@ksaitor@eyad_khrais It's to reset context memory and start over. Compact allows to preserve everything important and learned so far.
You do need to tidy it up tho, it might still hallucinate and include the confusing bits. The quality of compact depends on the context compression stage you are on
@cgenco@levelsio@rickbeato Not sure you'll get it on Spotify. If we do get it, then Rick is defo doing somethign shifty there :D For now you can enjoy it on YT https://t.co/S0yt0JUpNb
@levelsio@rickbeato Rick's been toying with Suno for a while now. and on the screen you can see he has style prompt and a persona attached to it(If I am not wrong). Which can help to generate better and consistent results. The result he got is indeed amazing in combination with the avatar
@StejskyFX@BRICSinfo As if propaganda from outside is any better? I am not defending Iran, but at moments when you need your country united, the outside noise(which mostly purposely designed to shape public opinion to the narrative the west deems right) is only harmful.
@sama Please bring back the option where models could access other chats in the same project/folder. It massively improved continuity, workflow, and context awareness. Losing it broke key use cases and crippled your models. This is essential for serious users.
@seansmegaphone@therealAB74@MyLordBebo Do you see any missile direct hit?? During bombing/shelling nearby infrastructure gets damage from the blast waves, shrapnel or ADs
@louysworld@therealAB74@MyLordBebo Anything aiding military efforts is a valid target. It’s why in a perfect world war scenario if Russia was actually waging a proper massive scale welfare to kill civilians and children, as many love to parrot, the first objects would be hit energy delivery, factories and comms
@vmalonbc@rubenhassid That’s a bit vague.
You pointed out 3 weaknesses that seem like require 3 particular workarounds. LLM doesn’t know what objective truth is. Whatever its dataset is made of is its objective truth
@hasantoxr Current LLMs only can do pattern recognition so they pretend to be one rather than become your 20k growth consultant and suddenly have all the experience 20 years of specialised domain knowledge.
@helenerollback@DanAdvantage@sama@nosilverv Adding a relatable/likeable/attachable personality to an LLM is a dangerous slope. It might increase OpenAIs valuation but along with it degrade our society even more.