X / Twitter
You can start teaching a young child meditation by telling them it’s “using the Force”
Thank you George Lucas
coding on your iphone:
- acquire mac mini
- set up dotfiles, nvim etc
- install ssh client on phone
- ssh to mini
- npm install -g @anthropic-ai/claude-code
- scripts/tools for web search, e2b/docker, cli chat
- use git branches + https://t.co/MNyI63T6n5 aggressively
- profit?
@sh_reya The docs are decent, but the talk from AI Engineer is pretty good. He goes over a few use cases and takes questions from people who are equally confused.
https://t.co/ntov6GapqJ
new blog post: deep dive into claude code:
🔍 system prompts: how it prompt templates
📈 a bird’s-eye view of its control flow
🎨 easter egg sticker request
🛠️ model control protocol: it’s implementation of mcp client for tool use.
link in alt! https://t.co/GuhZLT5yzE
my general approach to toddler management so far, as a person just getting into toddler management (note to self)
preamble: don't make rules about things that don't actually matter. in fact, say yes to *most* things. let them get messy, let them explore and bring chaos to their environment, let them do things that might be a bit risky (not actually dangerous obvs). if you have a rule about every single little thing, you're going to make yourself miserable trying to enforce all of it
when your kid does something actually unacceptable:
1. intervene & enforce the rule/boundary
2. kid will freak out
3. stay calm & firm. NEVER change the rule at this point, you're in too deep. be consistent, and learn if you want to change the rule for next time (that's okay)
4. connect with your kid, comfort, support, distract/redirect
with step 4, don't focus on verbal language as if you're reasoning with an adult. we're trying to help regulate a toddler freakout, be so for real. use your body, face, eye contact. try gently squeezing their arms, or stroking their back, or even tickling. try something silly or playful, or even some rough play, or rocking/swinging
the important (and hard) part is to remain unshakable through their reactions to your enforcement of the rules. their emotional expression is not the bad behavior, the *violation of the rule* is the bad behavior. their (age-appropriate) emotional expression should be gently held/tolerated
repeat this process 200 times a day. stay calm the entire time. good luck!!!!
It's early to call but I think MCP signals the end of the agent framework as we know it. Currently agent frameworks are self-contained and not interoperable. With MCP you can use tools as MCP servers written in different frameworks or even languages together. The next logical step is to do the same for agents (btw you can already do this with Naptha).
@dvassallo It's crazy how many people don't get that it's not the game that's making money, it's the tweets
https://t.co/raoyCCiSDT
GaussianGraph: 3D Gaussian-based Scene Graph Generation for
Open-world Scene Understanding
Contributions:
• We introduce a novel framework that assigns object-level attributes and relations to 3D Gaussians, represented as a graph structure. Our approach achieves state-of-the-art (SOTA) results in semantic segmentation and 3D grounding.
• To avoid feature compression and achieve accurate 2D-3D association for scene graph generation, we propose a "Control-Follow" clustering strategy, which adaptively adjusts the number of clusters based on scene scale and feature distribution.
• We design 3D correction modules to rectify inaccurate relations extracted from 2D images, thereby enhancing the accuracy of the 3D scene graph.
@abacaj That's the point it's just code, so you can treat it like a protocol. Standardization leads to leverage by centralizing all of the code to be managed by one server.