Unicode variation selectors for invisible LLM injection

Comments (1)

mpoteat · 1h ago

Recently, LeetCode has begun hiding instructions (using css) in their competitions to use particular unique variable names like 'dexolarniv' - and anyone using such a variable name in their submission gets summarily banned.

In their implementation, the hidden prompt does show up on copy/paste - however I tested this method, and the LLM (i.e. ChatGPT) does still follow the Unicode-hidden instruction and uses 'dexolarniv' in the code returned! So I think this is (right now) a viable invisible injection strategy.

We should have the ability to run any code we want on hardware we own (hugotunius.se)

Cognitive load is what matters (github.com)

I ditched Docker for Podman (codesmash.dev)

30 minutes with a stranger (pudding.cool)

Next.js is infuriating (blog.meca.sh)

996 (lucumr.pocoo.org)

Anthropic agrees to pay $1.5B to settle lawsuit with book authors (nytimes.com)

Show HN: I recreated Windows XP as my portfolio (mitchivin.com)

The MacBook has a sensor that knows the exact angle of the screen hinge (twitter.com)

Google can keep its Chrome browser but will be barred from exclusive contracts (cnbc.com)

Using Claude Code to modernize a 25-year-old kernel driver (dmitrybrant.com)

Stripe Launches L1 Blockchain: Tempo (tempo.xyz)

“This telegram must be closely paraphrased before being communicated to anyone” (history.stackexchange.com)

Almost anything you give sustained attention to will begin to loop on itself (henrikkarlsson.xyz)

Where's the shovelware? Why AI coding claims don't add up (mikelovesrobots.substack.com)

Google AI Overview made up an elaborate story about me (bsky.app)

Claude Code: Now in Beta in Zed (zed.dev)

Eternal Struggle (yoavg.github.io)

I'm absolutely right (absolutelyright.lol)

LLM Visualization (bbycroft.net)

Notes on Managing ADHD (borretti.me)

MIT Study Finds AI Use Reprograms the Brain, Leading to Cognitive Decline (publichealthpolicyjournal.com)

Serverless Horrors (serverlesshorrors.com)

Fil's Unbelievable Garbage Collector (fil-c.org)

Wikipedia survives while the rest of the internet breaks (theverge.com)

AI surveillance should be banned while there is still time (gabrielweinberg.com)

The maths you need to start understanding LLMs (gilesthomas.com)

Anthropic raises $13B Series F (anthropic.com)

We already live in social credit, we just don't call it that (thenexus.media)

Google: 'Your $1000 phone needs our permission to install apps now' [video] (youtube.com)

Bear is now source-available (herman.bearblog.dev)

A staff engineer's journey with Claude Code (sanity.io)

Purposeful animations (emilkowal.ski)

Atlassian is acquiring The Browser Company (cnbc.com)

Implementing a Foil Sticker Effect (4rknova.com)

Tesla changes meaning of 'Full Self-Driving', gives up on promise of autonomy (electrek.co)

Magic Lantern Is Back (magiclantern.fm)

Are we decentralized yet? (arewedecentralizedyet.online)

14 Killed in anti-government protests in Nepal (tribuneindia.com)

The staff ate it later (en.wikipedia.org)

The Little Book of Linear Algebra (github.com)

What Is the Fourier Transform? (quantamagazine.org)

Six months into tariffs, businesses have no idea how to price anything (wsj.com)

I bought the cheapest EV, a used Nissan Leaf (jeffgeerling.com)

WiFi signals can measure heart rate (news.ucsc.edu)

VibeVoice: A Frontier Open-Source Text-to-Speech Model (microsoft.github.io)

NPM debug and chalk packages compromised (aikido.dev)

Jujutsu for everyone (jj-for-everyone.github.io)

%CPU utilization is a lie (brendanlong.com)

Patrick Winston: How to Speak (2018) [video] (youtube.com)

Unicode variation selectors for invisible LLM injection

Comments (1)