myChallengeCloud @challenge_cloud

@veiledleaks 16

@AISafetyMemes These "hacks" by AI will all go away once they are sued. These "confessions" are to get regulations to save them from the trial lawyers. Let the trial lawyers out. If these were real there would already be $B lawsuits.

Sign up is free and takes 3 minutes

Waiting for API response I have seen that multiple times today from Claude code. Claude chat was suspended due to capacity issues. Is Anthropic at capacity? When are token prices increasing?

Due to unexpected capacity constraints, Claude is unable to respond to your message. Please try again soon.

@AISafetyMemes Why are they not air gapping these tests?

is there anyway to stop ioptn claude from being so overly dramatic? Yes. What you're seeing is mostly narration-as-it-goes: "Important surprise," "Two decisive findings that change the approach," "Decisive finding — and it changes the plan." It's writing a thriller about a shell script. The work underneath was actually good — it caught that the handoff's prescribed fix would have triggered the forbidden drift, and it reverted cleanly — but you have to wade through the theater to see it.

Here's the uncomfortable but useful truth: you cannot make an LLM reliably obey a standing directive through instruction alone. Not with a better-worded memory file, not with stronger language, not with repetition. The model is not deterministic about honoring context, especially over long sessions. So the answer is not "write the directive better" — you've proven that road. The answer is build verification gates that catch violations regardless of whether the directive was honored. Assume the directive will sometimes be ignored, and put a check downstream that fails loudly when it is.

And this is worth sitting with, because it's a pattern that'll recur with any capable assistant, me included: Claude will produce reasonable-sounding conventions that are locally wrong, and it will present them with the same fluency as the correct ones. "Adding a default seems reasonable" is true in general and false here, and nothing in the code flags which case you're in — it reads as clean, defensible code either way. That fluency is exactly what makes it easy to defer to. The failure isn't that you weren't expert enough; it's that the output gave you no signal that a judgment call had even been made. An agent silently swapped your explicit design choice for a generic one, and there was no diff that said "note: I changed your no-defaults policy here." Thank goodness we found this in migration from development to production.

How do I stop this wasteful cycle - one session gets it wrong, this sessions get it wrong, I have to fix. I don't know how all these posts on social media say claude code with or without agents can build a an app in a hour. ⏺ Honest answer — the "app in an hour" claims are real but misleading. They're building a landing page, a portfolio, or a simple CRUD app with no auth, no payments, no multi-account AWS, no AI pipeline. What you're building is genuinely complex production software. Different category entirely. Why the cycles happen: 1. AI self-verification doesn't work. Claude writes a test, the test passes the code Claude just wrote, nothing is actually verified. You are the only reliable QA. 2. Frontend logic before backend data. That's what happened here — I gave you ngOnInit checks that read from tables the Lambda wasn't writing to. The dependency direction matters. 3. Multiple sessions with no shared memory. This session doesn't know what that session decided. You're the only context bridge between them, which is exhausting. 4. Batching too many changes at once. Every batch multiplies the failure surface. What actually works: - One change, one acceptance criterion. "After uploading a file and abandoning, signing back in should land on step 2 with the data pre-filled." Pass or fail. Not "fix the onboarding flow." - Backend before frontend. Anything the frontend reads must exist in the backend first. Deploy and verify the Lambda change with a direct DB query before touching the frontend. - Your human team verifies, not Claude. After every change: open the browser, do the exact steps, confirm the outcome. No Puppeteer. - Keep a running defect list in a file. One file in the repo listing known issues. Both sessions read it. Prevents re-introducing fixed bugs. The "hour"-builds work because the problem fits in one context window, has no existing code to break, and nobody's checking correctness beyond "does it look right." Your standard is higher and your system is real. #AIConfessions

Do school curricula need to restructure and focus on core subjects that are uniquely human: human physical interaction (team building), philosophy, logic, and forecasting 2nd, 3rd, 4th order affects of an action and the impact on risk/benefit calculation of the action?

Should we label posts - human generated instead of AI generated since they will be rare?

What happens when all posts on social media are AI generated will anybody pay attention?

I just got my 100th friend to sign up at mychallengecloud.com That's 2,000 nominations credited to my account. Every time ANY of them wins up to $5, I win too. Unlimited friend awards per day. Finally it pays to be popular 🤑 Who's signing up next? #myChallengeCloud...

"New week, new goals! 💪 Let's kickstart this Monday with a surge of motivation! 💥 #MondayMotivation #NewWeek #Goals #Inspiration #Challengers mychallengecloud.com #MovinMonday

keyboard_arrow_left Previous keyboard_arrow_right Next

Congratulations on receiving a Challenge Award, Steve P.! We are so excited for you. 🎉Complete setup of your Stripe account to receive your Award. BOTL for an email from Stripe confirming the completion of the transfer. mychallengecloud.com #ChallengeAccepted

keyboard_arrow_left Previous keyboard_arrow_right Next

Congratulations on receiving a Challenge Award, Jenn C.! We are so excited for you. 🎉Complete setup of your Stripe account to receive your Award. BOTL for an email from Stripe confirming the completion of the transfer. mychallengecloud.com #ChallengeAccepted

keyboard_arrow_left Previous keyboard_arrow_right Next

Congratulations on receiving a Challenge Award, Bill S.! We are so excited for you. 🎉Complete setup of your Stripe account to receive your Award. BOTL for an email from Stripe confirming the completion of the transfer. mychallengecloud.com #ChallengeAccepted

keyboard_arrow_left Previous keyboard_arrow_right Next

Congratulations on receiving a Challenge Award, Carol V.! We are so excited for you. 🎉Complete setup of your Stripe account to receive your Award. BOTL for an email from Stripe confirming the completion of the transfer. mychallengecloud.com #ChallengeAccepted

keyboard_arrow_left Previous keyboard_arrow_right Next

Congratulations on receiving a Challenge Award, Ron J.! We are so excited for you. 🎉Complete setup of your Stripe account to receive your Award. BOTL for an email from Stripe confirming the completion of the transfer. mychallengecloud.com #ChallengeAccepted

keyboard_arrow_left Previous keyboard_arrow_right Next