Would you be interested if JetBrains releases a totally local AI agent, working 100% on your laptop, using our code insight engine and deeply integrated into the IDE?
Yes, it will be probably 1 month behind the very recent frontier models, but no token blood bath anymore
WDYT?
We’re recruiting for a Data Engineer to join our Research team at Liverpool FC. In my (perhaps biased) view, it’s a great opportunity to join an excellent team, work at the forefront of sports analytics, and have a tangible impact. https://t.co/JlFxXCUIsS
@juliendubois Yes, very much understand. I'm just wary of instructions that some models can take too literally. Instead of the negative/restrictive : "Do NOT ... etc" I'd recommend a firm positive instruction: "The user must review these changes before they are committed to git", (or similar).
Tip for the Google Gemini team: if you want to help Google truly get ahead in the AI era, use your hefty influence to get it so setting up API access to your own calendar doesn't involve THESE steps
Devious new attack class defined by @wunderwuzzi23: "Cross-Agent Privilege Escalation"
If someone has both Claude Code and GitHub Copilot on their machine you can prompt inject one and have it edit the config for the other in order to further escalate your attack's privileges!
LLMs love XML — the most instructive of configuration languages. Leaked AI-startup system prompts all use it, as does The Universal Jailbreak from @hiddenlayersec https://t.co/8KPytxMbhf
I updated https://t.co/KH5LcVaCJn to use XML tags in the persona instructions after a *quite interesting* conversation about reinforcing instructions to LLMs with @sublimino last week