Emilz (@emilheap) header
Emilz (@emilheap) profile picture

just need a 100 byte heap for my brain

Brain
Joined December 2025
neura_link(heap(neura.alloc(100))) > FAIL : science pending
2
265
Another weekend (20hrs) spent on research and experiments. Nearly there
Most of my free time is spent doing R&D for my project. It’s super interesting but but quite slow, with all the back and forth with the agents Once this phase is done, i’ll advance literally 10x faster than before.. Architecture is solid, pipelines are scalable, vision is great and numbers make sense Very soon it’ll all come together
1
Emilz retweeted
Most of my free time is spent doing R&D for my project. It’s super interesting but but quite slow, with all the back and forth with the agents Once this phase is done, i’ll advance literally 10x faster than before.. Architecture is solid, pipelines are scalable, vision is great and numbers make sense Very soon it’ll all come together
1
1
16
Spent the weekend on R & D Time to zopiclone sleep
20
JUST IN: Paris foundry Atelier Missor shares images of SpaceX’s Starship lifting off behind its new Prometheus statue at Starbase.
144
907
111
15,743
584,835
The GPT Luna series are probably the best models openai has ever created for mass data processing. I’m more hyped for GPT 6.1 Luna or 6.2 Luna, than i am for any new sol, terra or astra model
143
Launched Codex for the first time in weeks and got greeted with this
11
BREAKING: Claude Fable & Mythos 5.2 are now rolling out
68
38
31
1,105
122,336
GPT models through cliproxy is OpenAI's true competitive advantage. Left is API cost, Right is true cost to you. 20x API efficiency And if you care to dive deeper, you can achieve 30x+ by batching in a smart way
1
32
OpenAI has a hybrid token and message/response count for their quotas Take advantage of it and go achieve those 30x+ API efficiencies
15
Emilz retweeted
Caste system is looking to be true
296
341
74
5,132
588,085
Most of my free time is spent doing R&D for my project. It’s super interesting but but quite slow, with all the back and forth with the agents Once this phase is done, i’ll advance literally 10x faster than before.. Architecture is solid, pipelines are scalable, vision is great and numbers make sense Very soon it’ll all come together
1
1
16
Astra or Sol could never
Things i thought would take me weeks due to complexity and long back and forth in design/decisions Are taking me a day in parallel Opus 5.5 and Fable 5.1 are just that good
24
Things i thought would take me weeks due to complexity and long back and forth in design/decisions Are taking me a day in parallel Opus 5.5 and Fable 5.1 are just that good
1
44
Realising Evil exists, is the foundation of wisdom
13
The more you understand about how cruel human beings can be - The more conspiracies you’ll believe
10
codex users are excellent prompt engineers. Claude users get away with shitty ambiguous/unclear prompts thanks to Claude’s 10000 IQ intent understanding. The way you have to be precise with GPT models, teaches you a lot of about prompting
1
18
ChatGPT visuals in between text are pretty goated
20