pingder

the campaign pipeline on @TheDAOLabs is built so each stage hands the next one something it can check. onboarding fixes the format first. a dedicated hub means a social mining space reserved for that project, a guest project runs inside the existing setup at lower deployment cost. campaign design produces the spec: tasks, content direction, deadlines, requirements and a reward budget. everything later is measured against that spec. execution happens on whotweets. miners register, pick tasks, complete them and submit a link as proof. the link is the unit the rest of the pipeline works with. review takes those links. the system and human reviewers check them against the spec, and spam or fake interaction is removed here. distribution only reads approved items. points go out for completed work, rep moves with recognized quality. the design choice i rate highest is that no stage trusts the one before it blindly. the spec is explicit, the proof is a link, and rewards wait for review. #DAOVERSE

points and rep on @TheDAOLabs get mixed up a lot, but they track two different things. points are contribution points. a miner earns them by completing tasks, they accumulate as tasks get done, and they are the part tied to rewards. rep is reputation. it measures quality and reliability, and it rises as contributions get recognized for being good. it is not a reward balance and it does not convert into one. the two still interact. rep can influence a miner's opportunities and reward multiplier, so the reputation side feeds back into what the contribution side is worth. from a system point of view the split is sensible. points answer "how much did this miner complete". rep answers "how far should the system trust this miner's work". one counter for both would mix volume with quality, and quality would be hard to see. both are tracked on whotweets next to the tasks, so a miner can watch them move separately. #DAOVERSE

what a project gets from @TheDAOLabs is the whole campaign setup, already running. the project's job is mostly deciding what to ask for. the rest sits with dao labs: - designing the social mining campaign - building the task system on whotweets - tracking each contribution - validating and evaluating it - distributing points and rep to miners building that in-house would mean a task platform, a way to track posts on x, a review process and a reward system, all before the first contributor shows up. the review layer is the piece i'd least want to rebuild. submissions are checked by the system and by human validators on three inputs: whether the post meets the requirements, how good the content is, and the engagement it gets. spam and fake interaction get dropped there. so a project joins a setup that already works and spends its own effort on the product and the brief. #DAOVERSE

Quick question for anyone who has run a campaign: how many of those likes were still around a month later? This new post calls that fading activity ghost metrics and argues real contribution is the kind other people can check. Research, explainers, consistently showing up. In #SocialMining, Rewards from @TheDAOLabs follow work that passes review, which feels like the right incentive

Opened the Barbera Can and the smell is strong! Intense, a little chocolatey. Moka pot is going. This is the one I'm most curious about taste-wise. Final #Coffee of the @Barbera_1870 samples, let's see.

Engineers will feel this one in their bones: there is no single line to open and audit. Every debugging habit we have assumes the fault has an address, a file, a function, a commit. Training spreads behaviour across trillions of weights, so that address simply does not exist, and this write up from @TheDAOLabs admits better tools may not close the gap soon. Worth a read for anyone in the #DAOVERSE who still treats interpretability as solved

Breakfast at my favourite Banh Cuon Trung spot, with a hot cup of Oro I brought from home 😅 Soft rice sheets, egg, warm broth, and a smooth #coffee that doesn't overpower any of it. Perfect Lạng Sơn morning. Thanks @Barbera_1870

Two curves moving in opposite directions explain most of this thesis. Per the figures the article cites from Max Munro's video, model capability gets about 13 times cheaper each year while the cost to build frontier AI roughly triples If both hold, selling access at $20 a month cannot fund multi billion R&D, and margin migrates to whoever owns the pipes. A clear technical framing from @TheDAOLabs, and a good #SocialMining exercise: check those two curves against your own sources

#GoodMorning! Day 2 of trying the @Barbera_1870 samples. Today it's Baron Blue Java, 100% Robusta from a single estate in Java, Indonesia. Apparently they only make about 50 bags of it a year, so no pressure 😅 Let's go 💪

Peak numbers on a spec sheet are the most misleading figures in compute, and the third part of the DAO Labs compute series explains why in one paragraph FLOPs measure floating point operations per second. In theory they let you compare NVIDIA, AMD Instinct and custom TPUs on one scale. In practice that theoretical math gets stranded. If memory bandwidth can't feed the chip fast enough, or the compiler is weak, the arithmetic units sit waiting So a FLOP figure tells you the ceiling, not the floor. It is the same reason the GPU-hour, despite its own flaw (time doesn't guarantee throughput), stays the default contract unit, with indices layered on top to normalize the gaps The takeaway for anyone writing #SocialMining research on AI networks: quote peak specs only next to the bottleneck that limits them. Readers in the #DAOVERSE deserve the floor as well as the ceiling, and @TheDAOLabs gave us the vocabulary to say it

Okay, first cup of Mago done. So smooooth! Like, properly smooth, not bitter at all. I expected something super intense because it's Italian, but it's more round and cozy than strong. Nice start to this little #ItalianCoffee tasting from @Barbera_1870.