AJ🌐³

hmmm,

[AJ🌐³ @this_is_aj__]

https://t.co/ndTf30LoTI

[InventMoney @InventMoneyApp]

Are you positioned for AI Training jobs?

Don't miss your chance to earn $10+ / hour helping train AI.

Invent Money Train AI early access is live, but spots are limited.

Sign up to claim your spot and be the first to get matched with AI Training jobs as well as earn bonus rewards.

[AJ🌐³ @this_is_aj__]

@Kiwi_Nod @pharos_network @Kiwi_Nod As a Finance student,I’ve contributed by researching and promoting Pharos($PROS),stating its robust,high-speed,compliance features,and builderfriendly tools for developing realworld finance apps like payments and RWAs on-chain.
0xba7A43Efde7DA07Dc09FE7Bda0AbA6bb6f23300c

[AJ🌐³ @this_is_aj__]

@sidrachain For over months this has been my problem

[Pepe @pepehodls]

Gn fam 💚

Pepehodls #10670

Punk Pepe with $PEPE asset #NFT is looking for a place to call home on #solana blockchain.

Reply guy with most replies in 12 hours gets it.

Good luck 🍀

[AJ🌐³ @this_is_aj__]

apparently calling things based or cringe earns bananas 🍌
claimed my @basedappcom Press Pass
help me stack these → https://t.co/7OfEtmiFwq

[Sebastian Buzdugan @sebuzdugan]

layer normalization: training a giant model is a constant battle against numbers getting too large or too small layer normalization acts like a thermostat for the math by constantly rescaling the values so they stay in a healthy range for learning [19/100]

[Sebastian Buzdugan @sebuzdugan]

residual connections: as models get deeper the signal can get lost or distorted so we add shortcuts that let the original input skip ahead and mix with the output of a layer this keeps the gradients healthy and allows us to train networks with hundreds of layers [18/100]

[AJ🌐³ @this_is_aj__]

@cryptofergani @shubh_6161 @NotTK_Alternate @justhashira @voldrix99 @babs_jide01 @psysco_ @jennydotsol @b_nx45 @0homelandersol Sol wa

UcM6PHJdJ2wu7AcV2RKYCyLWiLFu8wrpn5WsDjML3yq

Done 🙏🥹

[Sebastian Buzdugan @sebuzdugan]

multi-head attention: instead of just paying attention once the model does it many times in parallel using different heads one head might focus on grammar while another focuses on logic and another looks for specific facts allowing for a multi-layered understanding [16/100]

[Sebastian Buzdugan @sebuzdugan]

attention scores: the model calculates a score for every pair of words in a sequence to decide how much focus to put on each one it divides these scores by a constant number to keep the math stable so the training doesn't fail when the model gets deep [15/100]

[fey8(❤️,🤍) @faith_the_first]

Gm gm CT 🔥

Can I get it back ?

[AJ🌐³ @this_is_aj__]

@faith_the_first Gm gm legend.

[AJ🌐³ @this_is_aj__]

@sebuzdugan Calling it a high-speed matching game is one of the cleanest metaphors I've seen for scaled dot-product attention.

[Sebastian Buzdugan @sebuzdugan]

query key and value: in the attention mechanism every token creates three different versions of itself one to ask a question one to provide an answer and one to hold the actual information it is like a high-speed matching game where tokens find their best partners [14/100]

[AJ🌐³ @this_is_aj__]

@sebuzdugan This really shows how models understand meaning beyond just words.

[Sebastian Buzdugan @sebuzdugan]

self-attention: this is the core process where each word in a sentence looks at every other word and asks how relevant are you to me this allows the model to resolve ambiguities like whether the word bank refers to a river or a building based on the surrounding context [13/100]

[AJ🌐³ @this_is_aj__]

@sebuzdugan Seeing the whole sentence at once is a major breakthrough.

[Sebastian Buzdugan @sebuzdugan]

the transformer: this is the engine that changed everything before the transformer ai had to read one word at a time like a human but now models can see the whole sentence at once in parallel which is what allowed us to train them on the entire internet [12/100]

[AJ🌐³ @this_is_aj__]

@martinweb3g UcM6PHJdJ2wu7AcV2RKYCyLWiLFu8wrpn5WsDjML3yq

[Sebastian Buzdugan @sebuzdugan]

dataset quality: we used to just scrape everything but now we realize that garbage in garbage out is the law of the land a few thousand pages of high-quality textbooks are worth more to a model than millions of pages of low-quality social media arguments [11/100]