Ross

Who am I to judge your idea? I quit my job for an idea. Stopped bug bounty hunting for it. Stopped talking to friends because of it. Wanted to quit high school for it. Spent all my savings on it. Worked day and night for two years on something that seemed delusional.

To be fair, suing OpenAI does seem like a more viable strategy than pretending Grok 4.3 is any good

When I work all day, every day, I actually make less.

@strix_ai I was thinking you guys use Caido.

We found a zero-authorization vulnerability in an a16z-backed DoD startup that exposed the data of active U.S. military personnel. We tried to report it. They ignored us for 150 days. Here is how our open-source AI agent found the ultimate OPSEC nightmare 🧵👇

There’s a version of me in another timeline who didn’t get into AI early. He’s a bug bounty hunter complaining about AI taking his job instead of learning it.

What’s the point of building features non-stop with AI if only 1% of users will use them? So much time and money gets wasted. And if you’re not tracking usage, it can feel like progress when it’s really just motion. Go outside and touch some grass. You’ll get better ideas.

I believe many would pay extra just to have Kimi 2.5 run faster. Throughput really matters for end users, even if the difference is small.

Kimi 2.6 is seeing 1.5× more usage on HackerAI than 2.5, and it’s now ranked #1 on OpenRouter. I called it. Fire model. Slightly more expensive, but still cheap compared to Claude. Big props to the team behind it.

@lmdev @ReactMiamiConf @FrontendMasters Damn. I should have taken more pictures like you. Just got one with Theo.

miami things happening @ReactMiamiConf

If you don’t achieve your goals, you’ll feel bad that you didn’t achieve them. If you do achieve your goals, you’ll feel bad achieving them didn’t make you feel good.

Kimi 2.6 is fire. Just by the fact it’s 10 TPS faster than Kimi 2.5. Waiting for Cursor to RL it ASAP so they can call it the fastest-cheapest-smartest-best-price-to-intelligence-ratio-in-the-game, aka composer-2.1

Opus 4.7 has the best command chaining and pipe usage of any model I've seen. We just added it to HackerAI, and no, you won’t get blocked or asked to verify for asking cybersecurity questions with us :)

Whatever you do, don’t be the guy who gives up at the exact moment when you should be fighting with everything you have. You’ll make it through either way, but there’s only one way you’ll look back and be proud of yourself.

I've been waiting for this for a few years already. Sadly, you have to provide your ID info and let them read your chats just to ask cybersecurity questions.

I find it ironic that people who once dismissed AI finding “real” vulnerabilities are now so worried about it, as if no one saw it coming. This was inevitable. As AI gets better at writing code, it also gets better at finding ways to hack that code. It’s not magic, it's data.

It would be nice to get access to Claude Mythos Preview model. Seems very good at hacking. Unfortunately, it can't happen in this universe. www-cdn.anthropic.com/53566bf5440a10…

You can always work more. You can always make more. You can always find more excuses to avoid doing X.

Too much hassle keeping customers happy with session and weekly limits, especially when you don't create your own models. Really glad we switched to usage-based billing. It's wild to see so many people complaining and angry about Claude's limits. Instead of getting $5,000 of model usage, you'll now get less, even if it's 10 times less, you still get $500 worth of model usage for $200.

Adding Codex to HackerAI sounds great, but unfortunately, nobody is using it. There are always many cool ideas that, once you implement them, you realize no one really cares. You just have to accept it and move forward.