Adrian Cockcroft#SW2con final thoughts. It’s a small event, doesn’t have crowds of attendees but with great content and hallway track it has enough to be worth it. I plan to be back next year. There were about as many people as last year’s Gluecon, although there have been much bigger Gluecon events in the past decade or so.
#SW2con had fun presenting my talk, and finished on time! Too much content - this is one of the talks where I “didn’t have time to write a shorter talk” as the saying goes.
Rachel Stephens#SW2con I’m up next - Thanks for the Memory - looking at recent changes in GPU architectures and the implications of that. Some thoughts on Enterprise Indigestion - as the pace of change in AI is too fast for most enterprises to consume. Also asking for advice/help with my meGPT idea. https://github.com/adrianco/meGPT
#SW2con Matthew Fields CEO of VMAccel talking about AI moving to the edge. In my opinion this is happening, but it’s in addition to growth in centralized AI not instead of it.
#SW2con Talk was short and could have gone into a lot more technical depth, there was some good Q&A afterwards.
#SW2con Marc Austin of Hedgehog talking about why AI needs a new network. Discussing Ethernet based solutions prior to the move to the high speed Ultra Ethernet Consortium vs. using Infiniband.
#SW2con Rob Zuber CTO of CircleCI talks about how to detect AI hallucinations.
Listening to Paige Bailey
talk about tradeoffs between small and large language models in terms of cost/latency vs quality of output. #sw2con
Small models today can compete with large models from 6-9 months ago.
She thinks smaller models augmented with retrieval is probably the sweet spot.
(also her general rule of thumb is that code assistants need to roundtrip in <500ms.)
#SW2con interview between Heather Joslyn of The New Stack and Paige Bailey of Google - discussing some of the new Google AI announcements and other recent news.
