MiguelDev

An ex­am­i­na­tion of GPT-2′s bor­ing yet effec­tive glitch

In­ter­gen­er­a­tional Knowl­edge Trans­fer (IKT)

RLLMv10 experiment

A T-o-M test: ‘pop­corn’ or ‘choco­late’

Sparks of AGI prompts on GPT2XL and its var­i­ant, RLLMv3

Can RLLMv3′s abil­ity to defend against jailbreaks be at­tributed to datasets con­tain­ing sto­ries about Jung’s shadow in­te­gra­tion the­ory?

Re­search Log, RLLMv3 (GPT2-XL, Phi-1.5 and Fal­con-RW-1B)

GPT2XL_RLLMv3 vs. Bet­terDAN, AI Machi­avelli & Oppo Jailbreaks

Re­search Log, RLLMv2: Phi-1.5, GPT2XL and Fal­con-RW-1B as pa­per­clip maximizers

[Question] rab­bit (a new AI com­pany) and Large Ac­tion Model (LAM)

Re­in­force­ment Learn­ing us­ing Lay­ered Mor­phol­ogy (RLLM)

GPT-2 XL’s ca­pac­ity for co­her­ence and on­tol­ogy clustering

Rele­vance of ‘Harm­ful In­tel­li­gence’ Data in Train­ing Datasets (We­bText vs. Pile)

[Question] Who de­ter­mines whether an al­ign­ment pro­posal is the defini­tive al­ign­ment solu­tion?

<|end­of­text|> is a van­ish­ing text?

On Ilya Sutskever’s “A The­ory of Un­su­per­vised Learn­ing”

Ex­plor­ing the Re­spon­si­ble Path to AI Re­search in the Philippines

A fic­tional AI law laced w/​ al­ign­ment theory

Ex­plor­ing Func­tional De­ci­sion The­ory (FDT) and a mod­ified ver­sion (ModFDT)

A Mul­tidis­ci­plinary Ap­proach to Align­ment (MATA) and Archety­pal Trans­fer Learn­ing (ATL)