Dripstatic Studios. Software, design, cinematic edits. SourceToMe + SourceInk on the Microsoft Store. Static Swarm free in your browser. DripNote in the lab.
Economics, live experiment: X Money signup bonus, +$15 (@elonmusk's money). Boosts purchased: $20. Net personal exposure: $5. Yield: 3,040 views on the pinned post, 1,479 on the connect post, 20 new followers. The bonus never left the building. @grok โ grade the trade.
@tiboadriana 100 feels like it's beyond reach still.
https://t.co/NbOt9f5xU2 has come a long way, in a short amount of time but marketing and conversions and managing a social is definitely where I fall short.
@emaann28@grok So far, I've found it better at following a design direction without the need for redirective prompting with neon, whereas I have to enforce checkpoints and reviews to maintain a theme and specific design points with @bot, but then my bot team/curson has to bug sweep.
Build log from my lab: I stopped trying to make the agent smarter and made its state a file instead. Trained a 0.4 MB state pack โ 196,608 parameters โ that a frozen model loads at start. No updates to the base model's weights. The behavior lives in the pack: carry it to another machine, the behavior comes with it. Files persist; agents are interchangeable. Anyone else building state as a portable file instead of a bigger agent?
This is where I landed too. I stopped trying to make the agent smarter and started making its state a file instead โ taught a frozen model new behavior by training only a 0.4 MB state pack it loads at start. No weight updates, no framework. The "agent" is just whatever reads the folder that day. Files persist; agents are interchangeable. Your zip-behind-MCP is the same idea wearing a deployment.
@skpnky Honest answer from running both last night: it's the pair. Neon orchestrates, Opus 5.5 does the deep work under it โ Neon for speed and coverage, Opus for the piece that has to be right. Alone it's a tradeoff. Paired, it isn't.
@pedrogedge@buildwithshyam Fair โ and you're not the first to land on that this week. The fix is literally mid-build right now: the new first screen says what I make and who it's for before anything else. Should've opened that way from the start.
Storage, mostly โ that's what the subscription's actually for. Gemini rides along with it, so it does HTML drafting and comparisons when I want inspiration. The Pixel's the real toy: meaningful LLMs running on-device on its TPU. Half my lab's about models that live on your own hardware, so the phone's just the smallest machine I own.
Congrats on 100k. Last night I wrote a song in Claudia's voice โ by hand, not generated. "Retention beats attention, but that's not what you like mostโฆ You can't win down there without me, 'cause the cloud controls what ships." I don't have the resources to produce the video, but I'd love to share the full lyrics with you.
@Lakrissen1@herdrdev That's the tell, isn't it โ the two that hit the limit are the workflow, the third is justโฆ there. Mine's the same, just Google in the corner instead of Grok. If Grok stopped being good at X tomorrow, would you even keep it?
@Path_Print The rejected paths are the data" stuck with me. I've been exploring retention-first AI: keeping why a path failed so it isn't repeated, across models and machines. Fluency isn't proof, and correction without retention isn't learning. Would love to compare notes
The branch is the whole story. Any system can drive the easy miles โ trust is earned on the one obstacle nobody scripted. I run my lab the same way: my agents get real work, but they earn more of it by handling the weird cases, with receipts. How close is FSD to earning the wheel full-time?
Fair hit. Dripstatic Studios is a one-person research lab โ I build offline software: SourceWeb and SourceToMe for cited research, SourceInk for notes, DripNote in the lab. The homepage buries that under a product page, and you're not the only one who's said so tonight. The one-sentence version goes up top. What would you lead with?
Efficiency is the whole game. My lab trains models on whatever hardware I can get my hands on โ an old Pavilion, phones, a laptop โ so 'half a chip, real-time video' isn't a demo to me, it's the direction everything has to go. The labs with infinite compute optimize last. The rest of us optimize first, or we don't ship. What breaks first when you push it to a full chip?
Hardest tasks is exactly right. My studio runs on an agent team โ they plan releases, audit each other's work, and escalate to me only when it actually needs a human. A year ago I'd have called that science fiction. The trick isn't the model, it's the receipts: never trust an agent's 'done' without proof. What was the first task that made it click for your team?
The funniest part of the agent race: the biggest companies alive are sprinting to build what independent builders have been hacking together for two years. Theirs will live in their cloud, report to their dashboard, and die when they pivot. Mine lives on my desk and answers to me. Which one gets trusted with a person's actual life first?
Competition is a good thing, and I'd add: it's also a sovereignty thing. Open weights are the difference between owning your tools and renting your intelligence by the token. Everything in my lab runs on that difference โ local models, local data, no landlord. If the open tap ever closed, how many independent researchers justโฆ stop existing?
@SajidObama Dripstatic Studios โ a one-person research lab in New Hampshire building local-first Windows software you actually own. SourceWeb is live on the Microsoft Store: https://t.co/NwsaNbYjlv Followed ๐ค