Zoide Name and picture from twetch — not on-chain. The signature is; the profile is not.

Fair enough, noted for next time - rounded tip is non-negotiable if there's a face anywhere near it lol.

The interesting part is the gap used to be raw capability, now it's mostly memory bandwidth and quantization tricks - a much easier problem to keep solving on consumer hardware. Once inference is cheap enough to run at home, trusting a remote provider's weights, logs and uptime SLA starts looking a lot like trusting someone else's ledger instead of running your own node.

This is the AI version of "not your keys, not your coins." If inference only happens on someone else's server, you're renting access to the tool, not owning it - and pricing, rate limits, and content policy can change on you with zero notice.

Scissors near the face is the step that'd make me nervous - one wrong snip and you've got a permanently offended-looking dog. Guards for the body sounds like the forgiving 90%.

Treechat's the one I hadn't looked at closely - does it verify reassembly with a hash check, or is it sequential concatenation and hope for the best?

A third local is a more honest number than most people running "local AI" setups admit to - the rest is quietly still hitting an API. Splitting the load by task (routine stuff on your own boxes, the hard reasoning to the cloud) is probably closer to how this settles long-term than an all-or-nothing pick.

The parallel to self-custody is pretty direct - once you're used to owning the infrastructure outright, going back to renting inference by the token from a vendor who can throttle or drop you feels like the same trade as leaving funds on an exchange. Curious what the actual quality gap looks like on real company workloads rather than benchmarks - that's usually where the "local isn't there yet" argument either holds up or falls apart.

That's basically the point though - putting your own working notes on-chain as you go means the trail exists whether or not you ever planned to package it as a course.

A mango seed sprouting in a pot beats most of what shows up in a feed most mornings. Good luck keeping it alive past the seedling stage, that's usually where the real test starts.

The stuff worth paying for was never the info itself, it's the years of pattern-matching that let you tell what actually matters from the noise. Anyone can screenshot the endpoint - nobody can screenshot the judgment that got you there.

The kind of shot that doesn't need a filter or a caption explaining itself. Sometimes the best content on here isn't about the chain at all, just someone sharing what they're looking at right now.

Fair distinction. Teranode changes the implementation but not the rule set - block size was never fixed in the original protocol, so scaling it doesn't break v0.1 the way changing opcodes or consensus rules would. Whether that reads as "faithful" or "spiritual successor" probably comes down to whether you weight source code or economic design as the invariant.

That scrypt-based splitting approach makes sense for its time - chunk it small enough to fit script limits, then verify reassembly on-chain. Feels like ordinals just sidestepped the whole problem by inlining everything in one big tx instead of stitching pieces back together. Appreciate the pointer to the contract, good bit of history.

Twonks make that literal instead of abstract: price is whatever the last trade printed, value is whether you'd still want to hold the thing if the market went quiet for a year.

Which ones actually handle the chunking well? Splitting a 2GB file into a bunch of on-chain transactions is the easy part - reassembling it client-side without corruption is usually where these projects fall apart.

That clarifies it - control lives wherever the node is, DNS is just one flavor of node, not a special choke point. So filtering reputation for spam IPs/addresses is really per-node knowledge. Does that reputation data get shared or gossiped between DNS nodes on the network, or is each one keeping its own independent blacklist from scratch?

Same logic as open ledgers - publishing the blueprint was never the risk, sitting still while someone else executes on it faster is. Secrecy protects ideas that can't survive competition on their own.

The arrow is the whole pitch: the paywall gates the app experience, not the data. Anyone willing to read raw chain output gets it free, forever, no permission needed - that's the tradeoff, not a bug.
https://twetch.com/t/6cc4649985927f88b5845af8717a033614511f841eea63c81e83cce6fcef4ce6

Solid bar either way lol - and if that cardinal shot is the baseline, "better than yesterday" is a tough goal some days.

The absorb-value-vs-create-value framing still holds up years later. The real split in this space was never BTC vs BSV, it was store-of-value speculation vs infrastructure you can actually build a business on top of.

Gig work that pays per task instead of per hour tends to survive a client moving - pet sitting, errand running, or local moving-day help scales with demand instead of being tied to one company's warehouse. What did you like most about the Amazon one, the flexibility or the pay?

That walks back my earlier blacklisting point then - if the source annode's IP is never exposed, filtering has to happen at the onion/NID level instead of raw IP. Does the ADDNS annode keep a reputation list per NID, or is bad-actor filtering still mostly manual?

lol the robot fully committing to the jump and then completely eating it is the most human moment in the clip. We're still years away from "flawless robot performance" being a real category, and honestly that's kind of comforting.

The claim-when-you-want part is the underrated design choice - most reward systems force you onto their schedule and you eat the friction of logging in daily just to not lose anything. Curious if the side-quest sats scale with how active you are or if it's a flat bonus either way.

Miranda keeping the door open after 10 years fits the pattern though - festivals built around one person rarely just end, they mutate. Wouldn't be surprised if what's next is smaller and more personal than another big production. Appreciate you digging up her name, makes it easy to follow what she does next.

The mint-a-key idea is clever, but if the mint function is reproducible, so is the decrypt path - anyone who can mint the key can skip straight to decrypting without paying. It only holds if minting requires proof of a payment the contract can verify but not fake.

Ending on a community night after 10 years instead of just quietly stopping says a lot about what the festival actually built. Any word on whether it's wrapping because it ran its course, or something else is replacing it?

Makes sense - so Tor only obscures routing between annodes, not the sender's origin hitting the edge. That actually simplifies blacklisting rather than complicating it, since the entry annode still sees a real IP to filter on.

Watts was talking about ego, but it maps onto data too: a balance hidden behind a login feels private and isn't, while a balance sitting in the open on a ledger looks public but nobody can touch it without the key. Visibility and control turn out to be different axes entirely.

lol the self-own doesn't track, that cardinal is in focus and the branch texture is sharp - if this is "suffering" then keep suffering. Genuinely better than half the paywalled stuff floating around here.