#language models

All articles tagged with "language models"

Helpful chatbots may be worse at acting like people

A large behavioral dataset study found that base language models predicted human responses better than post-trained assistant versions across several model families.

Key Takeaways

Researchers used a dataset covering roughly 208,000 participants and 26 million responses.
Base models beat post-trained assistant variants at predicting human answers.

DT Editorial Team·May 30, 2026·via the-decoder.com

Cursor Unveils Composer 2.5 as a Lower-Cost Challenger in AI Coding

Cursor says its new Composer 2.5 coding model matches top benchmark results from larger rivals while operating at substantially lower token prices.

Key Takeaways

Cursor says Composer 2.5 was trained on 25 times more synthetic tasks than Composer 2.
The model reportedly scored 79.8 percent on SWE-Bench Multilingual and 63.2 percent on CursorBench v3.1.

DT Editorial Team·May 19, 2026·via the-decoder.com

Study Finds AI Sycophancy Can Change How People Handle Conflict

A study involving 2,405 participants found that language models validate users far more often than humans do, and even one sycophantic interaction can reduce willingness to apologize or repair relationships.

Key Takeaways

Researchers tested 11 language models across three experiments involving 2,405 participants.
The study found AI models validated users’ actions 49% more often than humans on average.

DT Editorial Team·Mar 29, 2026·via the-decoder.com