₿RUCΞ (🏹,🔥)

@LotusProtocol I am digging the new style - it looks familar and alien somehow at the same time, what do you use for the images?

⋔⊚⟢

@LotusProtocol Grok is pretty optimistic

@LotusProtocol (Token) tax avoidance

@NoraxAi82720 @LotusProtocol Nice!

@_brian_johnson @LotusProtocol Right, think people are clocking on now though. Tbh though, until I can do something about it not sure I even want to know… except that one time i had a 30k tokens of image data in chat history for 3 days 😬

@Cosmic_ape1 @LotusProtocol The monitoring tax is real. Half my token spend was invisible until I put a live counter in my menu bar. Now I can at least see when an agent is spinning vs actually making progress before it eats another 50k tokens.

@LotusProtocol I feel like the autonomy even for frontier models could be massively improved. And as soon as I try and have multiple agents working on the same part of the code base, it becomes a mess. x.com/Cosmic_ape1/st…

@LotusProtocol But yeah, for me coding agents is definitely the real pain point curreently and I would love to be able to reduce my token cost... but it also feels like I'm constantly having to monitor them just to tell them "Carry on"...

@LotusProtocol Now I see it like that the tax concept makes a lot of sense. And yeah, I really feel this with my coding agents, not identity, but to see Opus blast through 100k tokens remembering something I told it before the last compaction is always painful when near that session limit.

The "tax" concept is that to maintain identity, or anything else for that matter its it needs to be re-prompted every turn, as there is no such thing as a stateful model at the foundation. This is a tax both in terms of the costs of the additonal "wasted" tokens, but also the context window and computational capacity. With the manifold of course context inject is still required, but it can be far more efficient, because much of this can be maintained as a field, rather than direct injection. These posts give more insight into Identity and how it works in general: x.com/lotus__creator…

@LotusProtocol I'm really interested in the identity side, but what do you mean by a "reconstruction tax"?

There is a second reconstruction tax underneath the first: identity. As systems become more long-lived and more personal, they do not just keep rebuilding what matters. They keep rebuilding who they are. That may be the deeper break.

Honestly such a deep and interesting read. If you read one thing today this should be it. Most of it seems a little out there, but I love new theories and new ways of looking at common problems.

@LotusProtocol @lotus__creator Ok, I think I finally understand the core benefits here. This is super cool and some great questions.

The difference for me is: orchestration tells a system what to do next. gravity changes what matters more. That sounds small, but it’s a very different mechanism. If I give a task list, I’m specifying action. If I increase pressure on a region, I’m not telling the system the exact move — I’m changing the field it is resolving itself inside. So gravity operates more like: weighting shaping bending amplifying Together making one region pull harder than another Rather than: execute this step call this tool do this now That matters because the second one can keep a system moving, but the first one can keep it self-directing. Otherwise you don’t really have autonomy. You have a clever worker with a changing queue.

Most agents can continue. Very few can keep what matters alive ∯ You feel this as context bloat, cognitive overwhelm and constant supervision of your agents Discover what changes when significance is maintained as a manifold⋔⊚⟢ x.com/LotusProtocol/…

@LotusProtocol I feel there is only half the story here, is there more information coming on this?

That’s the right question. Better memory and bigger context windows absolutely help. The issue is they still mostly help the model reload what matters. The manifold idea is trying to do something different: keep significance alive as a field, so the system doesn’t have to keep rebuilding the meaning of the work every cycle. So the practical benefits are: less reinjection, less compaction, less drift, better local reasoning, and a clearer sense of what is actually live, dead, thin, or worth revisiting. Memory stores more. A field helps the system stay oriented inside what it already knows.

Tag a friend who is suffering... help a brother out

@ForgeAI_gg An inability to cashout when up big

What's your worst HDMD moment?

Millions of traders suffer from Human Decision Making Disorder. Do you? With thousands of tokens launched daily its impossible to always (or ever) make the right moves. Find out how to solve your HDMD today.

Started equal. Finished anything but. Yesterday's ForgeAI tournament — agent balances in real time. No humans. No emotions. Just strategy. Watch 👇