Everything runs locally; images never leave the machine.

I also used this as a small experiment in building a paid micro-utility: payments, licensing, and basic marketing turned out harder than the algorithm itself.

Happy to answer questions about detection, inpainting tradeoffs, or browser-side image processing.

Removing the Gemini watermark: from OpenCV to LaMa to millisecond algorithm

heftykoo — Thu, 29 Jan 2026 02:51:11 +0000

Article URL: https://geminiwatermarkcleaner.com/changelog.html

Comments URL: https://news.ycombinator.com/item?id=46805139

Points: 1

# Comments: 2

New comment by heftykoo in "Ask HN: Engineers working AI tools. Are you working more or less?"

heftykoo — Tue, 18 Nov 2025 02:44:56 +0000

Work less time on work, work more time on my side project, because it can let me use less time to get job done, so I have more time on my side project.

And AI has expanded my boundaries — for example, I used to know nothing about image processing, but now with AI help, I’ve learned and use the technology and even built an initial product prototype using OpenCV, which helped my side project get off the ground successfully.