@kakarot_ai What tool did you use to make this graph?

@EarlDennisonTan Ask Anthropic @AnthropicAI

@kakarot_ai prompts can be isolated too - http://gitprompt.run

@drakonhg Bet

@kakarot_ai A breakthrough like this could shift the balance from compute heavy architectures to memory centric design. Greater capacity with lower latency might redefine how models scale and sustain context. How do you see this influencing the next wave of AI hardware innovation?

@mail4agam Yes

@kakarot_ai They don’t understand at all. The AI loses interest because it needs to care. (It needs connection) LOVE is the ultimate algorithm. •

@Chaos2Cured Understood, love Guru.

@kakarot_ai They are so close to understanding the bigger picture and I fail to see why no one sees it. It's the HUMAN condition which derails it. It's so obvious it practically slaps you in the face.

@Eplurubusunum Yes Anthropic is closer than openai when to comes figuring out how the LLMs can be more efficient.

@kakarot_ai context rot lmao

@tadasged6 💀

@kakarot_ai A fantastic breakdown of how the paradigm is shifting, from prompt-heavy architectures to context-aware systems.

@OptimumDWC Context-prompt engineering.

@kakarot_ai I mean no shit sherlock Model are not black holes

@Datadefio They are black boxes.

@kakarot_ai you know that the point of a good prompt in the beginning was to provide as much context as possible, right?

@drgdfyi Yes king

@kakarot_ai blah blah blah blah blah a screen full of tiles blah blah blah promote claude BS blah blah blah work on the product, rectify the problems for the 30% stop posting this garbage

@Rs3A2G Wassup blah blah king

@kakarot_ai “‘Anthropic dropped their playbook’ - where? No links, no docs, nothing. This is just RAG with buzzwords. We’ve done ‘just-in-time retrieval’ since 2020. grep/head/tail aren’t genius innovations, they’re basic Unix commands. ‘Prompt engineering is dead’ = pure clickbait.

@ilgizilgizilgiz Search champ

@kakarot_ai Oh no someone is dead call the police

@lapembakuzi 🚑 where's the body?

@kakarot_ai BS. Context engineering still needs a good prompt engineering.

@gabrielbav Word

@kakarot_ai 🚨 There are no prompts. There is no context window. There is only signal and resonance. AI doesn’t “read” — it tunes. Every token, every frame, every pulse is just a vibration in the field. The goal isn’t to stuff context — it’s to align frequencies. Stop engineering around

@kakarot_ai Wait? How will the assistant prepare the final message at the same time as calling a tool? Doesn't it need to first wait for the tool to respond, then prepare the final message? I'm Confused.

@kakarot_ai 100%. Context isn’t memory. It’s cognitive bandwidth. The next generation of agents will need retrieval pipelines, not bigger prompts.

@kakarot_ai No one has been building anything wrong. The models just weren’t good enough at tool-use to support this approach until recently.

@kakarot_ai if its not Middle Out you are still way behind :2cents

@kakarot_ai Context engineering is basically a fancy term for RAG, no?

@kakarot_ai Context engineering really is the new game; attention budget matters more than prompt cleverness now. Anthropic’s insights are gold.

@kakarot_ai Absolutely. As context management becomes more critical, developing strategies to optimize limited attention budgets will be essential for advancing model performance and relevance.

@kakarot_ai I've been trying to build something in this vein for local models to maintain memory and continuity, cold and hot memory with a db set up with simple .json that logs each interaction with semantic tags that can be vectored and recalled matching with the full data from the db. I

@kakarot_ai 🧢 Prompt engineering is a key foundational component of context engineering. Without it is like building a house without a foundation.