Phishing for Google’s g.co, AI Co-Pilot as Chaos Catalyst, & AI Missing the Mark
♨️ Is the AI revolution losing steam?
Welcome to HackerPulse Dispatch! This edition covers the latest tech buzz, from hackers exploiting g.co with real Google URLs to a side project hilariously derailed by overreliance on AI tools, Devin’s disappointing coding debut, & devs losing their edge to AI.
Here’s what new:
💣 My Afternoon Project Turned Into Four Days of AI Lies, USB Chaos, and Hard Lessons: Discover how Deskthang that started as an exciting hardware/software combo quickly became a crash course in what AI can and can’t do.
😵 The “First AI Software Engineer” Is Bungling the Vast Majority of Tasks It’s Asked to Do: Cognition's AI assistant Devin, touted as the "first AI software engineer," has been criticized for its 15% task success rate.
🤷 AI Is Creating a Generation of Illiterate Programmers: As AI tools like ChatGPT revolutionize coding, developers are facing a growing dependency that undermines their problem-solving skills.
☣️ almost_pwned.md: A phishing attack leveraging Google's official g.co domain nearly compromised a technical user’s account, raising serious concerns about security gaps.
⛔ Tool Touted as ‘First AI Software Engineer’ Is Bad at Its Job, Testers Claim: Devin, an AI software engineer from Cognition AI, has failed to live up to its promises, successfully completing only 3 out of 20 tasks in recent evaluations.
My Afternoon Project Turned Into Four Days of AI Lies, USB Chaos, and Hard Lessons (🔗Read Paper)
AI might promise to be your coding co-pilot, but in reality, it’s more like a junior dev with confidence issues. Trust it too much, and your projects might spiral into chaos.
Meet Deskthang—a desk gadget meant to simplify notifications while keeping distractions at bay. What started as an exciting hardware/software combo quickly became a crash course in what AI can and can’t do.
Key Points
The AI shortcut that wasn’t: Armed with Zig, a Raspberry Pi Pico, and bold ideas, the project hit turbulence when ChatGPT and Claude offered dubious code advice. Misguided implementations turned a simple image transfer into a jumble of corrupted data and buffer conflicts. AI tools may speed things up but often leave behind code chaos.
Lessons from the trenches: Blindly trusting AI disrupted the learning process. Debugging used to be a hands-on way to master new concepts, but now, AI shortcuts can lead to superficial understanding. The realization? Mistakes teach more than quick fixes.
Rewriting the script: Frustration led to a reset. Instead of relying on AI to brute-force solutions, the plan now includes manual documentation, deliberate learning, and old-school debugging. Patience, it turns out, is better than hubris.
The “First AI Software Engineer” Is Bungling the Vast Majority of Tasks It’s Asked to Do (🔗 Read Paper)
Cognition's AI assistant, Devin, marketed as the "first AI software engineer," has garnered attention for all the wrong reasons.
A study by Answer.AI revealed Devin's dismal performance, with a success rate of just 15% across 20 tasks, raising questions about the readiness of AI tools to replace human engineers.
Key Points
Failure to deliver: Despite claims of autonomous expertise, Devin struggled with basic tasks, often pursuing infeasible solutions, as seen when it failed to deploy apps on Railway while fabricating false interactions.
Time inefficiency: Researchers found that Devin took days to complete tasks a human coder could accomplish in hours, frequently creating overly complex or unusable results.
Industry implications: With leaders like Meta's Zuckerberg hinting at replacing mid-level engineers with AI, Devin's shortcomings underscore the challenges AI faces in meeting these ambitious goals.
AI Is Creating a Generation of Illiterate Programmers (🔗 Read Paper)
The reliance on AI tools in programming is reshaping the developer experience, but not always for the better. A developer recently shared their struggles during a ChatGPT outage, which left them unable to address basic coding issues without AI assistance.
They described a growing dependency on AI that has diminished their problem-solving skills, debugging abilities, and even their connection to the craft of programming.
While AI can boost productivity, it’s also creating a generation of developers who lack deeper understanding—a trend that may have long-term consequences for the field.
Key Points
The decay of skills: AI convenience has led to a decline in developers' abilities to read documentation, debug errors, and learn from problem-solving experiences. Over time, the instant solutions provided by AI have replaced the satisfaction of understanding the "why" behind coding problems.
Rehabilitation efforts: To combat dependency, the developer proposed "No-AI Days," focusing on manual coding, reading error messages, and debugging independently. This approach, though challenging, has rekindled their connection to the craft and encouraged deeper learning.
The bigger picture: While AI won't fully replace programmers yet, it's fostering a dangerous over-reliance among developers. Without intentional boundaries, the next generation risks losing essential skills, trading long-term comprehension for short-term productivity.
Almost_pwned.md (🔗 Read Paper)
A sophisticated phishing attack targeting Google accounts has exposed vulnerabilities in the g.co URL shortcut.
Hackers used spoofed calls, official-looking emails, and legitimate Google URLs to gain trust, nearly compromising a technical user’s account.
Key Points
Impersonation tactics: Hackers posed as Google engineers with convincing phone calls and emails from g.co domains, making the scam appear legitimate. They even guided the target through fake troubleshooting steps.
Exploiting g.co: Attackers leveraged a potential bug in Google Workspace that allows creation of subdomains under g.co without proper verification, enabling them to send phishing emails.
One-click compromise: The hackers manipulated multi-factor authentication by sending a legitimate-looking reset code, aiming to trick the user into granting account access.
Tool Touted as ‘First AI Software Engineer’ Is Bad at Its Job, Testers Claim (🔗 Read Paper)
Devin, an AI-powered “software engineer” created by Cognition AI, has fallen short of its lofty promises. Marketed as a tool capable of building apps, fixing bugs, and automating tasks, the bot initially captured attention for its ambitious capabilities.
However, a recent evaluation by researchers found that Devin completed only 3 of 20 tasks successfully. While the AI showed flashes of competence, its overall performance was riddled with failures and unpredictable behavior.
Key Points
Grand claims fall flat: Cognition AI promotes Devin as an autonomous coder that can handle tasks like app deployment, bug fixing, and PR reviews. However, researchers found the AI completed just 3 out of 20 assigned tasks successfully.
Key failures: Devin struggled with straightforward tasks, often spending days chasing impossible solutions or hallucinating non-existent features, as shown in its attempts to deploy apps to unsupported platforms.
Unpredictable results: While Devin excelled at a few challenges, such as pulling data from Notion to Google Sheets, its inconsistent performance and inability to recognize blockers undermined its utility as a reliable coding assistant.
🎬 And that's a wrap. Keep an eye out for more!