At some point, LLMs stop executing and start explaining

<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1s1ckkm/why_llms_sound_right_but_fail_to_actually_do/"> <img alt="Why LLMs sound right but fail to actually

<!-- SC_OFF --><div class="md"><p>I think LLMs have trouble with conceptual generalization because every dimension of their representations ALWAYS participates in every operation,

<p>You opened ChatGPT before Slack this morning.<br /><br /> Not for anything complicated. Just to figure out how to begin.</p> <p>That small move is already changing what

<!-- SC_OFF --><div class="md"><blockquote> <p>Wait, I need to be careful about the &quot;no_think&quot; tag in the system prompt. The system prompt says <code>/no_think</code>.

<!-- SC_OFF --><div class="md"><p>Something I keep noticing is that LLMs can give really strong, coherent answers in a single prompt, but as soon as the conversation goes on for a

<!-- SC_OFF --><div class="md"><p>I asked Gemini to give me all the days that have &quot;d&quot; in them. It returned -</p> <p>Monday, Wednesday, Thursday, Sunday<br />

<p>You've built the demo. It works. The LLM responds, the tools fire, the output looks great.</p> <p>Then you push it to production — and everything breaks.</p> <p>API calls fail

<!-- SC_OFF --><div class="md"><p>Built a free alternative to Claude Code ($20-$200/mo) that uses NVIDIA's open models through the same CLI framework (FREE!).</p> <p><strong>How