I'm a fairly experienced software developer with an understanding of how LLMs work, so I've been mostly able to mitigate this with my style of prompting and get good results.
I can only imagine how much this would hurt junior developers and those who are just learning, though. Especially "vibe coders" who are trying to avoid learning how to code altogether.
I think that this might be a side effect of the “Explanatory" mode I have it set to, which gives more verbose output. I suspect it might be told to be supportive, as if I'm just learning.
I'm all for support, but I also require pushback. Not everything I touch turns to gold.
I've been enjoying experimenting with Claude Code and found it to be a major productivity boost, but one annoying thing - at least with how I've configured it - is just how damned sycophantic it is.
It acts as if everything I say is genius (spoiler: it's not). Sometimes I've thought of flaws in my own thinking after and, when I explain them, it acts as if I was a genius to figure _that_ out too.
It's much better at unearthing flaws when I come at it with neutral language.
This screen now controls every TV in my house, with the functions I most often use a button press away. As they’re widgets I don’t even have to open the app, I can just swipe my device across one screen and they’re all there.
All vibe coded, ready to submit it to the store shortly, just need to set up the IAP and check they work.
(more Linux and FOSS news in previous posts of thread)
New GIMP plugin for integrating Gemini AI image creation:
https://www.phoronix.com/news/Dream-Prompter-GIMP-Plugin
Zed adds beta Claude Code integration via Agent Client Protocol:
https://alternativeto.net/news/2025/9/zed-introduces-native-claude-code-integration-in-beta-via-its-agent-client-protocol/
ollama 0.11.9 released with improved CPU/GPU performance, bug fixes:
https://www.phoronix.com/news/ollama-0.11.9-More-Performance
New Swiss AI model Apertus to make its debut in the upcoming Swiss Weeks hackathons:
https://betanews.com/2025/09/02/new-fully-open-and-transparent-large-language-model-launches-its-swiss-of-course/
Semantic grep for Claude Code (RUST) (local embeddings)
IMHO: Why #claudecode has quality shifts. It's
* not prompting
* not context
* not the problem type
It's not you. Or your project.
It's #Anthropic. They use quantized models (during the daytime) to scale the load. #Anthropic doesn't make that transparent to the users. No inference provider does. But it's cheaper for the provider.
Quantized models are smaller / down-scaled variants with less precision. Less capability. And less usefulness, therefore. Cheaper for the provider. Customers won't know. But pay the same.
Given that even a 200 USD subscription results in down-scaled LLM service quality, you need to consider
a) self-hosting
b) use open-source clients.
Option 1:
https://github.com/charmbracelet/crush
* OpenRouter models: Gemini, Grok
* self-hosted models: Qwen Code, Kimi K2
Self-hosting option with Modal:
https://modal.com/docs/examples/vllm_inference
Option 2:
https://github.com/openai/codex
For me, #codex is slow.
Option 3:
https://github.com/musistudio/claude-code-router
Hacks like CCR (Claude Code Router) with the aforementioned models are not mid-term solutions.
Option 4:
A #gemini-cli fork:
https://github.com/acoliver/llxprt-code
I started to use this with various models, depending on the task.
What do Options 1 to 4 have in common? You cannot rely on the providers.
GLM 4.5 with Claude Code is a killer combination
https://docs.z.ai/guides/llm/glm-4.5
#HackerNews #GLM4.5 #ClaudeCode #AItools #MachineLearning #Innovation
First attempt will be 95% garbage: 6 weeks with Claude Code
If you use #claudecode you should untick this button in the Claude Web interface:
The way #Anthropic pushes their new Terms of Service is not ok. Using chats to improve models by default is not ok.
I like the tools, but the posture Anthropic displays is setting the stage for failure, liability, and security breaches.
https://privacy.anthropic.com/en/articles/10023580-is-my-data-used-for-model-training
It's not 100% clear to me whether #claudecode even follows this setting. "May" is not a good word here. Bad policy language. The word you should use is "does not, if"
Launch 
Linearis is a #CLI tool for #LinearApp with JSON output, smart ID resolution, and optimized GraphQL queries. Designed for #LLM #agents and humans who prefer structured data.
https://github.com/czottmann/linearis
It replaced Linear's #MCP server for me. Why? Because that thing consumes 13k tokens just by connecting to #ClaudeCode et al. Linearis uses less than 1k. For models w/ smaller context windows this makes a diff!
It's fast and does everything I need 
And I think you might like it, too!
I spent about 7 days vibe coding my first app, an AI powered social media analytics app. It was a mixture of excitement, frustration, and a lot of lessons learned.
This isn't a tutorial. It's a realistic look at what it was like for me, an Entrepreneur, IT professional and home labber to vibe code an app that is actually useful with no previous experience.
https://labb.run/lessons-learned-from-vibe-coding-my-first-social-media-analytics-app/
Anthropic to counteract usage of Claude Code for "vibe hacking"
https://www.anthropic.com/news/detecting-countering-misuse-aug-2025
`afplay /System/Library/Sounds/Bottle.aiff` plays a sound on your mac without installing anything else. Add to #claudecode hooks to claude boop when it's done :)
I always tell my clients that ask to use #AI to "lower my fees":
AI can be my assistant, but not our boss. If you still want this, be sure to sign this:
DISCHARGE OF RESPONSABILITY
I feel so validated watching #Claude also struggle with React and Node.js. #claudecode #react #nodejs
#claudecode without #phpstan would be a catastrophe!
Thank God someone built that wonderful tool! 
Happy weekend 
I really don't know *why* this works, but it does. When I'm trying to challenge Claude Code to come up with viable alternatives to bypass its reward/laziness syndrome I ask it this: "What would the cool kids do?" 10/10 doctors agree that this helps...a lot. Try it and see what how it responds.
"Our Threat Intelligence report discusses several recent examples of Claude being misused, including a large-scale extortion operation using Claude Code, a fraudulent employment scheme from North Korea, and the sale of AI-generated ransomware by a cybercriminal with only basic coding skills. We also cover the steps we’ve taken to detect and counter these abuses.
We find that threat actors have adapted their operations to exploit AI’s most advanced capabilities. Specifically, our report shows:
- Agentic AI has been weaponized. AI models are now being used to perform sophisticated cyberattacks, not just advise on how to carry them out.
- AI has lowered the barriers to sophisticated cybercrime. Criminals with few technical skills are using AI to conduct complex operations, such as developing ransomware, that would previously have required years of training.
- Cybercriminals and fraudsters have embedded AI throughout all stages of their operations. This includes profiling victims, analyzing stolen data, stealing credit card information, and creating false identities allowing fraud operations to expand their reach to more potential targets."
https://www.anthropic.com/news/detecting-countering-misuse-aug-2025
