See @parth007_96
’s brilliant notes on reverse-engineering GitHub Copilot here: https://
thakkarparth007.github.io/copilot-explor
er/posts/copilot-internals
…
LLMS
-
Reverse-engineering GitHub Copilot: brilliant notes by @parth007_96
By
–
-
GitHub Copilot’s implicit prompt engineering creates a competitive moat
By
–
When you hit tab in GitHub Copilot, you’re FITM-completing a prompt of your file with an implicit comment header added, and it takes pages to describe what might be in that header. That’s prompt engineering. That’s a moat.
-
GPT-3/LLM prompts as assembly code, need prompt compilers
By
–
I increasingly see GPT‑3/LLM prompts as assembly code, not as human interface. We shouldn’t be writing prompts, but prompt compilers. A template string is not a moat.
-
Natural Language Processing NLP NLU NLG Speech Recognition
By
–
In this @Cognilytica #AIToday #podcast AI Glossary Series episode 'Natural Language Processing (NLP), NLU, NLG, Speech-to-Text, TTS, Speech Recognition' hosts @rschmelzer & @kath0134 define these terms & share how they fit into #AI. Full episode: https://
cognilytica.com/2022/12/16/ai-
today-podcast-ai-glossary-series-natural-language-processing-nlp-nlu-nlg-speech-to-text-tts-speech-recognition/?utm_source=dlvr.it&utm_medium=twitter
…
#NLP -
Code Capture as Foundation for RLHF Self-Bootstrapping
By
–
“capture code” could be a very cool basis for RLHF/retraining. this thing is bootstrapping itself!
-
ChatGPT++ for Scientific Discovery: Possibilities and Requirements
By
–
I guess I'm a bit more interested in chatgpt++ for scientific discovery more broadly and what that would take / look like.
-
ChatGPT Trust Issues and GPT-3 Regulation Discussion
By
–
Here is what #ChatGPT says about issues of trust and regulating #GPT3:
-
Steering LLMs with AI Alignment and Robotic Laws
By
–
Good reading on AI alignment, I've been wondering how one could steer LLMs with an equivalent of Three Laws of Robotics
-
Optimizing ChatGPT++ with Reinforcement Learning for Better Parameters
By
–
yes having no heat cycle is such a big deal chatgpt++ can probably help some, but i suspect RL for optimizing the input parameters can help a lot!
-
Model reveals current date despite training data end in 2021
By
–
Expanding on a (private account) reply: Its response reliably contains the current date, information it could not learn from its training data, which ends 2021-09. In all other contexts it reliably pretends not to know the current date and will guess incorrectly if asked.