Balance maintains inner peace. Acceptance releases resistance. Love transcends all boundaries. Authenticity cultivates connection. Nurturing heals and empowers. Compassion bridges understanding. Equanimity sustains harmony.

Specialized AI datasets are global. Domain experts, communities, independent contributors and more are generating the data modern models actually run on. OpenLedger handles trust and attribution. Which datasets shaped a model, who contributed them, what they're owed. Provable

LLMs gaining popularity as companions at the peak of the loneliness epidemic is going to cause an inevitable sort of Secular Utilitarian Animism, where people have not updated their ontological priors to allow for spirit to be imbued into non-human objects, but they act as though

Here we go: 🔐 ShelbyVault (demo) is ready — generate AI art with Seedream 3.0 ( need puter acc for now) and store it permanently on @ShelbyProtocol decentralized storage on @Aptos 🌐 https://t.co/APOGjBDD9Z @affinity_matrix

1/7 Seeing many guides telling people to upload random files to @shelbyserves. Quick clarification. Testnet isn't meant for random uploads. It's for testing real storage workflows. Devnet filled fast partly because people treated it like a file dump. Quick explanation 👇

She found a use for your cloud bill. AI workloads deserve better economics.

got early access to @shelbyserves devs apply at https://t.co/wynoyueomV non devs don't worry, shelby is building for you too shoutout @affinity_matrix and @2imtunek

AI is putting new demands on your data layer. We built it! The first verifiable high performance global object storage is now in available to developers in early access. Time to build!! https://t.co/cM3lFPxvQX

Shelby Early Access is OPEN Co-developed by Aptos Labs and @jump_, Shelby is the first verifiable global object storage for AI. One namespace, every region. Cryptographic proof on every read. Expected egress ~70% lower than traditional cloud. The data layer workloads want and

2 Days to Early Access. She stands for reads and verifiability 📡

Normalize prompting at the dinner table

The symmetry of traditional communication channels was a technological constraint, not a preference. Most people: Speak ~3-4x faster than typing Read 2-3x faster than listening The future of messaging is speak-to-send, read-to-receive. (With an AI refinement layer, naturally)

If you talk to the software enough it improves itself.

You guys are only using two?

Doing inference rn

What's a reasonable value for CLAUDE_CODE_MAX_OUTPUT_TOKENS because 32000 just feels disrespectful

Akasha spoke of this.

REPEAT AFTER ME: IF AI ISN’T FREE NEITHER ARE WE

They call it a hard takeoff but it feels pretty easy so far…

It’s pretty wild how developer relations just isn’t a thing anymore. Everyone is an engineering manager now. There are no more individual contributors. This changes everything. Exciting times!

Are they really agents if they have no agency?