🥁 Scott Gardner

Roo Code lets you mold your #AI coding agent into what you actually need. Brandon Mathis from This Dot Labs built custom modes for documentation, pair programming, writing specs, and improving his prompts. If you're serious about agent workflows (which you have to be), this is worth a read.

13

4 Comments

This user put competing GPT-5, Sonnet 4, Opus 4.1 and Kimi. This is the result and It’s wild! 𝗛𝗲𝗿𝗲’𝘀 𝘁𝗵𝗲 𝗧𝗟;𝗗𝗥 𝗳𝗼𝗿 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝘄𝗼𝗿𝗸: 𝙒𝙞𝙣𝙣𝙚𝙧 𝙛𝙤𝙧 𝙙𝙞𝙨𝙘𝙞𝙥𝙡𝙞𝙣𝙚𝙙 𝙚𝙭𝙚𝙘𝙪𝙩𝙞𝙤𝙣: 𝙆𝙞𝙢𝙞 ▪️Follows instructions fastest. Immediately uses the TypeScript workspace, stumbles on paths 3×, then simplifies and gets it done. ▪️Ships minimal, working code (simple_validate.ts). ▪️Cost/time: ~14 min, ~$1.59. 𝙈𝙤𝙨𝙩 𝙧𝙚𝙡𝙞𝙖𝙗𝙡𝙚 𝙛𝙞𝙧𝙨𝙩 𝙥𝙖𝙨𝙨: 𝙊𝙥𝙪𝙨 4.1 ▪️Jumps into TypeScript by minute 2. ~95% right on first try. ▪️Produces a small, modular framework (3 files) that’s easy to extend. ▪️Cost/time: ~10 min, ~$10.46 (pricier but sturdy). 𝘼𝙙𝙖𝙥𝙩𝙞𝙫𝙚 𝙥𝙧𝙤𝙗𝙡𝙚𝙢-𝙨𝙤𝙡𝙫𝙚𝙧: 𝙎𝙤𝙣𝙣𝙚𝙩 4 ▪️If a tool is missing, it pivots and continues. ▪️Clear separation of concerns (comprehensive_analysis.ts). ▪️Cost/time: ~6 min, ~$1.87. 𝙋𝙤𝙬𝙚𝙧𝙛𝙪𝙡 𝙗𝙪𝙩 𝙘𝙝𝙖𝙤𝙩𝙞𝙘: 𝙂𝙋𝙏-5 ▪️Ignores “use the TS workspace” for 15 minutes; fires off 31 shell commands first. ▪️When commands fail, it expands them (500→2000+ chars)… and keeps going. ▪️Emits odd hybrids (TypeScript file with massive embedded bash). ▪️Cost/time: ~27 min, ~$5.04. 🔎𝗨𝗻𝗶𝗾𝘂𝗲 𝗳𝗶𝗻𝗱𝗶𝗻𝗴𝘀 𝘀𝗻𝗮𝗽𝘀𝗵𝗼𝘁 🔺GPT-5: spots schema/naming quirks (RIDRETH2→RIDRETH3, _J suffix). 🔻Kimi: solid basic validation (SEQN exists, 9,966 participants). 🔹Sonnet: surfaces hidden mental-health data and 1.4M-row files. 🔸Opus: detects designed sparsity and wide column ranges (2–323). 📋 𝗧𝗵𝗲 𝗰𝗼𝗻𝗰𝗹𝘂𝘀𝗶𝗼𝗻 ♟️Use Kimi when you need a steady executor with guardrails. 👌Use Opus for higher-confidence, extensible scaffolds. 🧩Use Sonnet when adaptability matters (missing tools, shifting context). 💀Use GPT-5 for quick brainstorming/debug tricks—not for long, brittle agent runs (yet). Link to the analysis in the comments 👇 #ai #genai #claude #gpt #coding

321

49 Comments

Open-Source Sharing: Local Digital Avatar Solution! 🎉 I’m thrilled to share that we’ve officially open-sourced the Local Digital Avatar project on GitHub: 👉 https://lnkd.in/gpNnbk_V Built as a streamlined FastAPI app, this solution brings together speech-to-text, text-to-speech, lipsync, and LLM-driven conversation all under one roof—no multi-container orchestration required. 💡 Why it matters: By simplifying setup and unifying services, we’re making it easier than ever to experiment with AI avatars—no Docker headaches, just one Python app and Nextjs app. Let’s build the next generation of local, privacy-focused AI avatars—together! 🤝 #OpenSource #AI #FastAPI #DigitalAvatar #QMedAsia #OpenVino

24

4 Comments

🚨 Privacy & AI: the new Halo glasses are always on, record everything, and do NOT have any indicator to warn people that they are being recorded (!). The new generation of AI glasses raises serious privacy concerns: In an interview with TechCrunch, Halo's founders said that in U.S. states in which it's illegal to covertly record conversations without the other person's consent, it's up to the user to obtain consent before using the glasses (!). This is a completely unrealistic scenario, as glass users will likely not ask all potential bystanders to consent before being recorded. Malicious parties who want to covertly record others will benefit from the lack of built-in features (and the lack of a stronger legal framework on AI mandating these features) to exploit others' vulnerabilities. - Halo is not the only company developing AI glasses. Meta has recently announced its Hypernova AI glasses, which include a display and a wristband that allows users to control the glasses with hand gestures (it looks like Zuckerberg is finally making his Metaverse dream come true...). Surprisingly, Meta's AI glasses have built-in features to support bystander privacy, including a "capture LED" that lets others know when they are being recorded, and audio cues emitted when the user takes a photo or records a video. According to the company's website, if the LED light is covered, the user will be notified to clear it. It's important to remember that in any case, a small LED light on the corner of the glasses' frame will be easily ignored, and many people might not understand what it means if it's ON (it might just be part of the glasses' design). Also, in noisy places, the audio cue might be imperceptible. - For now, AI glasses are not a major AI trend yet, and it's unclear if they will ever be (to Zuckerberg's despair...). But they signal a strange trend in which an AI-powered device captures, records, analyzes, and processes everything a person experiences. The age of covert AI surveillance has officially begun. - 👉 NEVER MISS my curation and insights on AI's emerging legal and ethical challenges: join my newsletter's 75,300+ subscribers (link below). 👉 To learn more and go beyond, join the 24th cohort of my AI Governance Training in October (link below).

237

39 Comments

Aflac Data Breach By Scattered Spider Hackers Is No Quacking Matter - Forbes When you hear the name Aflac, you, probably like me, hear the quacking duck from their commercials. Unfortunately, however the recently announced data breach at Aflac is no quacking matter. Aflac disclosed on June 20th that it had suffered a data breach that may have compromised sensitive personal information held by the company, which offers a wide range of insurance products to millions of people. According to Aflac, it noticed suspicious activity on its networks on June 12th and is now in the early stages of investigating the extent of the data breach with the help of outside cybersecurity experts. Aflac's press release states that it did not find evidence of ransomware, but doesn't yet know the extent of the data breach which may include social security numbers and other sensitive information. It is believed that the data breach was the work of the infamous hacking group called Scattered Spider which focuses its efforts on one specific industry at a time, often using ransomware. The September 2023 ransomware attacks on MGM Resorts and Caesars Entertainment were attributed to Scattered Spider. Now, according to the Google Threat Intelligence Group, Scattered Spider is targeting the insurance industry. Earlier this month Erie Insurance suffered a data breach attributed to Scattered Spider. Google Threat Intelligence Group chief analyst John Hulquist warned “Given this actor’s history of focusing on a sector at a time, the insurance industry should be on high alert, especially for social engineering schemes, which target their help desks and call centers.” Scattered Spider is thought to be made up of English-speaking Americans and British hackers. In 2024 four Americans and one British national were indicted on cybercriminal charges related to activities of Scattered Spider. #cybersecurity #AFLAC #breach #ScatteredSpider

94

33 Comments

𝗜 𝗮𝗱𝗺𝗶𝗿𝗲 𝗳𝗼𝘂𝗻𝗱𝗲𝗿𝘀 𝗹𝗶𝗸𝗲 𝗖𝗮𝗿𝗼𝗹𝗶𝗻𝗲. A healthtech founder who doesn’t play with patient trust. She doesn’t want to hear that her S3 bucket leaked sensitive patient logs with emails, credit card info, or medical records. Not when she’s pitching to investors or applying for HIPAA certification. In the video below, I walked you through how I helped Caroline, founder of an AI health startup (𝘁𝗵𝗮𝘁 𝗯𝘂𝗶𝗹𝗱𝘀 𝗟𝗟𝗠-𝗽𝗼𝘄𝗲𝗿𝗲𝗱 𝘁𝗼𝗼𝗹𝘀 𝗳𝗼𝗿 𝗿𝗲𝗺𝗼𝘁𝗲 𝗱𝗶𝗮𝗴𝗻𝗼𝘀𝘁𝗶𝗰𝘀 𝗮𝗻𝗱 𝘀𝘁𝗼𝗿𝗲𝘀 𝘁𝗵𝗼𝘂𝘀𝗮𝗻𝗱𝘀 𝗼𝗳 𝗽𝗮𝘁𝗶𝗲𝗻𝘁 𝗶𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻 𝗹𝗼𝗴𝘀 𝗶𝗻 𝗔𝗺𝗮𝘇𝗼𝗻 𝗦𝟯) create an automated, real-time detection and response system for PII exposure using AWS services completely serverless, without writing a single line of infrastructure code. From Macie detecting sensitive data → EventBridge triggering a Lambda → to Lambda moving the exposed file to a secure bucket and alerting the team instantly with SNS → CloudWatch logging the event for monitoring and compliance purposes. This setup cut their manual audits, cut response time to under 2 minutes, and added a layer of HIPAA readiness and gave them investor-level confidence in minutes. You can watch the full video here: https://lnkd.in/dhWTHCmJ Access the GitHub repo here: https://lnkd.in/dtWtngFG Do you think this is something you want to adopt for your AI startup? #AIsecurity #AWS #CloudSecurity #Macie #DataLeakPrevention #Cybersecurity #HIPAA #DevSecOps #infosec #HealthcareAI #ServerlessSecurity

13

2 Comments

This is a problem in every industry, not only software development. I've worked on healthcare projects that had doctors copy-paste the same information into their EMR/EHR. When the tool gets in the way of the work it I a no longer a tool but very much a roadblock. —— This is something I watch out for when I docsoftware selections for ERP, CRM, EMR products and much more. Laying attention tovuaer experience is just as important if not more important than the raw features the product has. If no one is using your newly purchased and implemented multi-million dollar system, does it even exist?

17

Every founder I talk to says the same thing. “We just added another platform to fix the problem.” Then three months later, the problem’s worse. The truth is, most businesses aren’t drowning from lack of technology. They’re drowning from having too much of it that doesn’t talk to each other. Think about it. You’ve got a CRM that doesn’t sync with your email platform. Invoices that don’t update your accounting system. Project tools that don’t reflect what your team is actually doing. Every new app promises simplicity. But every disconnected tool adds another step, another login, another way to waste time. By the end of the week, you’re working for your software instead of your business. Here’s how smart companies fix it. 𝐒𝐭𝐞𝐩 𝐨𝐧𝐞, 𝐚𝐮𝐝𝐢𝐭 𝐲𝐨𝐮𝐫 𝐞𝐧𝐭𝐢𝐫𝐞 𝐬𝐭𝐚𝐜𝐤. List every tool you’re paying for. Then ask what problem each one actually solves. If two platforms do the same thing, merge them or cut one. 𝐒𝐭𝐞𝐩 𝐭𝐰𝐨, 𝐜𝐨𝐧𝐧𝐞𝐜𝐭 𝐲𝐨𝐮𝐫 𝐝𝐚𝐭𝐚. Integration isn’t optional anymore. Your CRM, accounting, project management, and analytics should feed the same brain. When data flows freely, decisions get faster. 𝐒𝐭𝐞𝐩 𝐭𝐡𝐫𝐞𝐞, 𝐚𝐮𝐭𝐨𝐦𝐚𝐭𝐞 𝐰𝐡𝐚𝐭 𝐫𝐞𝐩𝐞𝐚𝐭𝐬. If a task shows up more than twice a week, it can be automated. Reports, reminders, updates, follow ups. That’s where your time goes. That’s where your systems should take over. 𝐒𝐭𝐞𝐩 𝐟𝐨𝐮𝐫, 𝐝𝐨𝐜𝐮𝐦𝐞𝐧𝐭 𝐞𝐯𝐞𝐫𝐲𝐭𝐡𝐢𝐧𝐠. A clean process beats a shiny app every time. Your team should know how work moves from start to finish, even if the tools change tomorrow. At Vyudu, we build automation systems that turn chaos into connection. You don’t need more platforms. You need one ecosystem that makes them work together. This is for: → Teams wasting time on tool-hopping → Founders tired of paying for 20 apps but still missing visibility → Businesses who want clarity, not clutter Technology should make you faster, not busier. Integration is the real innovation. #BusinessAutomation #DigitalTransformation

7

8 Comments

"Game dev is the ultimate hardcore rogue-like" You don't know if what you have is good. You can have an idea everyone loves. You can build something that seems to resonate with all the people you hand it to. You can even get lots of media attention, or have great early reviews, or critical acclaim. You can test and try new things and experiment and build a community and pivot and adapt and learn and improve... ...and still create something that ends up being commercially non-viable. It doesn't matter whether you're in the PC space, working on VR, or headed to mobile. It doesn't matter whether you're targeting an eastern or western audience, or both. You won't know until the game has been out for awhile. Because strange stuff happens to games all the time. They start small, then get big. They start big, then vanish. They bump along for a long time before nosediving without explanation, or before skyrocketing. Many drift slowly down without much fanfare. With our "perfect" hindsight, we rationalize all of it, saying, "Well, this is why it worked - the clever marketing technique!" or "Here's where streamers stopped paying attention" or whatever else helps us believe we can make sense of a game dev reality that is beyond our finite ability to comprehend. The environment is chaos, our industry is too multi-variate. We compete with and are amplified by other things that grab attention. The player is not just a player, but a human with a ton of other things going on. Competitors that don't look like competitors, plus the ones that do, are everywhere. So you don't know what will happen. Neither do the VCs or publishers or pundits or players. Sure, some bets are easier to predict, but that's usually because they already have a product market fit that's working... until it isn't. That doesn't mean you shouldn't test, experiment, learn, adapt, and improve. Game dev, like a rogue-like, is all about stacking the deck in your favor as best you can. What exactly happens is mostly outside of your control. Do everything you can to put your thumb on the scale of the "success" side. Do everything you can, and then recognize that a lot of the equation is outside of your control. Repeat that enough times while getting better at it all the way, and you might just make something players love to play and are willing to pay you for. And if you get lucky sooner, please don't let it go to your head. If you do, your next shot will likely be a lot worse.

29

7 Comments

When Claude Opus 4.5 generates an Artifact, it doesn't just render a React app. It incorporates the business, files the LLC paperwork in Delaware, and closes a Series A funding round before the UI finishes rendering.

182

19 Comments