LLM
10 articles about "LLM".
Letting an AI assistant act without overreaching: action cards, a confirm gate and risk levels
In the upcoming MindThread 2.0 (on our preview site for now), we let Xiaomai, the in-app assistant, schedule posts, change posting times, toggle cloud patrol and pause accounts for members. The model never touches a tool: rules pick the intent, the model only fills card fields, the server stores the card, every card needs a click, high-risk ones need two, state is re-read before running, and everything is logged. Includes a confirm gate you can adapt.
Making an AI assistant quote only real numbers: the faithfulness check behind our in-app helper
The worst thing an in-app AI assistant can do is get a number wrong. For Xiaomai, the helper in MindThread 2.0, we built four layers: the model never gets tools, data is precomputed, every number and quote in the answer is checked against the data, and anything that does not match triggers one retry, then a plain "not found". Includes a ready-to-use answerIssues function.
AI-written outreach should read like the recipient wrote it: the quality bar we set for samples
We are designing a 'send a sample first' outreach flow: three Threads drafts written with MindThread in the recipient's own voice, or three findings from their website. The first version averaged 5.7 out of 10. This post covers the four gates we set: how we build the brand brief, computing voice with code, a source for every factual sentence, and a reviewer playing the business owner who scores the email, with anything at 7 or below rewritten. We have not sent a single email; this is a process design. Voice stats and overlap check code included.
When the model deprecation notice lands: how we split the gemini-2.5-flash migration into four steps
Google says gemini-2.5-flash shuts down on 2026-12-01 and requests will be redirected to gemini-3.5-flash-lite. It sounds like nothing to do, but when we called 3.5-flash-lite directly with a thinkingBudget, we got a 400. We split the migration into four steps: inventory, a tested parameter matrix, per-feature switching through config, and a quality comparison on real inputs. Along the way we found two links in our fallback chain that had been broken for a while. Probe and switching code included.
What Is GPT-6 Astra? Capabilities, Pricing, API Migration, and Five Safety Numbers to Read Before You Hand It an Agent
What is GPT-6 Astra? OpenAI's September 2026 model: pricing, API migration from GPT-5.6, and five system card numbers that matter before you give it an agent.
Prompt Injection Isn't Your Biggest Risk: 517 AI System Prompts Scanned, 11 Undefended Attack Vectors
Prompt Injection is 1 of 12 LLM attack vectors. Ultra Lab's UltraProbe scan of 517 prompts (sample unpublished) found 3.2 of 12 defended. Examples and fixes.
The Real Fix for AI Tech Debt: Don't Use Less AI, Limit Its Scope
A Dev.to article says AI is creating tech debt nobody talks about. We agree, but instead of using less AI, we redesigned our architecture: AI handles content strategy, humans build the system. Here's how we rebuilt UltraSite v2.
We Open-Sourced Our Prompt Defense Scanner: 200 Lines of Regex That Replace an LLM
Most AI security tools use LLMs to check LLMs. We built a deterministic prompt defense scanner: pure regex, under 1ms, zero cost (12 defense vectors at launch, 17 today). Here's why regex beats AI for this job, and how you can use it today.
How We Defend AI Against Comment Attacks: 5-Layer Prompt Defense in Production
When your AI auto-replies to hundreds of comments daily, Prompt Injection isn't theoretical: it's happening every day. This is the 5-layer defense architecture we validated across 27 accounts.
UltraProbe, a Free AI Security Scanner, Is Live: Find Your LLM App's Vulnerabilities in 5 Seconds
90% of AI systems are vulnerable to Prompt Injection, yet most developers have no idea. Ultra Lab launches the completely free UltraProbe, mapped to four OWASP LLM Top 10 (2025) categories: LLM01, LLM02, LLM05 and LLM10, making AI security testing accessible to everyone, not just enterprises.