“RAG performance can scale almost linearly w.r.t. log inference FLOPs”

Demonstration Based RAG (DRAG)

Method

Adding demonstrations as k in-context examples.

Prompt: documents, input query, final answer.

Parameters: number of documents, number of in context samples, number of iterations upper bound.

Iterative Demonstration Based RAG (IterDRAG)

Method

DRAG above, and then the model can generate a new sub-query. The model decides

Parameters: number of documents, number of in context samples, number of iterations upper bound.

identities allows another number to retain its identity after an operation.

What identities are applicable is group dependent. Identities are almost always object dependent.

<> NUS-HIST301 American History

The idea of identity politics is proposed, that politics became associated with sub-population of identities:

• Black Pride Movement
• Chicano Activism
• The American Indian movement
  • Termination of reservation system
  • Pan-Indian Rights
  • Alcatraz and Wounded Knee Occupations
• LGBT movement
  • Stonewall
  • GLF starts marching
• Asian American
  • Yellow Peril
  • Model minority movement
• NOW Femanism Acts
  • The Equal Rights Act almost possible, and then Phyllis Schlafly happened
• Environmental Movement
  • Silent Spring
  • Cuyahoga River on fire
  • Richard Nixon creates the EPA
  • Earth Day

to prove that something goes both ways: given \(A\Rightarrow B\), and \(A \Leftarrow B\), \(A \Leftrightarrow B\).

We really really want to prove:

\begin{equation} \text{BPP} \subseteq \text{P} \end{equation}

which will give \(\text{P} = \text{BPP}\).

How about we replace the truly random bits on the random tape \(r \in \qty {0,1}^{\text{poly}\qty(|x|)}\) with “pseudo-randomness” bits and prove that \(M\) can’t tell the difference.

Namely, a thing that is “pseudo-random” is easier to brute force over. So, we ideally can brute force over \(\text{poly}\qty(n)\) many outcomes instead of \(2^{\text{poly}\qty(n)}\) in the case of true randomness.