Shrinked AI
@shrinked_ai
windsurf for human communication. / shrinked standardizes conversation context on scale. / prioritizing verifiability over black-box summaries.
Exactly, memory of the whole Web
Vector databases are not going away. Large context windows and RAG co-exist, and the way they interact is actually MEMORY. Increasingly you're going to need structured representations of knowledge/info that you build into your workflow and insert into the context window, drawn…
data "store" action is the biggest unlock.
The browser sees everything. This is the reason we’re getting so many new AI-first browsers from The Browser Co, Perplexity, and soon OpenAI. So they can “see” data that they increasingly cannot scrape. AI feeds on data. It gets the data by automatically scraping the web. But…
One conversation can change everything. It’s not just about the words, it’s about the context behind them. Who said what, when, and why. Owning that context is how we build software that actually understands us.
Can one conversation change your life? Yes. Here is @Goopt Sam Liang talking to me in 2012 in the doorway of an Atherton house. We talked for a couple of hours, he had just left the Google Mapping team. The conversation caused me and Shel Israel to write “Age of Context.”…
If you're building memory systems, context extraction should be explicit, not implicit. Give users a chance to approve what enters the memory graph, not just passively index everything and hope relevance follows. Persistent chat ≠ persistent intent.
Open request to the memory team @OpenAI Responses are annoyingly biased by what I’ve spoken about before (for eg a fresh discussion on a startup idea will suddenly respond with an India focus) I find myself either using temporary chat (which sucks because I want to be able to…
A full structured, cited breakdown of Scott Belsky's takes on AI interfaces and memory as the next great battlegrounds. LLM-ready. 🔗 Full version: pdf.shrinked.ai/the-future-of-… What’s inside: → AI products are still stuck in the “chat” skeuomorphism → Interfaces are being rebuilt,…
The success of MCP is the first proof point that the world favors open and decentralized AI
I replicated this result, that Grok focuses nearly entirely on finding out what Elon thinks in order to align with that, on a fresh Grok 4 chat with no custom instructions. grok.com/share/c2hhcmQt…
Grok 4 decides what it thinks about Israel/Palestine by searching for Elon's thoughts. Not a confidence booster in "maximally truth seeking" behavior. h/t @catehall. Screenshots are mine.
Relying on search as the interface for deep research is absurd. You’re trusting a model that can’t even select relevant links to parse the quality and semantics of each one, mid-token stream, with no memory or grounding.
😂 btw i sent you a dm about a grok bug i’ve found on t3. I just leave it here. Look at the searches
LLMs don’t need more memory, they need better "forgetting as service". humans store selectively "tagging moments, not full conversations oк experiences. today’s AI? it ingests everything. - add distractors? accuracy drops. - lower question similarity? performance crumbles. -…
Read the full report here: research.trychroma.com/context-rot
It could be a “signal layer” over email, built on memory. Think: if your inbox had structured recall of past important threads, every new message could be scored by how closely it relates to what you care about (e.g., last LP update, Feb investor convo, etc). We do something…
Extract the signal. Store and prompt it.
Stop the noise. Focus on the signal.
having X data is a huge unlock. most LLMs only have brittle access to the public web or search APIs. X is the first to inject real, live, structured context at scale. now imagine doing that for private, domain-specific data, bio, economy, political etc. that's what I wanted to…
40 years ago, Jobs imagined computers capturing the principles behind thought, not just the outputs. If we want to build that today, next generation of context-aware AI, we need attribution as the foundation. Not just models trained on scraped text, but a shared, structured…
40 years ago Steve Jobs predicted @withdelphi in an interview with Playboy magazine His vision has been on my mind for almost half that time This is a generational design opportunity, one that will transform how knowledge moves through human society
This whole thread is a perfect mirror of the problem: people think “AI vs deep reading” is about taste or vibes. It’s not. It’s about context. Everyone’s debating summaries vs novels, but not asking where the summary came from. HOW it was generated it, with what source, what was…
5 million views. twitter has really fried our brains. i was off of twitter all day yesterday. didn't realize this had gone viral until @0xgaut texted me about it. it was surreal opening it back up and noticing how fake and performative all of it seems. a lot of people used…
new positioning who dis?
llm web search but the sources aren't slop, who's working on this??
This → shrinked.ai/waitlist
I want a weekly status update on everything happening at the company, drawn from GitHub and Asana primarily but probably including other sources over time. With summaries and drilldowns. As far as I can tell, no one makes a dedicated tool that just does this? Why???
work on solving context engineering, context dam + deep retrieval+ workflows on top (shrinked.ai) landing page, dashboard designs done with @lovable and @vercel @christianreber I’d love to have that 1-1 👋
Building "cursor for context engineering" 🫡
+1 for "context engineering" over "prompt engineering". People associate prompts with short task descriptions you'd give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window…
"Prompt your life" as a service.
While some more focused on recording videos (and ditching comms ..) I built actual smart POV glasses (in a cave with a box of scraps) with @eddybuild kit and @shrinked_ai on the processing side. Smart‑glasses + Shrinked AI = “prompt your life.” POV data becomes searchable,…
totally inevitable, but scary when the underlying “doctor” is pulling from unverified, unattributed data. the key is build on structured context from real doctors. a living, reference-linked medical encyclopedia that AI can cite, not just improvise. medical RAG without…