shoman3003

thanks, will do

interesting take. but its not because of elitism, but because humans are unpredictible. hiring is very hard, not at all reliable and even when you get to a star team, the leave.

not an american, not even in the western hemosphere.

nice one

enough bullshit with VC, enough allowing them to waste our time & worst of all our bend our thinking processes.

we are not here to do their job for them, we are not a lottery card for their x4 return.

enough with the shit, enough hyping them up. enough talking about fundraising as if it was the world cup. it is not.

this comment is so emotional its hard to read

best article i read on hackernews this year

craft is overrated. you clearly are a writer, which is noble, but most people aren't they just want to communicate as best as possible fast - the llm can just take their rambling and insert a summary.

i think it's a safe bet that in 10 years talking to Ai would be seen as normal, its already starting to be normalized. regarding privacy, at work it might be solvable, outside i am not sure.

bad title choice

i was building an agent to do outreach, but while i was leaning how basic memory works with letta - i noticed something; all memory architecture (letta/claw/claude) simply moves the pain to retrieval.

I started looking back at my own memories trying to understand how my mind remembers stuff as opposed to how agents do, and then i came across a pg essay that mentioned something similar.

Human memory doesn't store images in full resolution; just think back to any old memory you have - it's barely visible (100*100res). The older the memory is, the lower the resolution is. Faces are usually blurred; the images are there for context not for clarity. Same with text, when i look back at an old conversation, i don't remember most of the stuff that was said nor do i remember a "summary" of what happened. My memory of conversations is a lot like my memory of images; one part is ultra clear while the other parts are blurred (pain=weights).

Why would our minds blur stuff if they are saved to long-term memory! if my 10-year-old laptop has 500gb storage, my brain definitely has many terabits if not x100 that! Why do i have to relearn a skill that i used to be very good for years when i clearly remember almost all of the steps to how to do it. This sounds a lot like a context memory that went through compaction & then dumped to the disk. Even tho the log has been saved to long term memory; it's no longer usable unless rebuilt into a new context again, which is why i need to relearn the skill.

ONE POSSIBLE EXPLAINATION why the brain needs to care about memory saving to this a degree that it has to forget all of the details about my first car except its color & brand. It's that our biological memory is operating more like a set of context windows that switch between each other not retrieving data from a permanent long-tern storage!

Which means that we switch back and forth between different specialized context windows while we work (which is why we can switch takes in less than 60sec). Unless the context window is dumped into the long-term memory, we can still load it fast. The brain doesn't dump current sessions unless something major happened, when we change career, cities or we go through something big.

Whenever i am in a new city it feels like a new context window has loaded into my brain, my childhood memories suddenly feel different, the stuff relating to a different career i had in the old city feel alien. The new context window even makes me think in a different lens & analyze stuff from a totally different angle.

Why are we not doing the same with Ai, why do all memory structures either remove the full old sessions entirely or summarize them into oblivion? why not do what the brain is already doing & spread the work across 100s of context windows that each is concerned with 1 thinking pattern?!

why it feels like charismas when a new essay drops out

I talked to ~350 Ai founders in the past few months

~%40 are building generic Ai agents that does everything (aka, does nothing)

~%25 are building customer support agents with no differentiation whatsoever

~%20 building coding agents that claude already outshines

~%5 building an Ai CXO that doesn't really make sense in 2026

~%3 building financial agents that does reconciliations (what's new) ~%3 building agents for lawyers that writes/summaries documents (2023 says hello)

~%1 doing an Ai picture story maker for children (0 retention rate) ~%1 doing an Ai video editor that is less video more image+caption slop

~%2 doing Ai therapy/emotional tracking that doesn't really work

Not a single one is building anything new. Every single website is the same (ai agent that can save you xh/week), even the freaking UI is the same vibe coded purple/blue

Just fyi, you deserve what happened to you. I am Palestinian & what you are doing is turning our situation into a social trophy.

Maybe you guys should read about what you are supporting first.

English is not my first language; my native language has 28 letters & 6 variations of each letter. That gave my old culture more room to capture different types of thinking patterns, though they were mostly spiritual/metaphysical due to the influence of religion early on the language. That culture was too masculine for example, so they didn't really have many words for complex emotions, unlike French & German.

French & German do have a wide range of emotional language. You can literally express dozens of complex emotional states in 1 word where it would take 2 sentences to express in English. Still, the french/german words invented so far to express emotional states are fairly primitive compared to the actual emotional states we go through each day. There are still hundreds no mapped out, many have no word in any language. Imagine if English had no such word as Grit, Obsession or passion, would you really be able to consider someone speaking English emotionally intelligent?!

An Ai therapist app for example can't really do a good job when many of the emotions the patient feels do not have a word associated with them! which is why a human therapist is still kicking as due to her intuitive detection of that emotional state that needs 2 sentences to describe.

This is just 1 example. Language itself is the #1 limiting factor for how intelligent something can be (artificial or not)! What we call intelligence is the abstract ability to find new patterns in a given environment. An ai playing an alien game is unlikely to win if it were only allowed to define %50 of the objects in the game. Same with humans, if our ancestors didn't map all of the possible objects/emotions/items in the world into language, we can't ever pretend that a digital intelligence can navigate it, it literally has no access to %90 of it.

If we had a language with 50 letters for example, the 2 sentences needed to describe each emotional state (made of a dozen different individual emotions that we have a word for, and some we didn't map yet) would need only 1 word to describe them laser accurate it makes the reader feel the emotion without needing to experience it firsthand.

In a world where a 50-letter language is wildly used by agents, where the digital intelligence is literally able to remember an unlimited number of words - there wouldn't be a need to distort the truth by oversimplifying the thinking process to save memory or to consume less calories. -We can have a word for every type of American to "grand grandparent career" level, not just call someone black American or white American. -We can have a different word for every type of attraction, not call all Love. There is "you make me feel good love", "I like your apartment love", "you can be my future wife love"...e.t.c -We can have a different word for each new startup; a "$5 million ARR startup" is different from a "50M 2-year-old startup". -Each employee would have 1 word that describes their entire career right away to the HR Ai.