HACKERSCHOICE

it really depends on your goals, but web red teaming works perfectly fine and i think defense will lose after 2 years of productive so called "prompt engineering". its a question of your personal time, if you have the time to always compare the (almost daily released) different models. but you dont have to spend a single cent to use ai that is, at least, capable of doing stuff what the pr shit companys like asstrophic and closedai are telling us theyre working on. basically its a game of catching the most free tokens from either openrouter (https://openrouter.ai/models?max_price=0&q=free&order=newest), kilocode (https://kilo.ai/docs/gateway) and opencode (https://opencode.ai/docs/zen/). you can let bots create api keys and accounts for these gateways. with an nvidia card, you could use cuda, ollama and host uncensored, for example gemma 4 heretic/uncensored, on your local machine. or on a mash of victim mashinces if you use them as ollama nodes and split the load the dirtiest way is to use context7 mcp (https://context7.com/), which is huge collection of github repos and docs use qdrant with a 2048 vect0r size with this model https://openrouter.ai/nvidia/llama-nemotron-embed-vl-1b-v2:free and set up a vscode instance with kilo code extention for full fire capacaties you dont have to use agents, its bullshit. give your llm (less guardrails with chinese models like tencent, xiaomi, z-ai and deepseek and stuff) a well written prompt, point to online sources, give it context and a well structured folder and .md file base. you can get aditional mcp servers like https://github.com/CommonHuman-Lab/nyxstrike, https://github.com/SHAdd0WTAka/Zen-Ai-Pentest, or https://github.com/vxcontrol/pentagi. mcp servers are something like addons for llm but also bring skills an llm normally wouldnt have, like python scripts that do api stuff i think it gives you a bright variety of things and stuff you can do with it, which isnt linked to classical hacking. im not into instagram really but llm made it possible for me to trick their photodna automod system and made easy vp8 scripts to buffer videos every 2 seconds into a view-loop. or created a whole android click farm on software basis instead of using real phones (to get around googles integrity) within days instead of months, and so on. i think llm is great if it has full root and/or for creative people, but only to get work faster done. you still have to always observe and test. if you are not creative, and ive seen many, its not for you, they are never getting out of the testing phase, have one or two negative experiences and give up. but i think thats something that seperates us anyways from people that arent interested in knowledge. they dont tinker, but ai is for tinkers. i think and sometimes it gives you really good ideas on the fly... its hard to tell, you face different problems but its addictive just because of the speed you get shit done nowadays. if you dont understand the system, you cant really tell ai what you wanna achive. now hate me