dragosr

YouTube video by PromptorGTFO Dragos Ruiu - Multi-Model AI Orchestration and Prompt Leakage | Prompt||GTFO #1

A Multi-LLM MCP Orchestration Tool I'm working on, and some LLM safety testing I pulled together for Gadi's fun "Prompt||GTFO" presentations.
(3-5 min, real demos only, no slides)

The channel and events offer glimpses into "edgy" LLM security apps/projects. www.youtube.com/@PromptorGTFO

OpenBSD -current has a new driver for AMD machines, amdpmc(4), similar to intelpmc(4), for power management control needed to enter even lower power idle states during suspend-to-idle (S0i), which #OpenBSD has began supporting since 7.6.

Stefan Sperling (stsp@) has added initial 802.11n support to the #OpenBSD qwx(4) driver for Qualcomm ath11k wireless devices.

marc.info?l=openbsd-cv...

qwx(4) is found on machines like the ThinkPad x13s.

My parents hid in basements and forests while various armies rolled through and all the sides took turns bombing them. And had lots of problems getting food. We forget that evil that is war - unless you have friends in Ukraine or another modern war zone. It's horrible.

A custom-built quadcopter drone with four large propellers is resting on a wooden slatted bench. The drone is secured to a black TX16S remote controller using a black "FPV DONE" battery strap and several blue zip ties. The controller features dual joysticks, various switches, and a built-in screen, with an attached FPV monitor mounted above it. The setting appears to be an indoor or semi-outdoor lounge area with stone tile flooring and wooden furniture.

They are dropping lots of new Battlefield 6 preview videos. I note that it seems so ancient and dated now. Last century combat - too many obsolete tanks, not enough FPVs.

Also if your server panics at a few extra visitors a day, maybe it's trying to tell you it's time to graduate from dial-up on that raspberry-pi in your basement.

Since more and more web search is moving to "Web Search" modes on LLM queries, blocking LLM spiders is just shooting your own site visibility in the foot. Assuming those are machine traffic is an archaic view - you're blocking the humans behind queries. Go for it if you want no one to see your site.

Conversely tell me more about haven't you haven't learned to use modern tools.

i think you're holding it wrong then.

we not ever reach AGI but the advanced models are definitely already more clever than almost all humans on almost all topics.

Should We Respect LLMs? A Cross-Lingual Study on the Influence of Prompt Politeness on LLM Performance We investigate the impact of politeness levels in prompts on the performance of large language models (LLMs). Polite language in human communications often garners more compliance and effectiveness, w...

You get better results if you are polite, just like humans.

arxiv.org/abs/2402.14531

Photo of a Nanboku Line 乗り換え 出口案内 (transfer and exit guide) on a station platform.

The other day, I was telling friends and family how a housewife originated these Tokyo Metro transfer and exit guides, and they thought I was kidding. It used to be a well-known story. If you don't know it, here's a quick summary.

Illustration of the diffusion process: The "forward process" row shows a clear image of a gray tabby cat progressively becoming noisier through several steps until it is completely indistinct. The "reverse process" row begins with noise and progressively resolves into a distinctly different cream-colored cat, illustrating that the diffusion model generates a new, unique image rather than reconstructing the original input.

Diffusion models don't store images—just learned statistical patterns (textures, edges, color co-occurrences) in neural weights. Outputs are stochastic, unpredictable samples guided from noise; exact retrieval of training images isn't feasible, as no explicit copies exist.

A large, intimidating mech warrior with satellite-dish-like mouse ears and a helmet reminiscent of Darth Vader confronts a smaller, white humanoid robot wielding lightsabers in a courtroom setting, symbolizing a legal battle. The courtroom backdrop features wooden paneling and an indistinct audience observing the conflict.

Disney v. Midjourney: Until we assign copyright ownership to AI image-tool providers, we can't reasonably assign infringement liability either. Like pencil, paint, or photo editing software makers, tool providers remain neutral—liability belongs to users who commercially misuse their outputs.

Remember when people thought "prompt engineering" was going to be a career?

Turns out it will be a basic life skill everyone will need to learn, like using a search engine, or entering your wifi access point password.

US railroad industry's outdate radio protocol is vulnerable : Neil Smith has been trying to get the railroad industry to listen since 2012, but it took a CISA warning to get there

I am so glad Neil is finally getting some folks to listen. The bar for attackers is far too low here. Trivially spoofed packets from a SDR can remotely derail trains.

That works in the other direction too, making code from those other APIs fail on Gemini. It should at least accept both.

Same deal with "assistant" vs "model."

It's a minor but irksome little detail that Google's Gemini API expects a different message format than OpenAI/Anthropic. Google expects messages with a "parts" key instead of "content."

Why? Just to make cross platform code annoying? To hinder porting codebases over to other providers?

Patience, thoroughness, meticulous specification, and attention to detail are predictors of LLM code success.

Novice hand-wavy, hey read my mind as I give you an underspecified brief sentence describing code goal - what novices usually start with, will of course take longer, with poorer results.

But that's not what the LLM hype spinners that claim we won't need coders anymore and LLM "vibe coders" want to hear. They want a shortcut.

But you still have to do the work. No cheat code.
Experienced _LLM_ coders can build better, more sophisticated code, faster and get huge productivity boosts.

I agree it's interesting about novice LLM coders.
But not just reviewing code. Have several other LLMs review it, multiple times, before first of many human reviews.

It's not a magic wand. With experience, many prompts, much human integration of LLM generated chunks, experienced coders can thrive.

Experienced open source developers - and novice LLM coders. What's wrong with this picture?

Everyone - naively - thinks you can magically get any coder, or worse, non-coders to write great LLM generated code. It takes tons of work and experience before coding well with LLMs. Sorry no cheat code.

Wow, check out this clickbait non-story. You read the whole thing for the buried line that says "but this didn't happen."

Evidence, or go away sleazy hype, and deceptive headlines.

ChatGPT - Enhanced Joe Rogan Filter Shared via ChatGPT

In case anyone is curious about the steps that led to that code: chatgpt.com/share/686982...

Rogan Filter Proxy

Turns out, that's not such a hard problem. I'd do it with a proxy.

This probably has bugs, but will get you started:
chatgpt.com/canvas/share...

from gremllm import Gremllm # Be sure to tell your gremllm what sort of thing it is counter = Gremllm('counter') counter.value = 5 counter.increment() print(counter.value) # 6? print(counter.to_roman_numerals()) # VI?

The is diabolical... a Python object that hallucinates method implementations on demand any time you call them, using my LLM Python library github.com/awwaiid/grem...

In this era of trivial LLM language translation, this is stupid. A few lines of code could have had those documents served in French instantly.

Braid: Alberta's pro-Canada forces scoop separatists, set referendum question The pro-Canada referendum forces have scooped the separatists and succeeded in setting the question for a referendum on Alberta independence

Big LOL. Happy Canada Day.

The stench of desperation wafts from the fossil fuel propaganda machine, acrid as the fumes they peddle. | Dragos Ruiu The stench of desperation wafts from the fossil fuel propaganda machine, acrid as the fumes they peddle. Their latest fabrication? Transmuting BC Ferries' prudent policy on towing damaged electric...

The petroleum prophets can shriek their warnings and wave their doctored headlines. But you cannot propagandize physics, and you cannot lobby away thermodynamics.

The age of fire is ending.
The age of light has begun.

A story in which the country of Denmark discovers how often face appearance collisions happen and why face recognition is not a very unique biometric.

Administrators should immediately:

meshtastic --factory-reset-device

to purge stored state, flash ≥ v2.6.11, and, for secure deployments, regenerate keys offline:

openssl genpkey -algorithm x25519 -outform DER | tail -c32 | base64