Contact is Director of AI Research at $ADSK.
Pointed out that 1) they are very tied to $AMZN for GenAI because that is where their production data lives, and they need compute and data to be co-located for AI in production, 2) a killer app for GenAI doesn’t exist yet outside code generation, and 3) their most compelling GenAI project is a code generator for their 3D design tools.
Highlights
-Team sits within ADSK Research which consists of 200 people. Sub-team of 20 people focused on simulation research, manufacturing research, etc.
-Primary cloud is $Amazon (AMZN.US)$ AWS (75%) and secondary is $Microsoft (MSFT.US)$ is (25%)
-Very tied to $Amazon (AMZN.US)$ bc this is where production data is locked down – can’t just pick it up and move it, and need data and compute to be co-located. Think lots of businesses are limited by where their data lives – can always train a model for inference and then deploy it near the data.
-Use Open.AI via $Microsoft (MSFT.US)$ bc want their enterprise security constructs
-Fine-tuned own model based on GPT 3.5 to do natural language code generation for Maya which has 3D design tools used in movies and video games. Used to be a difficult to use Python interface and now complemented w/ natural language. In beta for now.
-RAG is useful for search. Creates a pool of content that LLM can reference, and thus have less hallucinations.
-Haven’t really found killer app for GenAI unless it is code generation. Think most of what GenAI will do will live within software workflows.
-Machine learning is great but also has been steadily commoditized as has become so easy to try different models on data and then deploy to production.
-Team sits within ADSK Research which consists of 200 people. Sub-team of 20 people focused on simulation research, manufacturing research, etc.
-Primary cloud is $Amazon (AMZN.US)$ AWS (75%) and secondary is $Microsoft (MSFT.US)$ is (25%)
-Very tied to $Amazon (AMZN.US)$ bc this is where production data is locked down – can’t just pick it up and move it, and need data and compute to be co-located. Think lots of businesses are limited by where their data lives – can always train a model for inference and then deploy it near the data.
-Use Open.AI via $Microsoft (MSFT.US)$ bc want their enterprise security constructs
-Fine-tuned own model based on GPT 3.5 to do natural language code generation for Maya which has 3D design tools used in movies and video games. Used to be a difficult to use Python interface and now complemented w/ natural language. In beta for now.
-RAG is useful for search. Creates a pool of content that LLM can reference, and thus have less hallucinations.
-Haven’t really found killer app for GenAI unless it is code generation. Think most of what GenAI will do will live within software workflows.
-Machine learning is great but also has been steadily commoditized as has become so easy to try different models on data and then deploy to production.
Disclaimer: Community is offered by Moomoo Technologies Inc. and is for educational purposes only.
Read more
Comment
Sign in to post a comment
