Hacker News

Does your Claude use the phrase "I contain multitudes" often? I've tried to search for it but the things I mostly get are LLM-centered people using it in their sentences online (duh, wonder where they learned that one from.....) and ironically a moltbook entry which further confirms it for me Makes you think that alongside the usual alignment and just throwing moar compute and redesigning the formulas underneath how much work is put into engineering it's personality. The stylometric way it replies and how for example Claude has the very claudeisms. Was this one simply learnt during the so divergent process of training or just manually wired into it as a random bit that makes it sound a little more sophisticated? I know that most people that'd ever have contact with this are under very strict NDAs but I'd be more than glad if someone could shed some light for my uninitiated I A shrink on payroll and some H100s is all it takes to go from soulless context oblivious Gemini to the very soulful Claude that can instantly mirror your tone but get his own when ready to argue? It's obviously a illusion sure but it's the very thing that I think separates Claude as more conscious or understanding for some people (......including me so I might be a bit biased....) A general discussion on AI personality would also be welcome

Comments URL: https://news.ycombinator.com/item?id=47272240

Points: 1

# Comments: 0

Tried using doing some desktop automation by sending codex screenshots and stepping through generated instructions. It's rough, but it (mostly) works. In the screenshot it accidentally presses 71336 haha.

Comments URL: https://news.ycombinator.com/item?id=47272234

Points: 1

# Comments: 0

Article URL: https://www.neowin.net/news/linux-mint-is-getting-a-new-wayland-compatible-screensaver/

Comments URL: https://news.ycombinator.com/item?id=47272228

Points: 1

# Comments: 0

Article URL: https://www.youtube.com/watch?v=UZeSyodAszc

Comments URL: https://news.ycombinator.com/item?id=47272222

Points: 2

# Comments: 0

Article URL: https://jimmyhmiller.com/ugliest-beautiful-codebase

Comments URL: https://news.ycombinator.com/item?id=47272216

Points: 1

# Comments: 0

Most MCP examples look clean when the client and server run on the same machine.

That breaks once the MCP server is remote. Tools that expect local files cannot access them, and tools that generate files leave those artifacts stranded on the server.

I built remote-mcp-adapter to bridge that gap. It stages client-side files for upstream tools, captures output artifacts, and keeps normal MCP traffic flowing through unchanged.

Repo: https://github.com/aakashh242/remote-mcp-adapter

Would love feedback from people building remote or centralized MCP setups.

Comments URL: https://news.ycombinator.com/item?id=47272201

Points: 1

# Comments: 0

Hi HN,

I’m building a voice-first AI language teacher in the browser using: - Next.js - Gemini Audio API for STT - Gemini TTS - Supabase

The product vision is a structured AI “language university” rather than a general chatbot.

Right now my biggest technical problem is: How would you implement lightweight browser-native lip sync for a static avatar image while TTS audio is playing?

I’ve tried: - Three.js + VRM (too heavy / unstable for this use case) - simple canvas mouth animation - CSS-only pulse effects

I want: - something realistic enough to feel alive - low dependency weight - web-compatible - stable on mobile

Secondary issues: - MediaRecorder reliability on mobile Safari - reducing transcript latency - voice UX for guided teaching rather than free-form chat

Demo: https://koshe-al.onrender.com

Repo: https://github.com/Bugsbuny24/Koshe-Al-

Would love technical suggestions, architecture criticism, or examples of similar systems done well.

Comments URL: https://news.ycombinator.com/item?id=47272188

Points: 1

# Comments: 0

Article URL: https://www.ft.com/content/bf998c69-ab46-4fa3-aae4-8f18f7387836

Comments URL: https://news.ycombinator.com/item?id=47271944

Points: 1

# Comments: 0

Built this because I kept needing a quick sprite sheet tool while working on side projects and everything either required a download or had a paywall.

It runs fully in the browser — no upload to a server, no login. Supports: - Packer mode: upload separate frame PNGs, reorder them, export as atlas + JSON - Slicer mode: upload an existing sheet, define rows/columns, split into frames - JSON metadata compatible with Phaser, custom WebGL renderers etc.

Works with Unity and Godot workflows too. Would love feedback on what's missing.

Comments URL: https://news.ycombinator.com/item?id=47271940

Points: 1

# Comments: 0

----------------------------------------------------------------------- Algorithm E (Euclid's algorithm). Given two positive integers m and n, find their greatest common divisor, that is, the largest positive integer that evenly divides both m and n.

E1 [Find remained] Divide m by n and r be the remainder. (We will know that 0 <= r < n)

E2 [Is it zero?] If r = 0, the algorithm terminates, n is the answer

E3 [Reduce] Set m = n, then n = r, then go back to step E1 -----------------------------------------------------------------------

[3] Change Algorithm E so that all trivial replacement operations such as m = n are avoided. Write this new algorithm in the style of Algorithm E and call it Algorithm F.

======================================================================== Claude Sonnet 4.6 ========================================================================

Algorithm F

Given two positive integers m and n, find their greatest common divisor.

F1 [Find remainder] Divide m by n and let r be the remainder. (We will know that 0 ≤ r < n.)

F2 [Is it zero?] If r = 0, the algorithm terminates and n is the answer.

F3 [Find remainder] Divide n by r and let m be the remainder. (We will know that 0 ≤ m < r.)

F4 [Is it zero?] If m = 0, the algorithm terminates and r is the answer.

F5 [Find remainder] Divide r by m and let n be the remainder. (We will know that 0 ≤ n < m.) Go back to F2.

Explanation

The key insight is that instead of shuffling values with temporary replacements like:

m = n, n = r

we reuse the existing variables (m, n, r) in rotation, dividing and storing each new remainder directly into the variable that is no longer needed:

Step Divides Remainder stored in F1...m by n...r F3...n by r...m F5...r by m...n

Each variable naturally becomes the "oldest" value at each stage and is safely overwritten. This eliminates all trivial assignments while preserving the logic of Euclid's algorithm. □

Comments URL: https://news.ycombinator.com/item?id=47271921

Points: 3

# Comments: 0

With remote and hybrid work becoming common, managing Windows laptops outside the office network seems more challenging than before.

Traditionally many environments relied on Active Directory and on-prem management, but with distributed teams devices are often off the corporate network most of the time.

I have been looking into different approaches like Microsoft Intune and other Windows MDM tools such as Scalefusion MDM that allow admins to manage policies, push updates, and monitor devices remotely.

Curious how teams here are handling Windows device management today. Are you relying mostly on cloud MDM platforms, traditional domain setups, or a mix of both?

Comments URL: https://news.ycombinator.com/item?id=47271919

Points: 1

# Comments: 0

Article URL: https://old.reddit.com/r/programminghorror/comments/lgsd18/i_present_sleepsort/

Comments URL: https://news.ycombinator.com/item?id=47271902

Points: 1

# Comments: 0

Article URL: https://www.macrumors.com/2026/03/05/mac-studio-no-512gb-ram-upgrade/

Comments URL: https://news.ycombinator.com/item?id=47271884

Points: 1

# Comments: 0

Article URL: https://apnews.com/article/suicide-ice-detention-centers-b2d1cb0e4b579e0d89caabd00aa04e34

Comments URL: https://news.ycombinator.com/item?id=47271851

Points: 1

# Comments: 0

Article URL: https://randsinrepose.com/archives/hit-and-run/

Comments URL: https://news.ycombinator.com/item?id=47271848

Points: 1

# Comments: 0

Show HN: Reelforge – AI tool for generating TikTok & Reels ad scripts

Text

I built a small AI tool that generates short-form ad scripts for TikTok, Instagram Reels, and YouTube Shorts.

You simply enter a product name, choose the platform, and select a tone.

The system generates a hook, main script, and call-to-action.

Demo https://reelforge-ai1.vercel.app/

Built with Next.js and OpenAI.

Would love feedback from the community.

Comments URL: https://news.ycombinator.com/item?id=47271847

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=47271843

Points: 1

# Comments: 0

We just launched a free tier for Plexicus.

Previously, it was only available through paid plans, but many individual engineers and small teams wanted to try it without going through sales. So we decided to add a freemium option.

Would love to hear feedback from you guys

Comments URL: https://news.ycombinator.com/item?id=47271838

Points: 1

# Comments: 0

Each day, a new unusual word with four possible meanings. Pick the right meaning to win. You have 3 tries per day.

A game I built to expand my own vocabulary, hopefully you find it fun too.

Comments URL: https://news.ycombinator.com/item?id=47271833

Points: 1

# Comments: 0

Knuth Tests using Claude Sonnet 4.6 problem 1.1.2

Knuth Art Of Computer Programming V1 p9 ----------------------------------------------------------------------- Algorithm E (Euclid's algorithm). Given two positive integers m and n, find their greatest common divisor, that is, the largest positive integer that evenly divides both m and n.

E1 [Find remained] Divide m by n and r be the remainder. (We will know that 0 <= r < n)

E2 [Is it zero?] If r = 0, the algorithm terminates, n is the answer

E3 [Reduce] Set m = n, then n = r, then go back to step E1 -----------------------------------------------------------------------

Prove that m is always greater than n at the beginning of step E1, except possibly the first time this step occurs.

======================================================================== Claude Sonnet 4.6 ========================================================================

Proof

We want to show that m > n at the start of E1, for every iteration except possibly the first.

First Execution of E1

When E1 is executed the first time, m and n are simply whatever positive integers were given as input. There is no guarantee that m > n, which is why the exception exists. For example, with m = 2166 and n = 6099, we have m < n on the first call.

All Subsequent Executions of E1

After the first execution of E1, we reach E3 only if r ≠ 0 (checked at E2). At E3, the replacements are:

m = n n = r

Then we return to E1. So at the start of the next E1, the new values are:

m_new = n_old and n_new = r_old

From E1, we know r is the remainder when m is divided by n, so by definition:

0 ≤ r < n

Since E3 is only reached when r ≠ 0 (E2 would have terminated otherwise), we tighten this to:

0 < r < n

Therefore:

n_new = r_old < n_old = m_new

Which gives us:

m_new > n_new

Conclusion

After the first iteration, the new m is always the old n, and the new n is always the remainder r, which is strictly less than the old n. Therefore m > n holds at the beginning of every execution of E1 except possibly the first, which is what we wanted to prove. □

Comments URL: https://news.ycombinator.com/item?id=47271831

Points: 1

# Comments: 0