Shrikanth Shankar

[Ali Ghodsi @alighodsi]

Love this integration between @huggingface and @databricks. Concretely, you will be able to train your own LLM from using Spark and Transformer/Dataset with this tight integration:
https://t.co/sj2NOn6x0W

[Matei Zaharia @matei_zaharia]

Very excited to return to UC Berkeley as a professor starting this week. I’ll be collaborating with the Sky Lab, @UCBEPIC, @berkeley_ai and others!

[Reynold Xin @rxin]

The upcoming Data and AI Summit keynote lineup looks insane: Eric Schmidt, Satya Nadella, Marc Andreessen, Nat Friedman (Github CEO introduced copilot), LangChain creator, PyTorch co-creator, DuckDB creator, and a few top researchers in AI and robotics https://t.co/POwT09c89c

[Shrikanth Shankar @ShrikanthSS]

@jevering yep, learning and learning quickly is a huge indicator of failing right. Sadly, I've seen things get to a stage where bets become too big to fail. if you hit 100% of your bets you arent taking enough risks imo (or gaming the result some way)

[Shrikanth Shankar @ShrikanthSS]

@gwenshap @sorenbs @sampullara @prisma can you do create unique index in a concurrent fashion (i.e. zero down time) within a transaction? I thought concurrent index required publishing a temporary index to all concurrent transactions?

[Shrikanth Shankar @ShrikanthSS]

@gwenshap @sorenbs @sampullara @prisma I'm interested in examples of migrations that are possible with txnal ddls and whether they support data changes. Simple things like adding cols to a single table dont need txnal DDLs. More complex ones need data migration and imo txnal ddls arent enough

[Shrikanth Shankar @ShrikanthSS]

@sampullara @sorenbs @DatabaseSponge @gwenshap any example you can point me to? My claim is that if you are mixing data and schema changes, transactional ddl is not sufficient for zero downtime (and imo not necessary).

[Shrikanth Shankar @ShrikanthSS]

@sorenbs @sampullara @DatabaseSponge @gwenshap exactly my point. Its not enough and it doesnt make the larger problem easier. Oracle EBR was designed without transactional DDL because it didnt make things simpler.

[Shrikanth Shankar @ShrikanthSS]

@sampullara @gwenshap Consider splitting name column in a table to first and last name, in a single txn you can add new columns, and update all rows. but if you update all existing rows, aren’t they locked for duration of txn? What happens to new rows? How do you catchup.?

[Shrikanth Shankar @ShrikanthSS]

@sampullara @gwenshap yep, but are transactional DDLs enough? mixing data changes with DDLs is tricky while there are ongoing txns. See https://t.co/hsSEWvVmev for a different approach

[Shrikanth Shankar @ShrikanthSS]

@sampullara @gwenshap you dont need transactional ddls for updating schemas without downtime. also, DDLs on other system are atomic (not sure if thats what you meant in the second point).

[Shrikanth Shankar @ShrikanthSS]

@gwenshap not sure i agree. From what I understand, PG's mvcc behavior has some sharp edges that can cause blow ups. txnal DDLs somehow never rose to that pain from what I know .

[Matei Zaharia @matei_zaharia]

Awesome to see Oracle adopt #DeltaSharing (from @DeltaLakeOSS)! This means you can share data between Oracle, Databricks, Apache Spark, pandas, PowerBI, Excel and any system that supports the open protocol. Open makes sense for a sharing solution.
https://t.co/6yFkhZR4Lr

[Shrikanth Shankar @ShrikanthSS]

@PatHelland @_Felipe @muratdemirbas :( . It's been nearly 23 years but in my second month at Oracle I was fixing a gnarly (for me) bug. Franco walked into my cube and spent 20 minutes walking over code with me. He didn't have to but he did.

[Shrikanth Shankar @ShrikanthSS]

@PatHelland @_Felipe @muratdemirbas I still tell sr engineers something I learnt (indirectly) from him. As you get more senior as an ic you should continue todo 2 things - write design docs and write tests and test infra. These days I add one more, do ops

[Shrikanth Shankar @ShrikanthSS]

@PatHelland @_Felipe @muratdemirbas I had no idea Franco had passed on. When I started as a fresh engineer at Oracle 23 years back he was the architect for my team (along many other things). It was an honor to have worked briefly with him and he left a huge impression on a very jr engineer.

[Ali Ghodsi @alighodsi]

Dolly LLM now on @huggingface, check it out! Cost under $100 to produce on a few machines for a few hours!

[Shrikanth Shankar @ShrikanthSS]

@elektrotal tbf bheem but still vayuputra. just got done seeing it too :)

[Shrikanth Shankar @ShrikanthSS]

@prchovanec 250k wouldnt cover working capital for a construction company for eg. also who defines sophisticated? Most startups are not financially sophisticated. They don’t even get full time cfo till much later. Their focus is tech which isn’t usually fintech