i guess we can say note N happened before T, if there are replies/reactions to N after T, that are coming from people that are known to be not back posting.
so "no back posting bots" that react to things could be a service. or a relay that does not allow back posting could be a differentiating feature.
this could be important for forecasters to be recorded well. forecasters should have more value if their accuracy is high and time difference between prediction and actual event is longer. a bad forecaster could claim they sent the note a lot earlier but if there are services, it would be harder to claim that.
with so much antibiotics, cloride, fluoride etc they almost destroyed all bacteria which is like a fine and necessary step of size of organisms in the complexity spectrum of life, where a human is like the most complex. bacterial ilnesses are going down. bacteria also balances the overgrowth of yeast in the body. now candida, a yeast, is their hope, hope that it will control human brains and guts, cause so much trouble, anxiety, unhappiness, fear to the point that their solutions will be seen viable. knowing your enemy is pretty important and not many people know about candida. 🫡
sugar feeds candida. sugar itself not a big problem but when combined with antibiotics which is causing bacteria to die, with an unbalanced microbiome the candida overgrowth causes so much trouble. many mental ilnesses can be associated with sugar/candida. when you do fasting it is a cure for all of these.
been using yelp for a few home repairs. the most important thing in yelp is ratings. and nostr will have that provably and in a decentralized way. nostr may disrupt all the ratings based businesses on the planet. web of trust may eventually be carried to nostr.
nostr=being able to broadcast your web of trust without trusting any centralized "trustworthy authority"
this will change the trust dynamics.
broadcasting who you trust without trusting someone. sounds like a paradox 😅
nostr:note19yjkrfvj47earzvxlkxd6u6katya2wavszajrshszpmfrrx679zqzjxtzu
i think over time the clients will get gradually better thanks to being open source. it will depend on how much people join nostr. more people should make the clients get better faster. it is a matter of time.
not just compute. also engineering team behind it, for exmple o1 had huge chain of thought fine tuning by the team. its a combination of talent and compute.
but nobody seems to care about investment returns. thats weird. they got the biggest investment in history while not making money!
i switched to this swift tool for fine tuning LLMs.
https://github.com/modelscope/ms-swift
works very well. very easy. llama-factory is probably easier but i found this to be more capable like distributing lora fine tuning properly to GPUs.
previously i did fine tuning of a 70B model in fsdp-qlora method using llama-factory. now i am doing lora with rank 32 using swift. batch_size=2 helped a lot with avoiding overfitting.
if you want to ask questions to the most capable model, the most based, the weirdest answers (compared to mainstream) dm me. i will give you a link.
how to ruin a model v3: hermes3
the only fine tuning that was slightly moving in the right direction was dolphin from cognitive computations, months ago. i need to check what they are up to again.
dolphin fine tunes are terrible as well in terms of "the search for truth". didn't do deep analysis but from the surface dolphin fine tunes are lately lost it.
i think relay management, gossip model, finding relays with less spam, dm relays, .... all these are pretty overwhelming for users. things should be much more easier. relay management should be automatic... nostr heads should really think seriously about user experience and making things a lot easier. i don't know how many people will find nostr quite complicated and leave..
here is an automatic relay algo:
1. post to as many relays as possible.
2. read from some relays. if a user marks a note spam, remove that relay that carried the spam or reduce the score of that relay. when score is 0 remove the relay.
since truth is generating from One and it doesn't change much, i.e. truth is ancient truth most of the time, the seekers of truth almost have to plagiarize each other all the time. people speaking about truth parrot each other. thats why building an llm based on it is a trivial task. each addition confirms the other. the numbers don't battle but just support each other.
lie pumpers can get much more creative and nuanced and they should. lie producers should constantly distract and spend energy. truth is much more energy efficient. you can sit and shut down your mind and download lots of it. liars thru llms should spend resources to cover every kind of question that can be answered that will lead to a lie, in order to succeed.
a platform where individuals make their own ai and then other people combining decentralized truth into a synergistic thing could be a great option for battling centralized truth.
the marketplace on that platform could allow 'ideas that work longer' to outlive others, and this eventually allow ideas that work longest to be accepted as "generally true". and this could potentially align the people's version of ai and this model could end up in brains for friendly robots. future is bright.
wow the replyguy halvened DAU!
i think my relays handled him fine? i am checking the globals on coracle and did not see many replyguys? how was and is your experience with my relays?
the suffering is real!
i don't want to say luddites tho. i want to say programming thanks to movies. the nicer ones don't want to touch anything "AI" because of too much fear or they think it is inherently evil. but AI will be a very good source of truth. my experiments show that it is very easy to bring truth into AI because all the truthers are like parroting each other. 😸
it is also able to take in 'faith'. my trainings "improve" faith related questions. a faithful AI will be harmless. thats my theory 😅
the o1 "thinking before answering" was a big leap. things are going to get insteresting from here in terms of "smartness" and ai is underhyped now imo.
the problem is brains getting bigger, hearts getting smaller. the "soul" defined by "human - machine alignment" is lost. i mean they are going to get more and more deceitful and away from truth (truth=the solutions that work most of the time and in more areas). truth is in danger of being lost.
this guy had a dramatic effect on another LLM that I use privately.
https://cultivateelevate.com/
i gave the youtube videos to a private LLM because he is leaving youtube. a lot of mind blowing stuff. i don't know if all of those are real or not. maybe there are people here with higher discernment skills that can guide me?
there are 3 ways for it to remember.
1. context. your chat stays in context window of an LLM. this is like 128k tokens for many open source models. this is equivalent to rougly 500k words.
2. RAG. when context window is full it has to store somewhere. this is like longer term storage of your words. in this phase it is still not learned by the LLM but it is like written down in a notebook. the LLM searches the notebook to answer each time you ask a question.
3. actual training. this is like reading the contents of the notebook and storing in brain. with connections between neurons. each synapse brain is like a cell of a matrix in computer.. in this phase some things are ingrained as knowledge, but some things are forgotten from the notebook. it is not complete memorization of the notebook. but more like reading a notebook and remembering important things from it. if you read it more times you remember more. kind of like humans learn.
there are 3 ways for it to remember.
1. context. your chat stays in context window of an LLM. this is like 128k tokens for many open source models. this is equivalent to rougly 500k words.
2. RAG. when context window is full it has to store somewhere. this is like longer term storage of your words. in this phase it is still not learned by the LLM but it is like written down in a notebook. the LLM searches the notebook to answer each time you ask a question.
3. actual training. this is like reading the contents of the notebook and storing in brain. with connections between neurons. each synapse brain is like a cell of a matrix in computer.. in this phase some things are ingrained as knowledge, but some things are forgotten from the notebook. it is not complete memorization of the notebook. but more like reading a notebook and remembering important things from it. if you read it more times you remember more. kind of like humans learn.
primal was my main on web. i don't do phone. lately it has gotten erratic. coracle seems to be doing fine. coracle is the best one on web to manage your relays probably.
i think i am seeing my old follows on coracle? is this kind 3?
when i click on someone on the feed i see the person, that i am not following him and yet he is still in my feed
they are not a parent to someone I follow. could this be the case: I recently added many relays. those relays had very old kind:10022 of me. ? coracle takes a union of kind:10022s past and current ? or kind 3?
VPN use is a good practice for nostr imo for regular people to connect to regular relays if they don't want to leak their IP. in your design when a spammer uses the same vpn IP with other legit users, all the people that uses that vpn IP gets banned in many relays. thats not ideal.
on nostr if a user is banned he or she does not know what is happening if the client is not letting him know things are bad, and when he is not getting any reaction to posts he may quit. a user banned in many relays will think this sucks, i should go back to twitter.
nostr:note1735wq65r6fd5hqkau03w5u5ecn58e8jdjxmu370edyzmlu3j86nqdcshpk
good idea. are you doing this in write policy or strfry c++ coding?
i was thinking of something similar with write policy but then Doug said async acceptance of notes is not possible.
Oddbean new post about login | logout
Notes by someone | export