Terence Parr
@the_antlr_guy
Followers
7K
Following
1K
Statuses
2K
Tech lead at Google (opinions mine), ex-prof CS/data sci. Prog languages and machine learning; creator ANTLR parser generator & more: https://t.co/CU6sxinDkv
San Francisco
Joined December 2010
Launched a new website, dedicated to giving deep but intuitive explanations of machine learning and related topics.
11
235
716
RT @poliloveslogic: Matrix Calculus You Need For Deep Learning The paper tries to explain all the matrix calculus necessary to understand…
0
89
0
RT @Bytebase: explained how we made it to built the SQL autocomplete framework with ANTLR4. 🦿 @the_antlr_guy 👇
0
2
0
@khmarbaise @googlecloud Yeah, I didn't have the motivation to dig into that since it's really only dealing with public data.
0
0
0
RT @IvanPribec: Started a discussion on using ANTLR for parsing Fortran: I think we would really benefit from havi…
0
1
0
@BlakPantherBabe Sorry about that. It's much better than it was but still gets overloaded I think. Someone gets naughty and pegs the CPUs :( Please try now.
1
0
1
@Budget is totally broken in Denver. What’s wrong with you guys making everybody wait in line while one person slow walks people to their car and slowly walks back. 96° out too. There’s a bunch of employees walking around and nobody is getting your customers to your cars!!
1
0
0
Yep, biggest hurdle for effective use of LLMs is dealing with the hideous non-determinism. Even when grounding an LLM with good examples or context, the same model with the same input and temperature 0 (assume top-k>1) can give different answers. Mixture of experts is key!
I've noticed a trend of more projects using LLM as Judge, but too many people are using just one LLM as a judge. Fortunately, LLMonPy makes is very easy to use multiple LLMs as judges. You can read more here:
0
0
3
RT @sftombu: You can read more about what I am trying to accomplish and check out the repo here:
0
1
0
RT @sftombu: @AnthropicAI 3.5 does much better than 3.0 at answering a question about a limerick buried in a list of 2500+ limericks. http…
0
1
0
RT @sftombu: The evaluation process for AI pipeline responses can generate data that can be used to improve future responses -- similar to…
0
3
0
Check out how much better GPT-4o appears to be at attention across input window than previous models. "Needle In A Needlestack" creates a prompt that includes thousands of limericks and the prompt asks a question about one limerick at a specific location.
5
3
5