AI Slop Just Won $25,000 from DeepMind — The Trust Crisis Hitting AI-Powered No-Code Tools
A Kaggle competition entry described as 'blatant AI slop' just won a $25,000 DeepMind Grand Prize. The 'Measuring Progress Toward AGI' competition was meant to evaluate genuine AI reasoning — and AI-generated submissions passed as human-quality work. This is the perfect metaphor for the AI trust crisis no-code builders face.
Table of Contents
Here's a sentence I never thought I'd write: a competition designed to measure progress toward artificial general intelligence just awarded one of its top prizes to AI-generated work. A Kaggle entrant submitted what the community is calling "blatant AI slop" to Google DeepMind's "Measuring Progress Toward AGI" hackathon, and walked away with a $25,000 grand prize.
You can't make this up. And if you build software with no-code tools, you shouldn't laugh it off either.
What Actually Happened
In March, Google DeepMind and Kaggle launched a $200,000 hackathon: "Measuring Progress Toward AGI: Cognitive Abilities." Build benchmarks that test AI models on learning, metacognition, attention, executive functions, and social cognition. Four $25,000 grand prizes, plus $100,000 in track awards.
Results landed 1 June. One grand prize went to Eros GenAI for "GAUGE," a benchmark measuring whether AI models recognise their own uncertainty. Ironic, given what followed.
A separate entry, one the Hacker News community describes as transparently AI-generated, also won. The thread hit 385 upvotes and 224 comments within hours. The complaint isn't just that AI slop won $25,000. It's that a contest designed to evaluate whether AI can think couldn't tell the difference between machine-generated filler and thoughtful human research.
AI evaluating AI. What could go wrong?
The Trust Problem Is Bigger Than Kaggle
The Kaggle incident is darkly funny. But it's also the canary in a coal mine. The same dynamic plays out every day inside the tools you're using to build software.
Three numbers worth losing sleep over:
60%. The proportion of AI search results that cite wrong sources, per the Columbia Journalism Review's Tow Center study.
25-45%. Depending on which study you pick, Veracode, Cloud Security Alliance, or Endor Labs, that's the share of AI-generated code containing security vulnerabilities.
$1 trillion. Alex Reisner's figure in The Atlantic for what he calls generative AI's "trillion-dollar engineering disaster."
Now connect those dots to your workflow. If a Google DeepMind-backed competition can't spot AI slop, what's catching the bugs, security holes, and hallucinated business logic in your AI-generated app?
Nothing. The answer is nothing.
Why This Hits No-Code Harder
Traditional software development has guardrails. Code review. Testing pipelines. CI/CD that catches regressions. A senior engineer squinting at a PR and saying "hang on, that doesn't look right."
No-code builders, especially the new wave of AI-first no-code builders, often have none of that. You prompt. The AI generates. You deploy. The entire review step is glancing at a preview and thinking "yeah, looks about right."
The Kaggle situation exposes the structural issue: AI evaluation itself is broken. When AI-generated work wins a prize in a contest designed to evaluate AI quality, the evaluator is compromised.
The Governance Gap
This is where the distinction between raw AI code generation and governed no-code platforms starts to matter. Cursor, Bolt, and Lovable give you raw output — code, components, logic that you own completely, flaws and all.
Platforms like Stacker, Bubble, and Webflow are architecturally different. They let you configure governed components, permission rules, validation logic, data access controls, audit trails, within a framework that enforces constraints by design.
If Cursor generates a login page with a broken authentication flow, you might not catch it until someone's data leaks. If Stacker's permission system governs who sees what, there isn't a code path that accidentally exposes the wrong data.
Where AI Belongs in Your Stack (and Where It Doesn't)
The Kaggle farce doesn't mean you should stop using AI. But you need a framework:
UI generation: fine. Let AI build layouts. If it's wrong, it looks ugly and you'll notice immediately.
Business logic: with caution, always reviewed. AI nails the happy path and bombs the edge cases.
Data model: human-designed, AI-assisted. AI doesn't know which fields are PII.
Permissions and access control: never AI-generated without human sign-off. If an AI hallucinates a permission rule, people's data gets exposed.
The Punchline That Isn't Funny Any More
The Kaggle prize is funny. It is properly, absurdly funny that AI-generated work won a competition designed to measure AI progress.
But it stops being funny the moment your AI-built app exposes customer data. Or charges the wrong amount. Or silently corrupts your database over six months because of a logic error nobody reviewed.
Governed platforms — Stacker, Bubble, Webflow — aren't just an alternative to AI code generation. They're the missing governance layer that AI generation desperately needs. They're the human in the loop, baked into the architecture.
If you're building business software with AI today, ask yourself one question: who reviewed the last AI-generated thing you deployed?
If the answer is "nobody," you've got a Kaggle situation waiting to happen. And unlike the hackathon winners, you won't get $25,000 when it does.
Want to read
more articles
like these?
Become a NoCode Member and get access to our community, discounts and - of course - our latest articles delivered straight to your inbox twice a month!
