[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

![curiousZeedX Avatar](https://lunarcrush.com/gi/w:24/cr:twitter::1668245944360370180.png) Ani [@curiousZeedX](/creator/twitter/curiousZeedX) on x 1756 followers
Created: 2025-07-25 10:58:03 UTC

Anthropic research just dropped a wild paper: giving AI more 'thinking time' can actually make it WORSE. For some logic puzzles and math problems, model accuracy decreases as you let it generate more reasoning tokens. 

Mechanistic Interpretability(understanding how neural networks, particularly large language models, perform computations by reverse-engineering their internal workings) is the hottest topic to do research in right now.

![](https://pbs.twimg.com/media/Gwsrl40bcAEy8tB.jpg)

XX engagements

![Engagements Line Chart](https://lunarcrush.com/gi/w:600/p:tweet::1948699260553494845/c:line.svg)

**Related Topics**
[networks](/topic/networks)
[neural](/topic/neural)
[coins ai](/topic/coins-ai)

[Post Link](https://x.com/curiousZeedX/status/1948699260553494845)