Gratz

@LeynaMusicx @Vishakha1801 maybe this is the answer to your recent question on why someone would want an open source model

@Vishakha1801 agreed, feeling the same. there's gotta be a middle ground between helpfulness and hand-holding that most people would actually prefer

unfortunately while claude is still my favourite model it seems to be getting increasingly annoying to use bec of anthropic overcorrecting on trust and safety. this i think is a hindrance to usability. i wish there was a better way to do this. i dont want to be lectured by my ai

@benlsage @xandervancil29 i disagree its a storage issue. i use claude code, not sure what youre on, but if its missing key information i tell it to examine the json of the other session to pull information. i save state with deep analysis of each session too. pulling is harder but skills can be p deep

@graematz @xandervancil29 All three imo

Most obnoxious part of vibe coding? Re-explaining the whole context to your IDE after switching chats

@xandervancil29 @benlsage @benlsage do yall think its a capture issue, a resurfacing issue, or a model issue?

@graematz @benlsage Yeah... Never works as well as I want

@BraedendotTECH replies looking like linkedin id blow my shit off

𝕏 gets way better when your feed is full of builders. People shipping projects. People solving problems. People obsessed with tech. Looking to connect with more people into: AI, SaaS, coding, startups, web dev, engineering & tech. Let’s connect 🎉

@xandervancil29 @benlsage state saving and handoff docs exist. project tracking in .md files

@benlsage who is working on a fix?

@babushkacc @sunlove I need to see your markdown project’s ui, your taste is so cool

@sunlove this hurts…

never forget what they took from us

@valhalla_dev Need lots of memory so price constraint. Apple 96gb unified seems like the system I’d use? That plus glm, id run skills to attempt higher cognition. I’m an obsidian glazer and thats the system I have so I’d tie that in? I don’t have this built out tho

Why would I or someone else not build am agent harness with open source models specifically built for offensive cyber operations so the next time the government or Anth or OpenAI says "we can't release this model it's too dangerous" they have way less of a leg to stand on

i feel like opus 4.8 is smarter until you start to see cotext rot and then its a lot dumber than 4.7 deep into sessions

@soulblocks lemme hop in

might try to make another GC again for nerds who aren't sleeping idk

this "ai not creating good designs" could be a specified taste corpus mismatch. ai is trained on shit as well as gold, so maybe it averages out? maybe if we pass enough priming context into prompts it can create phenomenal design?

got blocked today bc i recommended an artist vibe code video game mechanics instead of learning a new game engine. so interesting to see anti ai sentiment in the big q2 2026 i feel like at this point its clear that ai doesnt have direction or vision. it builds perfectly but i still have to constantly feed it direction. even when i have self improving systems, the things i want built come from me and the ai makes it. i havent seen standalone ai create a design that feels inspired. though this isnt unique to ai, i think 99% of creators dont create anything that feels inspired either. i feel like the proof of this shows across categories. nonzero taste, but most music from 2001 is worse than daft punk's discovery. curious where the state of art is moving forward. i have a gut feeling its relatively unchanged, just easier to create. means more slop, but means that high production capacity artists can create more which i think is good for humanity

I think cool people vibe coding on twitch is an underutilized niche

The human brain storing memories is a great starting point, I have an article on my page about the theory of congruence which is kinda similar, not tryna sell you anything but we’ve landed in a p similar place. I agree that it’s abt storing memories and loading them. The big thing imo is abt loading some key files into a startup skill that give the ai the context to call other related info.

@clankeruser Mine loads insights, past decisions, and current tasks into context at start and after that does searching via a vectorized index of the markdown files. I wonder how close our intelligence systems are

@clankeruser What’s your memory solution?

@thelemicphase Algo knows you too well