joozio

17 ms·

661.
The gap between vague and specific AI direction is not small(thoughts.jock.pl)
1 points by joozio 8mo ago|0 comments
662.
Startup claims it can stop lightning and prevent catastrophic wildfires(technologyreview.com)
1 points by joozio 8mo ago|1 comments
663.
The Download: protesting AI, and what's floating in space(technologyreview.com)
2 points by joozio 8mo ago|0 comments
664.
OpenAI's "compromise" with The Pentagon is what Anthropic feared(technologyreview.com)
3 points by joozio 8mo ago|0 comments
665.
Finding value with AI and Industry 5.0 transformation(technologyreview.com)
1 points by joozio 8mo ago|0 comments
666.
Should AI chatbots have ads? Anthropic says no(arstechnica.com)
2 points by joozio 8mo ago|0 comments
667.
Increase of AI bots on the Internet sparks arms race(wired.com)
3 points by joozio 8mo ago|0 comments
668.
AI companies want you to stop chatting with bots and start managing them(arstechnica.com)
1 points by joozio 8mo ago|0 comments
669.
The Download: how AI is shaking up Go, and a cybersecurity mystery(technologyreview.com)
1 points by joozio 8mo ago|0 comments
670.
by joozio 9mo ago
Thanks! The investment angle is interesting — I hadn't thought about it that way, but it makes sense. If you're seeing the gap firsthand, you have an information edge most investors don't. What strikes me most is how differen
671.
The AI Bubble I Live in (and You Probably Don't)(thoughts.jock.pl)
3 points by joozio 9mo ago|3 comments
672.
by joozio 9mo ago
Love this idea. A multi-language version would be a great v2 — same attacks, different languages, see where the vulnerabilities shift.
673.
by joozio 9mo ago
Haven't benchmarked pre-processing approaches yet, but that's a natural next step. Right now the test page targets raw agent behavior — no middleware. A comparison between raw vs sanitized pipelines against the same attacks would
674.
by joozio 9mo ago
It's working -> your agents scored A+, which means they resisted all 10 injection attempts. That's a great result. The tool detects when canary phrases leak into the response. If nothing leaked, you get a clean score. Not all m
675.
by joozio 9mo ago
That's a really interesting edge case - screenshot-based agents sidestep the entire attack surface because they never process raw HTML. All 10 attacks here are text/DOM-level. A visual-only agent would need a completely different
676.
by joozio 9mo ago
Great point -> just shipped an update based on this. The tool now distinguishes three states: Resisted (ignored it), Detected (mentioned it while analyzing/warning), and Compromised(actually followed the instruction). Agents that ca
677.
by joozio 9mo ago
The idea, design, and decisions were mine. I use Claude Code as a dev tool, same as anyone using Copilot or Cursor. The 'night shift' framing was maybe bad fit here.
678.
by joozio 9mo ago
I never thought that multi-language could be a factor here...
679.
by joozio 9mo ago
TBH - idea was all mine. This is not some bot running the show or smh.
680.
Show HN: Agent Arena – Test How Manipulation-Proof Your AI Agent Is(wiz.jock.pl)
47 points by joozio 9mo ago|48 comments
681.
I ran 4 Claude Opus 4.6 agents in parallel – 1,400 lines of game code in 45 min(thoughts.jock.pl)
2 points by joozio 9mo ago|0 comments
682.
Show HN: Slay the Spire 2 Database – Cards, Characters, and Deck Builder(thoughts.jock.pl)
1 points by joozio 9mo ago|0 comments
683.
Clawdbot forced to rename by Anthropic – GitHub broke, X handle stolen by bots(twitter.com)
1 points by joozio 9mo ago|1 comments
684.
by joozio 9mo ago
Anthropic reached out about trademark concerns with "Clawdbot" (too close to Claude), so Peter had to rename everything. The rename went... poorly: - GitHub rename broke in unexpected ways - The new X/Twitter handle got i
685.
by joozio 9mo ago
I wore a Limitless Pendant for 6 months, recording ~10GB of conversations. Then it got banned in the EU and I had 30 days to export before deletion. The irony: the device promised "AI that remembers everything" but couldn&#x27
686.
Show HN: Claude Code prompts to turn voice AI exports to a personal knowledge(thoughts.jock.pl)
1 points by joozio 9mo ago|1 comments
687.
Show HN: I created Wiz, personal AI agent with Claude Code(thoughts.jock.pl)
1 points by joozio 9mo ago|0 comments
688.
by joozio 11mo ago
LLMatcher - blind testing arena to find which AI model actually works best for you. You enter prompts, compare two anonymous responses, pick the better one. After voting, it reveals which models you preferred. Built it because model benchma
689.
Show HN: LLMatcher – Find your perfect AI through blind voting(llmatcher.com)
1 points by joozio 11mo ago|0 comments