– It's starting to do the whole "add this code here" thing where it's asking me to do the work — hadn't seen that before… now it's happening frequently
– I'm giving it simple code snippets for it to weave together into a system. Used to work perfectly on the first shot. Now it
@mattshumer_
-
AI Code Generation Quality Declining Frequently
By
–
-
AI Model Performance Degradation and Context Understanding Issues
By
–
It's making really dumb mistakes. Like, shockingly dumb. That wasn't happening before. Doesn't seem to use context as well as before.
-
Finding good enough off-the-shelf AI solutions for small projects
By
–
Damn, ok. Putting something small together and was hoping I could just use something off the shelf. Is there anything w/ mediocre perf. that is *good enough*?
-
Detecting End of Speech in Whisper Transcription
By
–
How do I detect when someone is done speaking when transcribing with something like Whisper? cc @batwood011
-
Few-shot learning enhances fine-tuned model performance
By
–
Never underestimate the power of few-shot on top of fine-tuned models
-
Building a Wirecutter for AI Model Reviews and Evaluation
By
–
I love this idea. Kind of like @wirecutter for models. At least for a little while, human reviewers might be needed. LLM-as-judge might be a bit too brittle for this. If anyone is working on this, let me know. Very interested in supporting.
-
Improve Fine-Tuned Models With Few-Shot Data Generation
By
–
High alpha trick to improve your fine-tuned models: Step 1: fine-tune your model on the data you have, do RLHF, whatever you want to do
Step 2: choose some of the best examples you have, and few-shot your model with those examples to generate more
Step 3: train on those! -
WizardLM 8x22B Removed from HuggingFace: What Happened?
By
–
The official WizardLM 8x22B was removed from HuggingFace, and the blog post link is 404ing. What is going on?