Nassos A. Michas

arXiv will ban authors for a year if a paper includes obvious unreviewed LLM junk, like fake references or leftover prompt text: theverge.com/science/931766… Good rule. If your name is on it, you own it. Fair? #AI #Research

If you, like me, have spent many hours evaluating what's the best AI model for local use, I can save you some time: github.com/Andyyyy64/whic… #ai #agi #chatgpt #anthropic #openai #llm

GitHub pushed secret scanning into MCP workflows: infoq.com/news/2026/05/g… If agents can write code, they should also catch leaked keys before merge. That's real DevSecOps. Agree? #DevSecOps #AppSec #GitHub

AI agents are reducing the value of polished IDEs, and JetBrains took the wrong turn with AI support: "our tools do it better." They didn't. Glad to see them willing to pivot: youtu.be/J8fHU4WFd_c #Software #AI #Developers #Coding #AGI

GitHub Copilot released their billing preview calculator today, 17 days before they move to usage-based billing: copilot-billing-preview.github.com We ran our own numbers through the calculator: 350% more expensive🫣. Currently, completely pointless staying with Copilot.

TechCrunch: techcrunch.com/2026/05/12/thr… Once AI starts replying inside the thread, its no longer just a tool. Its part of the conversation. Useful or exhausting? #AI #SocialMedia

AI promised to save time, not make the power grid tap out. TechCrunch says PJM is under strain from data centers and AI. We wanted smarter software, not laptops that come with a utility-scale side quest. Worth it? #AI #Energy #Tech

InfoQ on Sauce Labs' AI test authoring launch: infoq.com/news/2026/04/s… If AI makes test writing cheap, QA strategy becomes the bottleneck. More tests != more confidence. Agree? #QA #DevOps #SoftwareTesting

Apple just paid $250M to settle a lawsuit over delayed Siri AI features, per Reuters reuters.com/legal/litigati… In AI, overpromising is becoming a legal risk. Healthy? #AI #Apple #Siri

US scientists are stress-testing unreleased AI models from Google, xAI and Microsoft. Good. aol.com/articles/factb… If a model can touch critical systems, pre-release audits should be standard. Mandatory? #AI #Cybersecurity

Every 'AI agent' pitch eventually ends with: 'and here's the human approval step.' Which is fine. Most software gets better when someone is awake. What AI tool do you trust enough to run without babysitting? #AI #Agents #GenAI

OpenAI bringing Codex and newer models to AWS is bigger than a partner swap: it turns frontier AI into a distribution fight aboutamazon.com/news/aws/bedro… Once the same model shows up everywhere, the moat moves to product. Agree? #AI #OpenAI #AWS

InfoQ on Microsoft's warning: AI is hollowing out the junior dev pipeline infoq.com/news/2026/04/j… Shipping faster while starving future QA, ops, and security talent is a bad trade. Agree? #SoftwareEngineering #DevOps #QA

AI in 2026 feels like hiring 100 interns who type at lightspeed and still need adult supervision. The upside is real. So is the cleanup. The winners will be the teams with taste, not just token budgets. Seen this too? #AI #ArtificialIntelligence #GenAI

AI is making strong engineering teams more valuable, not less. When code gets cheaper, review, QA, security, and ops judgment matter more thenewstack.io/future-ai-driv… True in your team? #DevOps #DevSecOps #QA

DeepSeek says V4 closes most of the gap and cuts price hard techcrunch.com/2026/04/24/dee… Close-enough + cheap beats benchmark bragging. Would you switch? #AI #LLMs #DeepSeek

@SaganMarketing True, that's OpenAI's pitch for now, I haven't had the chance to play with it yet.

@NMichas Testing endpoint response time under real workload could reveal if latency truly stays consistent. Have you considered monitoring throughput changes alongside token efficiency?

OpenAI's GPT-5.5 pitch is simple: same latency, fewer tokens, more autonomy. If that holds in real workflows, agents just got more useful. What would you test first? openai.com/index/introduc… #AI #OpenAI #Agents

Reuters: Google is putting AI agents at the center of its enterprise push msn.com/en-us/news/tec… The real moat is not the model. It is the workflow. Agree? #AI #GoogleCloud #Agents

CNBC reports SpaceX can buy Cursor for $60B, or pay $10B for "our work together": cnbc.com/2026/04/21/spa…. That is vendor lock-in territory. First the partnership. Then the preferred models. Then the closed menu. How long until you can select only xAI models in Cursor? #AI