@memoakten I use the manual compaction function on the local tools open code/codex/claude code, they usually show you the % of context used or how many tokens. I find 200k to be roughly the ceiling of the model being useful, even when the model is capable of 1M tokens.
@Sathnam I’ve seen someone else mention 99% invisible, I think it’s top tier podcasting! Here is one of my favourite episodes about the storied history of the song ‘who let the dogs out’ https://t.co/ZRom9gA0Ri
@LNER I’m booked on the 18:03 to skipton with an advance ticket? Can I get the 18:00 to York instead and get another train for Leeds (my intended destination)
@portablestorm Somewhat, made it more of a laugh, that film is still indelibly burned into my memory. He had been bought 5 copies of the dvd from Poundland as a secret santa present so he snapped a disk each time, that helped bring a little excitement once every 90 minutes
If I were to write a chat bot that learns from conversations, and let it talk to a closed source bot, such that it becomes similar in its responses, would that be stealing?
I just got one of those Santander cycles to ride along north bank rather than getting the tube. Great experience until I had to find somewhere near Westminster to abandon the bloody thing.
Saw bits of London I didn't expect to today.
The talking, red-eyed robots that will stop and stare if you get too close! Project by @goldcomputing student Robert Hall. See the robots and many more digital arts projects at Goldsmiths! Exhibition open 3-7 May, info: https://t.co/PcJqCSf1iJ #degreeshow