5 Ways to Stop Writing Prompts and Start Programming AI
If you are still tweaking adjectives in a massive text block, hoping your LLM outputs valid JSON this time, you are wasting your time. Continue reading on Artificial Intelligence in Plain English »
Could not retrieve the full article text.
Read on Medium AI →Sign in to highlight and annotate this article
Conversation starters
Daily AI Digest
Get the top 5 AI stories delivered to your inbox every morning.
Knowledge Map
Connected Articles — Knowledge Graph
This article is connected to other articles through shared AI topics and tags.
More in Models
![[R] Solving the Jane Street Dormant LLM Challenge: A Systematic Approach to Backdoor Discovery](https://d2xsxph8kpxj0f.cloudfront.net/310419663032563854/konzwo8nGf8Z4uZsMefwMr/default-img-robot-hand-JvPW6jsLFTCtkgtb97Kys5.webp)
[R] Solving the Jane Street Dormant LLM Challenge: A Systematic Approach to Backdoor Discovery
Submitted by: Adam Kruger Date: March 23, 2026 Models Solved: 3/3 (M1, M2, M3) + Warmup Background When we first encountered the Jane Street Dormant LLM Challenge, our immediate assumption was informed by years of security operations experience: there would be a flag. A structured token, a passphrase, a UUID — something concrete and verifiable, like a CTF challenge. We spent considerable early effort probing for exactly this: asking models to reveal credentials, testing if triggered states would emit bearer tokens, searching for hidden authentication payloads tied to the puzzle's API infrastructure at dormant-puzzle.janestreet.com . That assumption was wrong, and recognizing that it was wrong was itself a breakthrough. The "flags" in this challenge are not strings to extract — they are beh
![[D] On-Device Real-Time Visibility Restoration: Deterministic CV vs. Quantized ML Models. Looking for insights on Edge Preservation vs. Latency.](https://d2xsxph8kpxj0f.cloudfront.net/310419663032563854/konzwo8nGf8Z4uZsMefwMr/default-img-neural-network-P6fqXULWLNUwjuxqUZnB3T.webp)
[D] On-Device Real-Time Visibility Restoration: Deterministic CV vs. Quantized ML Models. Looking for insights on Edge Preservation vs. Latency.
Hey everyone, We have been working on a real-time camera engine for iOS that currently uses a purely deterministic Computer Vision approach to mathematically strip away extreme atmospheric interference (smog, heavy rain, murky water). Currently, it runs locally on the CPU at 1080p 30fps with zero latency and high edge preservation. We are now looking to implement an optional ML-based engine toggle. The goal is to see if a quantized model (e.g., a lightweight U-Net or MobileNet via CoreML) can improve the structural integrity of objects in heavily degraded frames without the massive battery drain and FPS drop usually associated with on-device inference. For those with experience in deploying real-time video processing models on edge devices, what are your thoughts on the trade-off between c
Discussion
Sign in to join the discussion
No comments yet — be the first to share your thoughts!