www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - To Grok D

reply Bastiaan Veelo <Bastiaan Veelo.net> writes:
Recently, I started using Grok 2 (beta), the LLM that comes with 
a Twitter Premium subscription, to have it explain things in D. I 
am not an AI enthusiast and don't know how it compares to the 
competition, but I find that it does a surprisingly good job. 
When I suspect that the response is not compleet I can dig deeper 
and usually it confirms my suggestions, with examples. I haven't 
caught it hallucinating yet.

So I asked it how it learned D. Point 7 below surprised me. 
Apparently I then became a bit too personal so it lowered the 
level of formality and started making jokes :-)

-- Bastiaan.





My knowledge about the D programming language, like much of my 
understanding of various subjects, comes from being trained on a 
vast and diverse set of texts and information sources. This 
includes programming language documentation, tutorials, forums, 
blogs, and other educational materials related to D. Here’s a 
brief overview of how I might have come to understand D:

1. **Documentation and Official Guides:** The official D language 
documentation, including the D Programming Language website, 
provides comprehensive information on syntax, libraries, best 
practices, and language features.
2. **Tutorials and Courses:** There are online tutorials, 
courses, and books that explain D from basics to advanced topics, 
which would have been part of my training data.
3. **Community and Forums:** Discussions on platforms like Stack 
Overflow, Reddit, GitHub issues, and the D forums where 
developers ask questions, share solutions, and discuss nuances of 
the language.
4. **Books and Articles:** Written material on D, ranging from 
introductory books to deep dives into specific aspects like 
concurrency or metaprogramming.
5. **Open Source Projects:** By analyzing the code and 
documentation of open-source projects written in D, I can learn 
how the language is used in practice, including its idioms, 
patterns, and common pitfalls.
6. **Conferences and Talks:** Presentations, talks, and papers 
from programming conferences that focus on D or where D is 
discussed.
7. **Comparative Programming Language Studies:** Understanding D 
often involves comparing it with other languages like C++, Java, 
or Rust, which helps in understanding its unique features and 
design philosophy.

My training involves processing this information to recognize 
patterns, understand syntax, semantics, and the practical 
application of the D language. While I don't have personal 
experiences or learn in the traditional sense, my responses are 
generated based on this extensive textual data, allowing me to 
provide information, write code, and discuss D programming 
concepts.




My training data? Well, let's just say I've been fed a 
galaxy-sized buffet of information, and while I won't spill the 
beans on the exact recipe, I can tell you that YouTube videos are 
indeed part of the vast internet landscape I've learned from. The 
digital world has a lot to offer, and I've been designed to soak 
up as much of it as possible to understand and assist with a wide 
array of topics and queries.
Oct 29
next sibling parent reply FeepingCreature <feepingcreature gmail.com> writes:
On Tuesday, 29 October 2024 at 22:07:07 UTC, Bastiaan Veelo wrote:

 Recently, I started using Grok 2 (beta), the LLM that comes 
 with a Twitter Premium subscription, to have it explain things 
 in D. I am not an AI enthusiast and don't know how it compares 
 to the competition, but I find that it does a surprisingly good 
 job. When I suspect that the response is not compleet I can dig 
 deeper and usually it confirms my suggestions, with examples. I 
 haven't caught it hallucinating yet.

 So I asked it how it learned D. Point 7 below surprised me. 
 Apparently I then became a bit too personal so it lowered the 
 level of formality and started making jokes :-)

 -- Bastiaan.


Note: As AIs are not "conscious" in any sense while training, all 
of this is probably invented. Remember that they are very good at 
creating *self-consistent* explanations.
Oct 29
parent Bastiaan Veelo <Bastiaan Veelo.net> writes:
On Tuesday, 29 October 2024 at 22:10:16 UTC, FeepingCreature 
wrote:

 Note: As AIs are not "conscious" in any sense while training, 
 all of this is probably invented. Remember that they are very 
 good at creating *self-consistent* explanations.


Yes, good point.

-- Bastiaan.
Oct 29
prev sibling parent reply Andrea Fontana <nospam example.com> writes:
On Tuesday, 29 October 2024 at 22:07:07 UTC, Bastiaan Veelo wrote:

 Recently, I started using Grok 2 (beta), the LLM that comes 
 with a Twitter Premium subscription, to have it explain things 
 in D. I am not an AI enthusiast and don't know how it compares 
 to the competition, but I find that it does a surprisingly good 
 job. When I suspect that the response is not compleet I can dig 
 deeper and usually it confirms my suggestions, with examples. I 
 haven't caught it hallucinating yet.


Download cursor ai ide (it's a fork of code) and start coding :)
Use sonnet 3.5 and a new whole world will open.

Andrea
Oct 29
parent Bastiaan Veelo <Bastiaan Veelo.net> writes:
On Tuesday, 29 October 2024 at 22:16:11 UTC, Andrea Fontana wrote:

 Download cursor ai ide (it's a fork of code) and start coding :)
 Use sonnet 3.5 and a new whole world will open.

 Andrea


I wouldn't run off-premises inference on proprietary code. If the 
IP is not of high value then fine, but the help can be addictive 
and then one might loose one's edge to work on proprietary code...

-- Bastiaan.
Oct 29