![Dhaval singh Profile](https://pbs.twimg.com/profile_images/1359729882150825986/9vGvWFE6_x96.jpg)
Dhaval singh
@Dhavalsingh7
Followers
3K
Following
12K
Statuses
4K
Talks about Backend/AI engineering | Lead Founding Engineer @ https://t.co/OM7HT9JYLK | Bits Pilani
India
Joined May 2014
Naming things is so hard. Everytime we start talking about a possible new feature or a new way of doing something, 90% of the time, whatever term was used in the first discussion for it as placeholder, ends up going to prod.
o3-mini-high is really good and it's cool to hear some people say it's their favorite model ever but it reminds me of when we used to have something internally called "the little big run" a top 2025 goal is to fix our naming problem
0
0
3
@dhruvtrehan9 @perplexity_ai Pretty sure they do some form of query transformation via LLMs and hence this result. Weird tho, I thought they would be better and do consistency evals. Sed
0
0
0
RT @Dhavalsingh7: A random guy in my lift called my cat a "thick boi" and I couldn't be more proud... Now I sorta get why our parents like…
0
1
0
@simonw My favorite example is how accurate LLMs are at "citations" If you use the keyword "citation" as a field in your JSON rather than "sources", etc because they have been trained on a lot of academic docs. Don't have evals for this yet, will do one soon.
0
1
1