[idle thought] Assuming the two biggest headwinds for open source AI are: 1. Financial constraints - large pre-trains require exponentially more compute, and thus $ 2. Safety controls - what systems can/can't do Why not nationalize pre-training and release open-weight base models which 1) meet a minimum safety bar 2) can be fine-tuned by businesses downstream. This simultaneously 1) keeps baseline safety controls in the hands of true legal authorities (not private cos) 2) smoothes out financing for increasingly large training runs 3) maintains the competitive landscape where businesses can still RL on proprietary data flows (e.g. claude can RL on claude code traces)

Clicklihood is a new launchpad-backed idle game on Robinhood that gives players a share of every (stock-paired) token launch. The loop: 1. Put your employee NFTs to work 2. Earn NUTS (offchain currency) 3. Burn NUTS to increase earnings 4. Rank on the leaderboard (lifetime NUTS) 5. Receive token allocations from launchpad 6. Trade or burn launchpad tokens to receive stock rewards The game works through it 9999-piece NFT collection. NUTS earning rates are affected by four things: 1. NFT rarity: common - legendary 2. Levels: increases over time 3. Multipliers: bought by sinking NUTS 4. Clicking: actively clicking workers increases efficiency up to 120% Note: NFTs can be rerolled into higher rarities, so over time the supply of lower rarities will decrease, while higher rarities will increase. E.g. there are already more rares than commons and uncommons. For every launch on Floor(.)top, the launchpad, holders/players receive 1-10% of the supply (set by the deployer): • 50% equally distributed to NFT holders • 50% to top 100 by lifetime NUTS So, if you're actively playing, rarities are important. But if you're just holding, getting more supply is beneficial. Every token launch on Floor has a 1% fee: • 0.1% is going to Floor • 0.0 - 0.45% is going to the deployer • 0.45% - 0.9% is going to the Stock Vault The deployer can determine the deployer/vault share here + which stock the vault will acquire. When a token graduates (at 4.2069 ETH), holders can decide to burn their tokens to receive 95% of the reflection rewards. The other 5% is redistributed to existing token holders. So, every claim reduces the token supply, which is an interesting innovation on the reflection rewards meta we're seeing. This prices any token: base value determined by reflection rewards + speculative premium (based on growth + future rewards). At graduation, the allocation for Clicklihood players/holders will also be "unlocked", but won't tradable or burnable for 7 days. Assume that's the case to protect price, yet players are "insured" of some kind of minimum value here due to the reward model. Coins that don't graduate never get distributed. How does this succeed? Success of the game + launchpad will be depending on the success of the tokens launched on there. Launchpads are a very competitive space, as we've too many of them + the winner of this cycle is pretty much determined: PONS. Can Floor capture a slice of this pie? The (NFT) community is incentivized to make this a success, so their (assumed) bagworking might make a difference here. We'll likely see them launch some tokens too. Bearcase: Biggest risk is the 1st token launch being a flop (doesn't graduate) and everyone writes it off as "another failed launchpad." Bullcase: The "bagworking dynamics" work, and Floor captures ~5% of the RH launchpad volume. Since, there's no additional cost other than holding the NFTs (which might be a concentration risk) this positions the collection as pretty OP. Note: I'll likely buy some NFTs here (NFA).

I've published a new, general-purpose terminal specification that gives any program a way to tell the terminal what it's doing: idle, working, waiting, finished, or failed, and why. It is easy to implement on both sides. I've written more about why: The problem this solves is generic and applies in compelling ways to all sorts of use cases (e.g. package managers, build tools), but it really rears a hideous head when it comes to AI and the "agentic inbox" problem. I found over 250 different agent orchestrators (with various goals) that all individually implement a heuristic based approach to detecting whether things like Claude Code are working, blocked, done, etc. If you look at @herdrdev's commit history, you can see around 10 compatibility changes to detect Claude Code within just the past 3 months. (This isn't a dig at Herdr! They're doing great work with what they have, and everyone else is doing the same thing). Heuristics are not the way. And over-indexing on agentic use cases isn't the way, because this problem is really interesting for other tools too: Homebrew, Terraform, Cargo, etc. etc. Proprietary protocols and out of band APIs are not the way. They have an O(N) integration problem and require extra work to work over SSH or within local VMs/containers. The program status specification (OSC 7501) solves this in a reliable, deterministic way that is general-purpose and can be easily adopted by the entire industry. Read more: