Construct (on a suitable product of probability spaces) two independent sequences $(X_n)_{n\in\mathbb N}$ and $(Z_n)_{n\in\mathbb N}$ of positive random variables such that $E(X_n) \to 0$ but $X_n$ does not converge a.s., $E(Z_n)=1$ and for each $\omega\in\Omega$ there is $n\in\mathbb N$ such that $Z_m(\omega)=0$ for $m\ge n$. Then $Y_n=X_nZ_n$ converges to $0$ (even point-wise) and in $\mathscr L_1$ (because of the positivity and $E(Y_n)=E(X_n)E(Z_n)$) but for the $\sigma$-algebra $\mathscr G =\sigma (X_n:n\in\mathbb N)$ you can pull out the measurable factor to obtain $E(Y_n | \mathscr G)= X_n E(Y_n|\mathscr G) =X_n E(Y_n)=X_n$.

Although this is known it is not as well-known as it should be (a colleague once showed me a book on probability theory where the a.s. convergence of the conditional expectations was claimed).

Construct (on a suitable product of probability spaces) two independent sequences $(X_n)_{n\in\mathbb N}$ and $(Z_n)_{n\in\mathbb N}$ of positive random variables such that $E(X_n) \to 0$ but $X_n$ does not converge a.s., $E(Z_n)=1$ and for each $\omega\in\Omega$ there is $n\in\mathbb N$ such that $Z_m(\omega)=0$ for $m\ge n$. Then $Y_n=X_nZ_n$ converges to $0$ (even point-wise) and in $\mathscr L_1$ (because of the positivity and $E(Y_n)=E(X_n)E(Z_n)$) but for the $\sigma$-algebra $\mathscr G =\sigma (X_n:n\in\mathbb N)$ you can pull out the measurable factor to obtain $E(Y_n | \mathscr G)= X_n E(Z_n|\mathscr G) =X_n E(Z_n)=X_n$.

Although this is known it is not as well-known as it should be (a colleague once showed me a book on probability theory where the a.s. convergence of the conditional expectations was claimed).

Construct (on a suitable product of probability spaces) two independent sequences $(X_n)_{n\in\mathbb N}$ and $(Z_n)_{n\in\mathbb N}$ of positive random variables such that $E(X_n) \to 0$ but $X_n$ does not converge a.s., $E(Z_n)=1$ and for each $\omega\in\Omega$ there is $n\in\mathbb N$ such that $Z_m(\omega)=0$ for $m\ge n$. Then $Y_n=X_nZ_n$ converges to $0$ (even point-wise) and in $\mathscr L_1$ (because of the positivity and $E(Y_n)=E(X_n)E(Z_n)$) but for the $\sigma$-algebra $\mathscr G =\sigma (X_n:n\in\mathbb N)$ you can pull out the measurable factor to obtain $E(Y_n | \mathscr G)= X_n E(Y_n|\mathscr G) =X_n E(Y_n)=X_n$.

Construct (on a suitable product of probability spaces) two independent sequences $(X_n)_{n\in\mathbb N}$ and $(Z_n)_{n\in\mathbb N}$ of positive random variables such that $E(X_n) \to 0$ but $X_n$ does not converge a.s., $E(Z_n)=1$ and for each $\omega\in\Omega$ there is $n\in\mathbb N$ such that $Z_m(\omega)=0$ for $m\ge n$. Then $Y_n=X_nZ_n$ converges to $0$ (even point-wise) and in $\mathscr L_1$ (because of the positivity and $E(Y_n)=E(X_n)E(Z_n)$) but for the $\sigma$-algebra $\mathscr G =\sigma (X_n:n\in\mathbb N)$ you can pull out the measurable factor to obtain $E(Y_n | \mathscr G)= X_n E(Y_n|\mathscr G) =X_n E(Y_n)=X_n$.

Although this is known it is not as well-known as it should be (a colleague once showed me a book on probability theory where the a.s. convergence of the conditional expectations was claimed).