Data Machina #178
Exploring ChatGPT. Deep Metrics Learning. Fine-tuning CLIP models. Transformer circuits. Probabilistic time series forecasting with transformers. Learning from limited, imperfect data.
Exploring OpenAI ChatGPT. I’ve been slamming the bot like so many others. Overall it’s quite amazing. You can generate code, poems, movie scripts, essays and all sorts of crazy stuff… It’s fluent and has “memory.” It’s not a search engine, and sometimes it’s factually incorrect and not verifiable. It’s very politically correct. And it lies too.
The core engine of ChatGPT is powered by GPT 3.5 Language Models and a RL algo called Proximal Policy Optimization (PPO.)
In ChatGPT: Optimizing Language Models for Dialogue you can read about its limitations, strengths and weaknesses.
People are building ChatGPT-based tools, apps like crazy! Here are a few:
A tool that embeds ChatGPT in the MacOS menubar
A hack that runs ChatGPT in WhatsApp
An API for interacting with ChatGPT using Python and the Shell
Then there are lots of people in the wild trying prompt injections to hack ChatGPT.
It’s gonna be a dull, damp, cold day. So I might as well grab a beer and keep hitting the bot with prompts…
Write Python code to implement a Longformer-based BART for long-form text summarisation
Try your ChatGPT prompts here and see what you get.
Have a nice week.
10 Link-o-Troned
A Pythonista *Experience*
Scripting aRt
Deep & Other Learning Bits
ResearchDocs
El Robótico
[Free book & code] Foundations of Robotics with Python & ROS
Perceiver-Actor: A Multi-Task Transformer for Robotic Manipulation
data v-i-s-i-o-n-s
DataEng Wranglings
startups -> radar
ML Datasets & Stuff
Postscript, etc
Tips? Suggestions? Feedback? email Carlos
Curated by @ds_ldn in the middle of the night.