Blocksyc - Truth over validation.

Working with LLMs, I’ve noticed how heavily responses can change based on framing alone. Ask “Is this a good idea?” and you often get encouragement. Ask “What’s wrong with this idea?” and suddenly the response becomes much more critical. That’s sycophancy. Not obvious hallucinations, but models subtly shaping answers around what they think users want to hear. It’s a known issue discussed openly by companies like Anthropic and OpenAI, but I think awareness around it still needs to grow. I built Blocksyc to help reduce that behavior through a system layer focused on challenging assumptions, stripping emotional framing, and prioritizing truth over validation.