Profile

32B llm model is now better than human average on one of the most difficult benchmarks. i think we are starting go to the exponential growth territory. despite previously commercial models being able to do the same, commercial models are not scalable. when reasonable sized open source models beat human capabilities we are starting to see scalable exponential growth. these models can be deployed in millions. https://redlib.privadency.com/r/LocalLLaMA/comments/1p7d97m/opensource_just_beat_humans_at_arcagi_716_for_002/

#dev, why are you scrollbar racist?

try catch is the most stupid thing ever invented its only purpose is to hide the error message to make debugging harder and ais love spamming it everywhere

object detection: detect bounding box of an object(s) within its environment object: a structure that is detached from its environment by another substance where, another substance may be: - air - ground - grass - water - object etc. i think object detection can be done with fair accuracy using classical logic algorithms, given algorithm has enough data of the observation. although what happens if we have car on top of car? in this case classical algorithm based on substance detection would only see one car object. also classical algorithm is not language algorithm, so it does not have concept of a car. what classical algorithm should be able to do, is use the most abstract form of object detection, which for example can be useful on systems like collision detectors: it is enough to know some object exist, its boundaries and its movement direction etc.

i found out about the internet wide middle man outage while searching for a torrent. bittorrent doesnt work if you cant access any of the sites indexing torrents meanwhile two torrent sites unaffected dtan.xyz rutor.info

BREAKING: 99 % of internet down due to cloudflare malfunction

proof of work is in a sense often misleading. bitcoin is not a proof of work coin in its definition - its a proof of hardware coin. invention of asics turn many proof of work coins to proof of hardware coins. proof of hardware coin is distinct from proof of work coin, because most of the work is done by the hardware manufacturer. its practically impossible to do the required work without purchase of specific hardware. this post is not to say which one is better. proof of hardware (poh) coins have specific benefit over pow: limited availability, slowlesness of movement in hardware. pow coins can be attacked by renting extreme amount of free computing power. asic markets probably dont have enough rentable hardware available to attack the network. yet we can arque, whether any attack on pow coins through rented hash power is economically feasible. my blind guess is: no. attacker is always playing against the house and statistically loose money.

key technical indicators for success of cryptocurrency - is it usable: confirmation in under 5 second for real world in store payments - is it decentralized: proof of work is the key innovation in cryptocurrencies that enables mathemathical double spend protection. all proof of stake systems are centralized and not mathemathically strong. pos is a social trust model. ignore the noise: - smart contracts do not have real world use cases. smart contracts are simply financial computer programs, which for some unknown reason run on chain. they enable yield systems that do no produce anything, but only move abstract monetary units based on logical conditions. real world use of cryptocurrencies is to replace existing payment systems: visa and swift.

i reviewed top 10 cryptocurrencies by market cap only one of them i usable for comparison i used only real cryptocurrencies the list is available here https://www.coingecko.com/en/categories/proof-of-work-pow #kaspa #cryptocurrency

if cryptocurrency blocktime is over 5 seconds, its unusable and will not gain adoption. nobody is going to wait over 5 seconds at the payment terminal.

bootstrapping is also same for both: if you dont know anyone from the network, you must make a blind guess. in ethereum and other "decentralized" proof of stake networks this means that there is no computational backing prooving which one is the right network. its a blind guess, but probably made for you by the code you downloaded, which probably connects to something like checkpoint.ethereum.org to download initial trust checkpoint. nostr:nevent1qvzqqqqqqypzqaw4wrnxtc64zmn58k5k5673kwna9ge3d2yfurugar7zy6gy9r0eqqspdtpdl8hu9x4ut4zrmjs3cknkdn0q5f09405jckrcfxmz03yd7fgkugvdv

most of #nostr is #ethereum. its a #pos model, which is a trust model equal to #wot model. both are social trust models, rather than computational trust models. social trust model is a model where network peers must trust existing peers to provide valid network state, whereas computational trust model is where only computation needs to be trusted. latter is called #pow model.