Databricks
LLMs “tend to fail” when faced with the “messy data, ambiguous language, and nuanced complexities of actual data analysis."
But Apache Hudi chief sounds a cautionary note over this week's $1 billion "Tabular" acquisition
Inference “up to 2x faster than LLaMA2-70B” for new model, trained on 12 trillion tokens.
"We haven’t disclosed a specific timeline to reaching profitability, and we are continuing to invest aggressively"
Whilst the intent to innovate with data is there, often the reality is a different story...
Fresh 100TB TPC-DS record is 2.2X greater than previous.
"We put it on a few web pages and said: ‘Hey download this!’ No one downloaded it..."