New Tutorial: Fine-tune #Llama3 to Automate Customer Service Learn how you to efficiently #finetune Meta Llama-3 to automatically classify customer support issues and generate a response. Notebook included!
LLMS
-
LMSYS Anonymized Model Names and OpenAI Attribution
By
–
The LMSYS explanation is that they pick "anonymized" names for new blind tested preview models, so it's possible OpenAI themselves didn't pick this name (assuming it's one of theirs, which still seems likely to me)
-
Tokenizer Design Evolution: Why Revisit Outdated 2019 Architecture?
By
–
Why would they do that though? The state of the art in tokenizers has moved on enormously since 2019, why go back to an old design?
-
Prompt Optimization and RAG Over Model Fine-Tuning
By
–
I'm not patient enough to fine-tune my own models – plus I have a hunch that for most applications you can get better results from prompt optimization and techniques like RAG that don't require any additional fine-tuning at all
-
New Math Benchmark: 387 Challenging Problems Tests GPT-4 Limits
By
–
As benchmarks continue to get saturated, it's great to see a no-frills benchmark of 387 challenging math problems: https://
github.com/protagolabs/od
yssey-math/tree/main
… GPT-4 is 66% on high-school subset, 42% on college subset, and only 11% on high-school competition subset. -
AI Summarizer Bot Saves Time via Text Messages
By
–
— Matt Shumer (@mattshumer_) 30 avril 2024
I made myself an insanely useful AI summarizer bot. It lives in my text messages. I just send it a link to an article, video, etc., and it texts me back a fantastic summary, in seconds. This is going to be a massive time-saver.
-
Med-Gemini Achieves State-of-the-Art Medical Task Results
By
–
The multimodal capabilities of the base Gemini models shine in evaluation of multimodal medical tasks in Med-Gemini, showing state-of-the-art results in 5 of the 8 benchmarks we examined.
-
Med-Gemini Outperforms Human Experts in Medical Text Tasks
By
–
Med-Gemini results are preferred over human expert results on long-form text-based tasks like medical summarization, doctor referral generation, and medical simplification (where the goal is to generate a plain language summary that can be understood by most readers without a
-
Med-Gemini Advances Clinical Diagnosis on NEJM Benchmark
By
–
On the NEJM CPC benchmark (an open-ended diagnosis task), the input is a text-based, challenging clinico-pathological case (CPC) report, and the output is a differential diagnosis list, comprising 10 potential diagnoses. The Med-Gemini model does well on its own and improves
-
Med-Gemini-L 1.0 Achieves State-of-the-Art Medical Reasoning Performance
By
–
The Med-Gemini-L 1.0 model is state-of-the-art on 10 of the 16 text-based medical reasoning tasks that we looked at.