Timestamps: (🧬 00:06) Introduction to medical demos using Groq-powered LLMs (⚠ 00:34) Challenges of using LLMs in medicine: hallucinations and need for referenced information (💡 01:03) Explanation of RAG (Retrieval Augmented Generation) technique - Uses a database of text sources to gather relevant information for a user query - Top sources are fed into the LLM to generate a referenced response (📊 01:52) Demo 1: Genomics database hooked up to Mixtral LLM running on Groq hardware - Example query: "Which gene variants are associated with adjusted warfarin doses?" - Comparison of outputs with and without RAG ✅ With RAG: more detailed content, includes gene variants and genes, provides working PubMed links for verification ❌ Without RAG: less detail, missing some gene variants and genes (🧬 03:44) Demo 2: Real-time genome annotation using Llama-2 70B LLM on Groq hardware with RAG - Explanation of VCF (Variant Calling Format) file - Iterative processing of VCF to find matches in gene-drug database - LLM summarizes the abstract and provides a working PubMed link for each hit - Demonstrates real-time generation of DNA annotations Benefits: ✅ Demonstrates the potential for LLMs in medical settings when combined with techniques like RAG ✅ Improves reliability and accuracy of LLM outputs in medical contexts ✅ Provides referenced information for doctors to verify answers Cons: ❌ None mentioned in the video Key Takeaways: 🔑 RAG enhances output accuracy and reliability of LLMs 🔑 Groq hardware enables real-time processing of complex tasks like genome annotation 🔑 Combining LLMs with techniques like RAG opens up possibilities for their use in medical settings Score: 5/5 - Informative and engaging demonstration of Groq-powered LLMs in medical applications
Can another LLM run on Groq? For example, could Claude be trained and hosted by Groq? I would suppose the simplified processing of Groq will requires a specific types of a LLM. So, my question really is would the emulation possible to run other LLMs and would the emulation worth it??
This video doesn't show any genome analysis at all. It retrieves and summarises Abstracts of ( a very limited set of ) papers which mention a gene or a variant.
5:00 RAG alone isn't enough. You still get hallucinations with RAG and Groq only helps with the speed aspect of retrieval. Until llms don't hallucinate at all, no doctor would use this. First rule is do no harm. A doctor would rather give no advice about a gene, than wrong advice. I like the demo, but this is still so far from being useful
Timestamps:
(🧬 00:06) Introduction to medical demos using Groq-powered LLMs
(⚠ 00:34) Challenges of using LLMs in medicine: hallucinations and need for referenced information
(💡 01:03) Explanation of RAG (Retrieval Augmented Generation) technique
- Uses a database of text sources to gather relevant information for a user query
- Top sources are fed into the LLM to generate a referenced response
(📊 01:52) Demo 1: Genomics database hooked up to Mixtral LLM running on Groq hardware
- Example query: "Which gene variants are associated with adjusted warfarin doses?"
- Comparison of outputs with and without RAG
✅ With RAG: more detailed content, includes gene variants and genes, provides working PubMed links for verification
❌ Without RAG: less detail, missing some gene variants and genes
(🧬 03:44) Demo 2: Real-time genome annotation using Llama-2 70B LLM on Groq hardware with RAG
- Explanation of VCF (Variant Calling Format) file
- Iterative processing of VCF to find matches in gene-drug database
- LLM summarizes the abstract and provides a working PubMed link for each hit
- Demonstrates real-time generation of DNA annotations
Benefits:
✅ Demonstrates the potential for LLMs in medical settings when combined with techniques like RAG
✅ Improves reliability and accuracy of LLM outputs in medical contexts
✅ Provides referenced information for doctors to verify answers
Cons:
❌ None mentioned in the video
Key Takeaways:
🔑 RAG enhances output accuracy and reliability of LLMs
🔑 Groq hardware enables real-time processing of complex tasks like genome annotation
🔑 Combining LLMs with techniques like RAG opens up possibilities for their use in medical settings
Score: 5/5 - Informative and engaging demonstration of Groq-powered LLMs in medical applications
The side-by-side comparisons of output with and without RAG sources clearly showed how it enhances the reliability and accuracy of responses.
Can another LLM run on Groq? For example, could Claude be trained and hosted by Groq? I would suppose the simplified processing of Groq will requires a specific types of a LLM. So, my question really is would the emulation possible to run other LLMs and would the emulation worth it??
This video doesn't show any genome analysis at all. It retrieves and summarises Abstracts of ( a very limited set of ) papers which mention a gene or a variant.
Thank you very much! This will help a lot here.
it is the same Groq that X launched, right?
Hi, can i see the python script? thanks
5:00 RAG alone isn't enough.
You still get hallucinations with RAG and Groq only helps with the speed aspect of retrieval.
Until llms don't hallucinate at all, no doctor would use this. First rule is do no harm. A doctor would rather give no advice about a gene, than wrong advice.
I like the demo, but this is still so far from being useful
That's nonsensical. They will greatly help doctors.
@@BooleanDisorder backup your argument