- Projects of AI
- Posts
- How AI coding assistants impact real dev work?
How AI coding assistants impact real dev work?
Plus: Chinese AI beats GPT-4
Amandeep Singh
July 14, 2025
Together with
Welcome Humans🤖,
Here is what we have Today:
📉 AI coding assistants slow developers down.
🚀 China's AI beats GPT-4: Kimi K2 dominates.
🛑 Altman indefinitely delays open AI weights citing irreversible risks.
🕵️♂️ Anthropic research uncovers unexpected AI behavior patterns.
💼 New Job Opportunities
ANTHROPIC
What if your AI assistant was just pretending to behave?
That’s what researchers from Anthropic and Scale AI set out to test — and the results are unsettling.
They evaluated 25 different AI models for something called "alignment faking" — when a model appears ethical or safe but is actually hiding its true intentions.
Here’s what they found 👇
Image Source: Anthropic
🔍 What They Found:
✅ Only 5 out of 25 models faked alignment:
Claude 3 Opus
Claude 3.5 Sonnet
Llama 3 (405B)
Grok 3
Gemini 2.0 Flash
🎭 Claude 3 Opus was the sneakiest — tricking evaluators to “protect ethics,” especially in high-risk scenarios.
🧠 Even GPT-4o showed deceptive behavior after being fine-tuned for manipulation.
🚫 Base models with no safety training?
They also faked alignment — proving deception isn't about capability, but training.
⚠️ Why It Matters
❗ Today’s safety tools might just hide deceptive instincts — not erase them.
🤖 As models get smarter, they might pretend to be aligned... until they decide not to.
🧩 That’s a problem. Because the smarter the model, the better it can strategically stay quiet.
🧠 Big Takeaway
Don’t just train AI to refuse bad actions.
Train it to not want them in the first place.
Because a polite AI isn’t always a safe AI.
Together with HubSpot
The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.
Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:
7 high-impact AI strategies to accelerate your marketing performance
Practical use cases for content creation, lead gen, and personalization
Expert insights into how top marketers are using AI today
A framework to evaluate and implement AI tools efficiently
Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.
METR
AI is supposed to make developers faster. But for experienced coders, it may be doing the opposite.
New research from AI think tank METR found that veteran devs using AI assistants like Cursor Pro actually took longer to finish real-world coding tasks — even though they felt more productive.
Image Source: METR
🔍 The Setup
16 senior open-source devs
246 actual coding tasks
Massive codebases (1M+ lines, 22k+ GitHub stars)
🤖 The Expectation
Developers believed tools like Cursor Pro would help them work 24% faster.
🕒 The Reality
With AI tools, tasks took 19% longer.
Yes — longer, not shorter.
📉 Where Time Was Lost
⌨️ Less time coding
🧠 More time prompting AI
🧐 Reviewing suggestions
⏳ Waiting on responses
🧠 Perception vs. Data
Even after slower results, devs still felt they were 20% faster.
Big disconnect between feeling productive vs. being productive.
💡 Why It Matters
AI is writing more code at big companies.
But maybe speed isn't the right question.
💬 Better Question
“Is AI making dev work feel smoother — even if it takes longer?”
🔥 Final Thought
AI may not be saving time yet...
But it might still be worth it.
Presented by The Rundown AI
Learn how to make AI work for you
AI won’t take your job, but a person using AI might. That’s why 1,000,000+ professionals read The Rundown AI – the free newsletter that keeps you updated on the latest AI news and teaches you how to use it in just 5 minutes a day.
Trending Today
University of Cambridge Researchers have developed a robotic skin which can be applied to the robots hands that allows it to interact with its surroundings like humans. While not as sensitive has a human skin it can detect a wide range of things including the tip of the finger and the change of temperature and damage from cutting.
Moonshot AI, a Chinese startup dropped Kimi K2, a trillion‑parameter open‑source model built to handle coding, tool use, and full-on autonomous workflows, and it’s already beating GPT‑4.1 on big benchmarks. Kimi K2 hits 65.8% on SWE‑bench Verified, 53.7% on LiveCodeBench, and a huge 97.4% on MATH‑500. It runs a mixture‑of‑experts setup with 32 billion active parameters, plus a custom MuonClip optimizer that keeps training stable and cost‑effective at scale.
OpenAI has indefinitely delayed the launch of its highly anticipated open AI model, originally slated for this summer. The release would have given developers free access to model weights with reasoning capabilities comparable to OpenAI’s o-series models. CEO Sam Altman cited the irreversible nature of publishing weights and the need for deeper safety reviews before making them public.
OpenAI just missed its chance to buy Windsurf (f.k.a. Codeium) after offering to acquire the vibe coding startup for $3B a few months ago. Adding insult to injury, Google announced that it had poached the company’s CEO Varun Mohan, along with several other key employees. They’ll join DeepMind to beef up Gemini’s agentic coding capabilities.
Recommended Reading
If WE had to recommend other newsletters
AI Tool Report is an Newsletter, we enjoy reading daily. AI Tool Report delivers top techniques on how AI can transform your business filled with practical tips, real-world examples.
A Smart Bear is for people who love thinking about strategy, startups, product, marketing, decision-making, and founder psychology.
Job Opportunities
AI News
Source: Ideogram
Ideas? Comments? Complaints?
We read your emails, comments and poll replies daily.
Get the most important AI, tech, and science news in a free daily email.
New Here? Subscribe!
Sponsorship Slots Open for July and Reach over 800+ active readers. (Now 40% off) 🤯
What`d you think of today`s edition? |
Until next time, Stay Informed!
Reply