1,074 Photos and videos
Brittle if/then/else logic gives way to robust or even anti-fragile logic with predicates.
github.com/monetic-labs/an-t…
1
4
529
notdevin retweeted
Apr 25
I've been running the same AI agent every day since JULY 🚨
- 297 days running 24/7
- 993,115 seconds of compute time automated
- 5,020,623,362 tokens generated
- 127,743 workflows ran
- 605,292 tool executions
Comment AGNT below to steal my setup 👇
198
27
228
28,937
The india builders in ai are way ahead of the frontier shops, which actually makes a lot of sense if you think about it
1
2
73
it actually makes 0 sense to use the big boys to do your hosting anymore. I do all my devops work from chat, deal with no IaaS dashboards of infinite choices and dead ends. this saves me a ton of time and money every month so i can now manage more infra and services as a solo dev -> over 18 projects, 150 rust servers, dbs, networking, access, git, and security : less than 1hr per week
the only people tied to their IaaS is when you start relying on their software, which is also wild to do now given that nearly everything they provide software wise is already available and better from an open source offereing
1
49
You will be familiar with Persona because they just got done leaking ALL of your most sensitive personal data to the world a few months ago
Claude now requires government ID verification (via Persona) before subscription.
ChatGPT doesn't.
Gemini doesn't.
Anthropic just handed their competitors a gift.
Community note
Misleading. This is only in some specific cases, and other companies like OpenAI also do this.
x.com/markvalorian/s…
71
Architectural patterns and practices that enable building with stability and confidence, andf importantly all transformers that are useful for code know these:
- Domain Driven Design
Why you might care:
If you want to know the boundary of an important component, knowing your domains will make that a zero effort challenge
- Event Source
Why you might care:
This design gives you the ability to rewind time on the system its wired as a by product of how the pattern is used to construct state
- CRDTs
Why you might care:
This gives you a way to standardize data payloads that flow on critical paths with all the provenance you need to build SLAs
- CQRS
Why you might care:
You can optimize the systems performance of reading and writing by making them atomically separate
- Factory
Why you might care:
You get to build with standardized legos decoupled from client logic, your objects are just toys without a proper factory
24
what i like about composer 2 in cursor is that its more usefully detailed in ways that sonnet 4.1 was. There is the right amount of effort required in each prompt to make sure it stays coordinated which keeps me more in tune with the code being produced
whereas gpt 5.4 i can't figure out where it actually but i know the work is right. which then leads to forward progress and maximum opacity which isn't my favorite
1
3
118
Been calling this out since the last week of January. Turns out when you use a billion tokens every couple days you find the changes and problems in models really quickly.
Apr 8
SOMEONE ACTUALLY MEASURED HOW MUCH DUMBER CLAUDE GOT. THE ANSWER IS 67%.
the data shows Opus 4.6 is thinking 67% less than it used to.
anthropic said nothing until the numbers went public. then suddenly Boris Cherny (creator of Claude Code) shows up on the GitHub issue.
users are calling it "AI shrinkflation" (same price, less intelligence)
we already know from the leaked source code that they have an internal switch that keeps the models working to their full extent for anthropic employees.
in the last week Claude went from WOW to being a more restricted and expensive version of ChatGPT.
people are saying Anthropic is deliberately downgrading Opus to save compute for training Mythos, their next model.
1
121
FWIW your data also has a decay rate and most of it is less than desirable. You can reclaim your data, don’t forget to download your data before you delete your data, you want to own an control your root set
Delete your search history, delete your bookmarks, delete your reddit, medical records, 12 yr old tumblr, delete everything.
Every photo on the cloud, every message on every platform.
None of it is safe.
It will all become public in the next year
Local storage and compute 📈
1
2
255
So the longest wave in crypto was the last cycle where nobody did much of anything new so they renamed the same things differently 10,000 times and then pretended they were all different because semantics
Ai is already doing similar things with RAG and other harness components, which is all everything non-model is, harness components
Models are like L1s and L2s, that mental model is easier, you mostly don’t see about anything you’ve never heard of before and the more desperate the shills are the less likely the model does what they say it does.
Transaction speeds shill look like token speed shills
I bet these archetypes overlap harder than some might think because the cycles are so close and so many chance for literally the same people to rinse and repeat
1
2
89
RAG is “broken”, but the solution is not a wiki 😂
Apr 6
🚨 Andrej Karpathy thinks RAG is broken. He published the replacement 2 days ago. 5,000 stars in 48 hours.
It's called LLM Wiki.
A pattern where your AI doesn't retrieve information from scratch every time. It builds and maintains a persistent, compounding knowledge base. Automatically.
RAG re-discovers knowledge on every question. LLM Wiki compiles it once and keeps it current.
Here's the difference:
RAG: You ask a question. AI searches your documents. Finds fragments. Pieces them together. Forgets everything. Starts over next time.
LLM Wiki: You add a source. AI reads it, extracts key information, updates entity pages, revises topic summaries, flags contradictions, strengthens the synthesis. The knowledge compounds. Every source makes the wiki smarter. Permanently.
Here's how it works:
→ Drop a source into your raw collection. Article, paper, transcript, notes.
→ AI reads it, writes a summary, updates the index
→ Updates every relevant entity and concept page across the wiki
→ One source can touch 10 to 15 wiki pages simultaneously
→ Cross-references are built automatically
→ Contradictions between sources get flagged
→ Ask questions against the wiki. Good answers get filed back as new pages.
→ Your explorations compound in the knowledge base. Nothing disappears into chat history.
Here's the wildest part:
Karpathy's use case examples:
→ Personal: track goals, health, psychology. File journal entries and articles. Build a structured picture of yourself over time.
→ Research: read papers for months. Build a comprehensive wiki with an evolving thesis.
→ Reading a book: build a fan wiki as you read. Characters, themes, plot threads. All cross-referenced.
→ Business: feed it Slack threads, meeting transcripts, customer calls. The wiki stays current because the AI does the maintenance nobody wants to do.
Think of it like this: Obsidian is the IDE. The LLM is the programmer. The wiki is the codebase. You never write the wiki yourself. You source, explore, and ask questions. The AI does all the grunt work.
NotebookLM, ChatGPT file uploads, and most RAG systems re-derive knowledge on every query. This compiles it once and builds on it forever.
5,000 stars. 1,294 forks. Published by Andrej Karpathy. 2 days ago.
100% Open Source.
1
87