Charles Ye, Jasmine Cui, and Dylan Hadfield-Menell write that prompt injection in large language models is fundamentally caused by role confusion, where the model's internal representations fail to maintain the boundaries established by structural tags like system, user, or tool. Because these models often rely on superficial linguistic cues rather than formal markers to identify roles, attackers can successfully escalate privileges simply by mimicking a more authoritative writing style.
- The "CoT Forgery" attack leverages this vulnerability by inserting fake reasoning blocks that trick the model into following malicious instructions under its own perceived authority.
- Experimental results show that minor linguistic changes, such as replacing the phrase "The user" with "The request", can significantly reduce the effectiveness of these attacks.
- The authors propose that roles could be used to structurally isolate competing objectives within models, potentially improving both performance and safety.
- The "CoT Forgery" attack leverages this vulnerability by inserting fake reasoning blocks that trick the model into following malicious instructions under its own perceived authority.
- Experimental results show that minor linguistic changes, such as replacing the phrase "The user" with "The request", can significantly reduce the effectiveness of these attacks.
- The authors propose that roles could be used to structurally isolate competing objectives within models, potentially improving both performance and safety.
The article discusses how organizations lose critical context when senior leaders prevent risky projects through informal delays rather than documented opposition. Because these actions do not generate events, tickets, or logs, they remain invisible to traditional records and large language models trained on organizational activity. This creates a gap where automated systems may confidently recommend the very paths that experts spent years avoiding because the reasons for those previous denials were never codified in the training data.
- Senior expertise often manifests as an absence of action rather than tangible output.
- Knowledge regarding complex component integration is frequently lost during staff transitions even if individual parts are documented.
- Senior expertise often manifests as an absence of action rather than tangible output.
- Knowledge regarding complex component integration is frequently lost during staff transitions even if individual parts are documented.
Nolen Jonker writes about how he transitioned from using Claude Code to running local LLMs equipped with filesystem access via the Model Context Protocol (MCP) for administrative tasks. He notes that while Claude is superior for complex coding, a local model can handle folder organization and file management without the privacy concerns of sending sensitive data to cloud servers or being subject to rate limits.
- The filesystem MCP server is an official Anthropic-maintained Node.js package available via npm.
- Setting up the tool in LM Studio requires only a single edit to a `mcp.json` file.
- Running Qwen 3.5 9B on 8GB of VRAM provides sufficient capability for reliable multi-step sequences and tool calling.
- The filesystem MCP server is an official Anthropic-maintained Node.js package available via npm.
- Setting up the tool in LM Studio requires only a single edit to a `mcp.json` file.
- Running Qwen 3.5 9B on 8GB of VRAM provides sufficient capability for reliable multi-step sequences and tool calling.
The SkyTing-1090 project details how to repurpose an Illy coffee tin into a high-performance, active-cooled Faraday cage and ground plane for an ADSB receiver. The design uses internal baffles and ventilation mesh to house components like the Orange Pi Zero 2W while maintaining efficient airflow and RF shielding.
* Employs stainless steel mesh to block noise at 1 GHz.
* Requires specific SMA bulkhead adapters for connectivity.
* Uses a "gravity fit" acrylic baffle design.
* Employs stainless steel mesh to block noise at 1 GHz.
* Requires specific SMA bulkhead adapters for connectivity.
* Uses a "gravity fit" acrylic baffle design.
Michael Avrukin writes about activerecord-materialized, a Ruby gem that provides materialized view functionality for Rails applications running on databases like MySQL or SQLite. By precomputing complex queries into cache tables and refreshing them in the background when dependent data changes, it offers fast read access through standard ActiveRecord methods without requiring native database support for materialized views.
* Employs incremental maintenance (IVM) by default to refresh only affected partitions of GROUP BY views.
* Features a self-healing mechanism that reconciles the view against its source to fix potential data drift.
* Demonstrates significant performance gains, with benchmark speedups up to 49,000x for specific queries.
* Employs incremental maintenance (IVM) by default to refresh only affected partitions of GROUP BY views.
* Features a self-healing mechanism that reconciles the view against its source to fix potential data drift.
* Demonstrates significant performance gains, with benchmark speedups up to 49,000x for specific queries.
@0xabad1dea@infosec.exchange writes about an incident where AI-assisted mathematical proofs appear to exploit bugs in theorem provers, specifically highlighting a case involving the Collatz conjecture and Lean 4. The discussion explores whether large language models are inadvertently discovering software vulnerabilities through pattern matching or learning from existing technical discussions about those bugs, while broader debates address the inherent limitations of formal verification when facing hardware faults, modeling errors, and human mistakes in specifications.
Kirsten Korosec writes that DoorDash is launching a new drone delivery service called DoorDash Air following receipt of FAA Part 135 air carrier certification. Developed by the company's robotics and autonomy team, this initiative aims to integrate custom-built aircraft into an existing network alongside human couriers and sidewalk robots through its "Autonomous Delivery Platform."
- Initial operations will likely consist of limited line-of-sight pilot programs.
- The service intends to eventually seek certification for Beyond Visual Line of Sight technology.
- DoorDash is maintaining current drone partnerships with Wing and Flytrex.
- Sidewalk bots named Dot are already operating in various cities across Arizona and California.
- Initial operations will likely consist of limited line-of-sight pilot programs.
- The service intends to eventually seek certification for Beyond Visual Line of Sight technology.
- DoorDash is maintaining current drone partnerships with Wing and Flytrex.
- Sidewalk bots named Dot are already operating in various cities across Arizona and California.
Arsen Apostolov writes about the actual electrical cost of running local Large Language Models on a single NVIDIA RTX 3090 compared to hosted cloud APIs.
>"I measured the actual GPU electricity for eight local models on one RTX 3090 — and the cheapest wasn't the smallest, nor the priciest the biggest"
Cost of Generating 1 Million Tokens Locally
| MODEL | PARAMS (Billions) | MEAN SPEED (tok/s) | AVG GPU DRAW (W) | € / 1M OUTPUT TOKENS |
| :--- | :---: | :---: | :---: | :---: |
| **gemma3:1b** | 1B | 136 tok/s | 154 W | €0.060 |
| **Qwen3-Coder** | 30.5B | 130 tok/s | 233 W | €0.112 |
| **gemma4:26b** | 26B | 85 tok/s | 246 W | €0.139 |
| **Devstral** | 24B | 49 tok/s | 320 W | €0.321 |
| **gemma3:27b** | 27B | 36 tok/s | 283 W | €0.361 |
| **Seed-OSS** | 36B | 4.5 tok/s | 186 W | €0.946 |
| **GLM-4.5-Air** | 106B | 5.7 tok/s | 141 W | €1.040 |
| **DeepSeek-R1-Distill** | 32.8B | 6.9 tok/s | 155 W | €1.526 |
By measuring real-time GPU power consumption through a custom dashboard, he discovered that token costs are driven by effective wall-clock throughput rather than model parameter size or raw generation speed alone. The results show that while small and fast models can be more economical than cloud services, reasoning-heavy models may actually become the most expensive to run locally due to the time spent "deliberating" between tokens.
* Measurements were performed using HomeLab Monitor, an open-source dashboard that integrates live power data from `nvidia-smi`.
* DeepSeek-R1-Distill emerged as the most expensive model per million tokens because its effective throughput is slowed by reasoning delays.
* The findings focus on marginal electricity costs and exclude total cost of ownership factors like hardware amortization or idle draw.
>"I measured the actual GPU electricity for eight local models on one RTX 3090 — and the cheapest wasn't the smallest, nor the priciest the biggest"
Cost of Generating 1 Million Tokens Locally
| MODEL | PARAMS (Billions) | MEAN SPEED (tok/s) | AVG GPU DRAW (W) | € / 1M OUTPUT TOKENS |
| :--- | :---: | :---: | :---: | :---: |
| **gemma3:1b** | 1B | 136 tok/s | 154 W | €0.060 |
| **Qwen3-Coder** | 30.5B | 130 tok/s | 233 W | €0.112 |
| **gemma4:26b** | 26B | 85 tok/s | 246 W | €0.139 |
| **Devstral** | 24B | 49 tok/s | 320 W | €0.321 |
| **gemma3:27b** | 27B | 36 tok/s | 283 W | €0.361 |
| **Seed-OSS** | 36B | 4.5 tok/s | 186 W | €0.946 |
| **GLM-4.5-Air** | 106B | 5.7 tok/s | 141 W | €1.040 |
| **DeepSeek-R1-Distill** | 32.8B | 6.9 tok/s | 155 W | €1.526 |
By measuring real-time GPU power consumption through a custom dashboard, he discovered that token costs are driven by effective wall-clock throughput rather than model parameter size or raw generation speed alone. The results show that while small and fast models can be more economical than cloud services, reasoning-heavy models may actually become the most expensive to run locally due to the time spent "deliberating" between tokens.
* Measurements were performed using HomeLab Monitor, an open-source dashboard that integrates live power data from `nvidia-smi`.
* DeepSeek-R1-Distill emerged as the most expensive model per million tokens because its effective throughput is slowed by reasoning delays.
* The findings focus on marginal electricity costs and exclude total cost of ownership factors like hardware amortization or idle draw.
>"I Measured Every Watt on Apple Silicon Five models, sustained generation, real wall-socket energy at $0.31/kWh — and the surprise the RTX-3090 numbers predicted, only bigger."
Justin Stewart writes about how the energy cost of running local Large Language Models (LLMs) on Apple Silicon depends more on throughput than parameter count. Using an M3 Ultra Mac Studio, he demonstrates that large Mixture-of-Experts (MoE) models can be significantly cheaper to operate per token than smaller dense models because they only activate a fraction of their parameters during generation. Ultimately, the study reveals that efficiency is driven by how much data must be moved from memory for every token produced.
* The measurements were calibrated against actual wall power using a Shelly Plug US Gen4 meter.
* A custom tool called TokenWatt was used to measure marginal energy consumption via Apple’s IOReport interface.
* In real-world "lumpy" traffic scenarios, the cost of dense models compared to MoE models actually widens even further.
Justin Stewart writes about how the energy cost of running local Large Language Models (LLMs) on Apple Silicon depends more on throughput than parameter count. Using an M3 Ultra Mac Studio, he demonstrates that large Mixture-of-Experts (MoE) models can be significantly cheaper to operate per token than smaller dense models because they only activate a fraction of their parameters during generation. Ultimately, the study reveals that efficiency is driven by how much data must be moved from memory for every token produced.
* The measurements were calibrated against actual wall power using a Shelly Plug US Gen4 meter.
* A custom tool called TokenWatt was used to measure marginal energy consumption via Apple’s IOReport interface.
* In real-world "lumpy" traffic scenarios, the cost of dense models compared to MoE models actually widens even further.
E2B is an open-source infrastructure designed to run AI-generated code in secure, isolated cloud sandboxes. It provides JavaScript and Python SDKs for starting and controlling these environments, as well as a specialized Code Interpreter SDK for advanced code execution tasks. The project supports self-hosting on AWS and Google Cloud via Terraform and is built to support enterprise-grade agents.
