METR ran a rare randomized trial on AI's impact in real-world dev work.
Result?
Tasks took 20% longer with AI tools. Even though devs felt 20% faster and experts expected +40%.
No hype - big open-source repos, seasoned devs, Claude 3.5–3.7, Cursor Pro, proper metrics & stat sig.
Turns out - AI slows down experienced devs on real projects.
Full study:
https://metr.org/Early_2025_AI_Experienced_OS_Devs_Study.pdf
Es ist so weit: Die nächste KI-Stufe ist gezündet! OpenAI hat heute GPT-4.1 vorgestellt – und das Upgrade ist beeindruckend.
Mit einem Kontextfenster von bis zu 1 Million Tokens!
Alle Details und Hintergründe findet ihr hier:
https://kinews24.de/gpt-4-1-api-ai-modelle-mini-nano-coding-kontextfenster/
Ein echter Gamechanger für Entwickler, Unternehmen und alle, die mit KI arbeiten. Mehr dazu in Kürze – bleibt dran!
OpenAI has released GPT-4.1 models with better coding, instruction following, and support for long-context tasks up to 1 million tokens
#OpenAI #GPT41 #GenAI #AIDevelopment #AIModels #CodingAI #InstructionFollowing #LongContextAI #OpenAIAPI
"Vibe coding" is a security disaster.
"His business, built entirely with AI assistance and 'zero hand-written code,' was experiencing bypassed subscriptions, maxed-out API keys, and database corruption."
It's gratifying to see my insights about the #techEconomy—especially #ITlayoffs in #cybersecurity and #softwareDevelopment partly driven by the current #AI hype cycle—amplified by peers & media.
"AI can't support what it doesn't know," says James Stanger, the chief technology evangelist at CompTIA, a nonprofit trade association for the US IT industry…"If you've got toxic companies that are interested in that binge-and-purge, on-and-off hiring of developers, I'm not sure they're going to create very good products." (Hoover)
It supports my own #codingAI #metrics that show a human solving a problem the AI never completed even with 400% more time & expert prompting. Confirmation bias? Possibly. Feeling seen & heard? You bet!
Hoover, A. 2025. The career ladder for software engineers is collapsing. Business Insider. https://www.businessinsider.com/career-ladder-software-engineers-collapsing-ai-google-meta-coding-2025-2
Jacobs, T.A. 2025. AI can’t replace IT professionals yet. LinkedIn. https://www.linkedin.com/posts/todd-a-jacobs_layoffs-itprofessionals-ai-activity-7298780638222929921-XuOA
New research on AI and code quality. The people who did the study last year showing AI increases "churn" did a follow-up study this year.
AI enables developers to produce new code substantially faster at the cost of long-term maintainability. Code is added about ~20% faster, but "churn" is ~50% higher, code duplication is ~30% higher, and refactorization is ~80% lower.
Chinese artificial intelligence lab DeepSeek has introduced DeepSeek V3 a open-source AI model that nearly closes the gap to GPT-4o and Claude 3.5 Sonnet. #AI #OpenSource #OpenSourceAI #DeepSeekV3 #DeepSeek #AIResearch #CodingAI #AIModels
GitHub’s Copilot goes multi-model and adds support for Anthropic’s Claude and Google’s Gemini. #CodingAI
Posted into Coding is Living @coding-is-living-mobileatom
Here's an #AI** **thought experiment: today** **#codingAI (s) are pretty bad still. In 5 years, they'll be good at what we're trying to get them to do now.
On that trajectory, in 10 they will be able to do a lot more. So here's my question: when AI will be the primary coder, will it develop its own coding language? More performant and better than anything we've ever created so far?
Or will it just converge/loop back to Assembly 'cause that's the root and it's more efficient?
OpenAI-powered debugging assistant
JamGPT is debugging alongside you, so you can identify the bug, receive the code to fix it, and share with your team – all in one link.
Coding your own AI in 2023 with fastai
by Vincent Will @wweb_dev
#CodingAI
#fastai
#webdev
#coding
#python
#deeplearning
#ai
#beginners
https://dev.to/vincenius/coding-your-own-ai-in-2023-with-fastai-2pbk
