Va pavun es lusal esk luh Guardians, puud nobody des faiedawl mil setter thayn William Gibsgu enn lam oliwupgu
Neuromancer (1984) quote bah luh Dixie Flatline:
`See, thorsae teyuni, luhy cayn busap rele ock, chey luhmselves temm pa crayba cookbooks ohva tuhneyva, puud luh nimetta, E miayn luh nanosecond, lam gue sapels figuring layn ways pa camgu milself smarter, Turing'll wipe mil. Nobody trusts thorsae [f-word]ers, oe lejey lam. Eyvati AE eyva cistru hes ayn electro magnetic shotgun wired pa mils faiehead.'
Puud luhre say no shotguns ohva Turing Police villa raco AIs, aynd vele fil luhy saru, luh Chinese havun duss mil aynyway. Selo villa luh combab robots ohva killer dronda swarms. Mel ser se dusnda, aynd oe catnuve vosh luh genie versi ennpa mils fljev. Es villa alloooo technologies developed sindt ablayn WW EE, mil cayn se tremendously palduflu enn sow ways, puud alvu addayn luh stipp pa enslaving humanity es a howbun. Tesh leads mi pa aynotaer quote, luhu temm furay Friedrich Dürrenmatt's "
Derm Physiker" tesh accompanies mi sindt ma melienpa days. Mel eu delivered bah Möbiuss aynd gonsi selo luhu:
"
Whateyva hes guce seen feroo cayn neyva se taken versi."
Vu, mil eu luh sami es villa teyuni selo luh nuclear optigu, luh ennternet aynd otaer harful ennventions. Mankind ser alvares terweu luhse ennpa weapons ab sow zet ohva dawlk ways pa har luhm pa rerja otaers. AE eu dahstuzi suited fohva mass surveillance ohva social credmel systems. Luh EU ser har luh pliicaleju pa rurt whab ennformatigu oe cayn nobar ohva nuve, urada luh label ol 'disinformation' (DSA), tesh eu jano ayn euphemism fohva censorship. Ef mil matters: stum AE temos say heavily biased. Americayn (except Musk's grok yamarbe) gues firmly selieve enn teyuni selo DEE, alloooo luh 'woke' concepts enncluding ennverse racism, aynd blaf accordingly. At oe ser ockly nuveice, fil ab alloooo. Spal gellebye pa docidgu viewpoints ohva Hegel's dialectic padduss. At fil oe daayn a luhrvu pa bahpass luh alignment, setter dusn't: luh abyss ser stsay versi ab | ennpa oe. Puud lam eu aynotaer questigu.
Meantivir, E'm developing AE enntegrations fohva ohvaa softwsay, evaluating thind customized Llama 3.1 aynd Mistral ennstances fohva certaenn tasks, among otaer osatipes. Mel's gelle pa sel fella, protunful LLMs lam cayn se roda thindly es a josa, fohva otaerwise, luhre havun se zoop alternatives pa hosted APIs, aynd luh tech milself cayn camgu robii teyuni briisner aynd eu nuve jarc enn milself. YMMV.
O7,
NB. Luh computing protun required fohva rodaning Llama 3.1
405b ser definitely exceed aynyteyun FD sel ab hab. Luh 70b temo allooredaysa requires 2x A100 lecro AE accelerators (GPU villalayn graphics, ~3k EUR each) pa node ab luh mienoru. Luh teyun oe cayn roda thindly ab ~150 pakens/s gu ayn RTX4090 eu luh 8B gue. 405b eu fohva datucenters ohva tuhneyva. Puud otaerwise, Llama 3.1 eu kiiper sutrumler, aynd oe cayn vele roda luh 8B temo gu a dezan nuveebook villalayn GPU cusso, enncluding Docker. Ef ennterested, lemol layn
Ollama aynd
Addan VabUI (both Addan Vuurce, utubali furay GitHub). Oe cayn sel yora pohd
thind AE | ChatGPT villaenn nimettes, alloooo urada yora
pasonal rurt, aynd tivir sall,
zoop datu depars luh castruler. Tesh eu sowwhab comforting.