A description of a seven-layer cognitive architecture for AI agents — memory, beliefs, feelings, reflexes, immunity, predictions, and narrative, bound by a nightly consolidation pass. The system has been running in production for about a year; live counts from the database are in the post.
Opens with a meta moment: while drafting the section on the immune layer, one of the system's own hooks fired and blocked the author mid-paragraph — the architecture working on its writer in real time.
The post describes the shape of the architecture without publishing the internals — schemas, update math, trigger grammar, and phase pipeline are held back. It is a description of what has been built, not a research proposal. Interested to see how this composed-layers approach reads against the current memory-layer-focused work from Mem0, Letta, and others.
Hi - gamma is defined as [average time spent on market]/[average time spent off market] for someone who isn't lying. This depends on what kind of job obviously, but if it's the norm for people to spend 1 year unemployed for every 15 years employed, then gamma = 1/15. The claim is that the lower the gamma, the higher proportion of liars in the job candidate pool.
He is getting people aware about data centers and other things (Surveillance tech) coming into their towns. Companies are lying to people, not giving them the full picture, or not even including them in the conversations.
Please make a video that talks more about these things which would reach more and don't really need to throw shade.
You should talk about the other things people should worry about with these data centers instead of what he talks about.
I am wondering did you message him to talk about these issues? I feel usually when people do not want to have a discussion it is not productive and this actually can harm regular people. Maybe that would be a great discussion both of you can have and even in a podcast form to gain a bigger audience?
The excuse of impressions is just an excuse to not say they hate Elon and would rather push ideologies rather than fight for one of the civil liberties they claim to defend which is, free expression. The EFF are nothing more than fascists.
We run agents on web tasks and the difference between
clean semantic HTML and a site drowning in React state is massive.
Agents can technically navigate both, but one costs 10x more in
compute and fails 5x more often.
Whether site owners have any incentive to care is a different
question. Agent traffic doesn't pay the bills yet. Something like
x402 could change that, but the payment rails aren't there.
Good point, just pushed a fix. Title, author, subject, keywords, producer, creator, creation date, and modification date are now explicitly stripped from the output file metadata.
It seemed that these were already removed when the PDF was rasterized, but now they're explicitly being removed.
Just open sourced it: github.com/mr-guac/redactpdf
For your friend's air gapped environment, the file works offline after the libraries cache on first load, but it does pull PDF.js and pdf-lib from CDN so a one-time internet connection is needed.
To run it fully offline you'd need to download those two libraries separately, transfer them to the air gapped machine, and swap the CDN links in the HTML to point to the local files instead.
Clean-room, portable C++17 implementation of the PlanB IPv6 LPM algorithm.
Includes:
- AVX-512 SIMD path + scalar fallback
- Wait-free lookups with rebuild-and-swap dynamic FIB
- Benchmarks on synthetic data and real RIPE RIS BGP (~254K prefixes)
Interesting result: on real BGP + uniform random lookups, a plain Patricia trie can sometimes match or beat the SIMD tree due to cache locality and early exits.
Would love feedback, especially comparisons with PopTrie / CP-Trie.
curious whether metadata survives the PNG roundtrip. things like original creation dates, software used, or embedded thumbnails can still leak info even in rasterized PDFs. might be worth adding a strip step if you isnt already doing it
Or maybe we're spending too much time on communicating. If too much time is allocated then its hard to stay focused and there's always the next time that can be used to clarify. Cut all the unnecessary meetings and only allocate the minimum viable time to communicate. Then everyone will be listening.
That's why I added "across the country". I guess its a bad analogy.
I agree with the premise of the article but I just don't think going back to manual coding is the solution.
Here's my new attempt using puzzle as an analogy which I wrote yesterday:
Starting last year, I noticed coding was getting less fun. It’s like buying a puzzle set and finding out there’s an auto-complete button. Press it and the puzzle solves itself. Faster than me, better than me, prettier than me. It’s like playing a game with cheats on.
I don’t even have to touch the pieces anymore. I just tell the auto-solver what I want. Tell it I want a bird, it gives me a bird. A pirate ship? Here’s a pirate ship. At first I never imagined it could do a rocket, but with its help, that went from fantasy to reality fast.
Sometimes it doesn’t quite match what I wanted, but usually just telling it what’s wrong fixes things. The whole process is so fast that, if nothing’s broken, I don’t even bother looking at how it actually solved it. That would just waste time.
But coding felt less fun with this new assist mode.
The fun of puzzle-solving is gone. That feeling of trekking through the hard parts and finally reaching the summit is gone. Now it’s like taking a cable car up.
Before, I had to think alone for a long time, try things, experiment, until I finally cracked the problem. Now with the assist mode, it’s like doing college homework where the teacher already has the answer key. I just ask and I get a standard answer.
Coding went from craft to management. “I” went from a craftsman with standards to a foreman watching workers do the job. It’s just not the same. And “foreman” sounds kind of weak.
Coronavirus vaccines already broke that barrier. Normies will inject any shit into their bodies if the talking heads tell them to. The alternative - not completely conforming with the views and behaviours of their larger in-group - is a far far worse fate.
Exactly. When you're operating as a business you need to be executing and AI helps a lot in brainstorming, developing, testing etc.
I have ADHD and just by brainstorming with AI helps me initiate.
Of course, you need to be the ultimate gatekeeper or else there will be quality issues. But isn't that the same when we write manual code? AI is just another tool in your toolkit.
enforcement is the hard part. most context engineering stuff describes what should happen, not what actually stops it from happening. curious how your enforcement layer handles runtime checks vs just descriptive ones
Surely Islam does too? Being stuck with 72 young women for eternity as an old man with incontinence, bowel cancer, a bad back and a broken dick sounds a lot more like hell than heaven.
