# May 17, 2026

## Frontier AI has broken the open CTF format
- **Frontier AI has fundamentally altered the CTF landscape**, rendering traditional skill measurement obsolete as AI tools can now solve many challenges with minimal human input, leading to a shift in competition dynamics.

## Every AI Subscription Is a Ticking Time Bomb for Enterprise
- **AI subscriptions are currently subsidized, creating a significant gap between what enterprises pay and the actual costs, which will lead to financial shocks when prices adjust.** Major providers like OpenAI and Anthropic are operating at a loss to capture market share, with reports indicating that companies could face bills between **$15,000 and $40,000 a month** for AI usage that currently costs only **$1,000** at subscription rates.

## δ-mem: Efficient Online Memory for Large Language Models
- **$δ$-mem** introduces a **lightweight memory mechanism** that enhances large language models by integrating a compact online state, allowing for efficient historical information reuse without the need for costly context window expansions.

## Kioxia and Dell cram 10 PB into slim 2RU server
- **Kioxia's LC9 SSDs** enable Dell to create a **10 PB all-flash storage server** in a compact **2 RU** form factor, enhancing storage density for enterprise applications.

## DeepSeek-V4-Flash means LLM steering is interesting again
- **DeepSeek-V4-Flash** revitalizes interest in **LLM steering**, enabling engineers to manipulate model outputs by adjusting internal activations, making it accessible for local models to compete with frontier agents.

## Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep
- **Semble** is a code search library that enables agents to retrieve relevant code snippets **instantly**, utilizing **~98% fewer tokens** than traditional methods like grep+read, achieving indexing in under a second and query responses in approximately **1.5 ms**.

## Unknowable Math Can Help Hide Secrets
- A graduate student, **Rahul Ilango**, has developed a novel type of **zero-knowledge proof** that leverages the limits of mathematics to enhance cryptographic security, overcoming long-standing challenges in the field.

## Self-Distillation Enables Continual Learning [pdf]
- **Self-Distillation Fine-Tuning (SDFT)** allows models to learn from demonstrations while maintaining existing skills, effectively addressing the challenge of **catastrophic forgetting** in continual learning.

## Do you agree with Judea that learning from data is not everything? [D]
- **Judea Pearl asserts that there are mathematical limits to learning from data alone, emphasizing that correlation does not equate to causation, which is often misunderstood in machine learning.**

## Recent Developments in LLM Architectures: KV Sharing, mHC, and Compressed Attention [P]
- **Recent LLM architectures** like Gemma 4 and DeepSeek V4 are innovating with techniques such as **KV sharing** and **compressed attention** to enhance long-context efficiency, significantly reducing memory and computational costs.

## DeepSeek Exposed: Users Can Access Each Other's Conversations with a Special Input[D]
- **DeepSeek's critical privacy flaw** allows users to access each other's conversations by entering a specific character, exposing serious vulnerabilities in session isolation and data security.

## #1 on memory benchmark LongMemEval with Gemini Flash, not Pro [R]
- **Gemini 3 Flash** achieved a **96.4%** score at top-50 in the LongMemEval benchmark, outperforming other models like Mem0 (94.8%) and Honcho (92.6%), showcasing the effectiveness of a smaller answering model to enhance retrieval quality.