Parv Mahajan

Karma: 1,005

parvmahajan.com

Several frontier models are substantially prefill aware

yeedrag, Parv Mahajan, David Africa, alexsouly, Jordan Taylor and RobertKirk

17 Jun 2026 17:41 UTC

55 points

2 comments5 min readLW link

AI Safety’s Biggest Talent Gap Isn’t Researchers. It’s Generalists.

Topaz, agucova, Alexandra Bates and Parv Mahajan

13 Apr 2026 14:47 UTC

59 points

2 comments6 min readLW link

The current SOTA model was released without safety evals

Parv Mahajan and yeedrag

8 Mar 2026 1:51 UTC

110 points

12 comments5 min readLW link

Three ways to make Claude’s constitution better

Parv Mahajan2 Feb 2026 21:48 UTC

36 points

3 comments2 min readLW link

Turning 20 in the probable pre-apocalypse

Parv Mahajan21 Dec 2025 10:14 UTC

444 points

66 comments3 min readLW link

Seriously, use text expansions

Parv Mahajan9 Dec 2025 5:08 UTC

12 points

0 comments1 min readLW link

(parvmahajan.com)

TastyBench: Toward Measuring Research Taste in LLM

Parv Mahajan, Yilin and yix

2 Dec 2025 23:26 UTC

33 points

2 comments6 min readLW link

Parv Mahajan’s Shortform

Parv Mahajan7 Aug 2025 18:44 UTC

3 points

20 comments1 min readLW link

Lessons from a year of university AI safety field building

yix, afterless, Parv Mahajan, Andersehen, Tuna and neverix

6 Jun 2025 14:35 UTC

35 points

3 comments7 min readLW link