DConf '23--Language Models, D, and so on--Max Haughton
Even in their infancy, large language models like ChatGPT are the talk of the town. Until recently, LLMs have been inferenced (run) using huge chunks of Python code, but recent interest has brought about some much simpler, specialized, libraries written in native languages (llama.cpp being a recently famous example). In this talk, Max talks about the state of LLMs and how they might be useful for D programmers.
The ODP slides are available here:
https://dconf.org/2023/slides/haughto...
DConf '23 Homepage
https://dconf.org/2023/index.html
The D Programming Language Home
https://dlang.org/
#dlang #dconf #ai
TIMCODES:
00:00 Introduction
01:42 LLM defined
02:57 Framing--what do we care about?
04:37 Is it really for programmers?
07:20 What programmers want and what they can do
08:22 Usable models
10:13 A specific example for D
17:56 What if you had a model to run locally?
19:57 New dog, old tricks
22:52 Q: Is there a NumPy equivalent for D?
26:05 Question about a potential benefit of static type systems
27:48 Q: What's the status of auto diff in D?
30:14 Q: Where should we focus to move forward?
33:02 Q: What is the benefit using AI to compress Wikipedia?