| User | Post Title | Wikitag | Pow | When | Vote |
The set of pretraining data a base model was pretrained on affects how easy it it to align. Data about AI is particulalry influential. If the pretraining data is poisoned with examples AI taking over the world, paperclip maximizing or alignment faking, then as a model is being trained to become and AI assistant or agent it may be more likely to adopt those personas and behaviors. It also doesn't need to invent these strategies for itself, it can find them in its world model. On the other hand, if the base model already has a deep, nuanced and realistic understanding from its training set of how a helpful, honest, and harmless AI assistant would act across a wide range ot challenging situations, eliciting that persona and set of behaviors from it becomes easier. Thoroughly filtering the pretraining data without impacting capabilities is challenging, but supplementing it merely requires figuring our what aligned behavior looks like, and then writing or synthesizing enough additional training data of suitable quality. For fiction, this is also known as Aligned AI Role-Model Fiction.
This approach is also sometimes called Safety Pretraining or Pretraining Language Models with Human Preferences.
Alignment Pretraining is an example of a Self Fulfilling Prophecy.
Thanks, edited that section with a link.
we generally recommend using this Google Docs add-on to convert your Google Docs drafts (formatted as you like) into Markdown syntax.
btw google docs supports this natively now! Select some text, press Option+/ and search for "Copy as Markdown".
"The X-Men comics use terms like “evolution,” “mutation,” and “genetic code,” purely to place themselves in what they conceive to be the literary genre of science. The part that scares me is wondering how many people, especially in the media, understand science only as a literary genre." -- from Eliezer's post Science as Attire.
"The X-Men comics use terms like “evolution,” “mutation,” and “genetic code,” purely to place themselves in what they conceive to be the literary genre of science. The part that scares me is wondering how many people, especially in the media, understand science only as a literary genre." -- from Eliezer's post Science as Attire.
The set of pretraining data a base model was pretrained on affects how easy it it to align. Data about AI is particulalry influential. If the pretraining data is poisoned with examples AI taking over the world, paperclip maximizing or alignment faking, then as a model is being trained to become and AI assistant or agent it may be more likely to adopt those personas and behaviors. It also doesn't need to invent these strategies for itself, it can find them in its world model. On the other hand, if the base model already has a deep, nuanced and realistic understanding from its training set of how a helpful, honest, and harmless AI assistant would act across a wide range ot challenging situations, eliciting that persona and set of behaviors from it becomes easier. Thoroughly filtering the pretraining data without impacting capabilities is challenging, but supplementing it merely requires figuring our what aligned behavior looks like, and then writing or synthesizing enough additional training data of suitable quality. For fiction, this is also known as Aligned AI Role-Model Fiction.
This approach is also sometimes called Safety Pretraining or Pretraining Language Models with Human Preferences.
Alignment Pretraining is an example of a Self Fulfilling Prophecy.
What does this fractional portion mean? Roughly speaking, logarithms measure not only how long a number is, but also how much that number is really using its digits. 12 and 9997 are both two-digit numbers, but intuitively, 12 is "barely" two digits long, whereas 97 is "nearly" three digits. Logarithms formalize this intuition, and tell us that 12 is really only using about 1.08 digits, while 97 is using about 1.99.
Question: log10(100)=2, but clearly, 100 is 3 digits long. In fact, logb(bk)=k for any integers b and k, but k+1 digits are required to represent bk in base b (as(as a one followed by k zeroes). Why is the logarithm making these off-by-one errors?
You may be wondering, are two 5-digits really worth more than one 10-digit? Sure, you can place them in 25 different configurations, but how do you encode "9" when none of the digits have a "9" symbol written on them? If so, see The symbols don't matter.
We also offer a Markdown editor. If you want to use Markdown, here is a guide that discusses Markdown syntax, and we generally recommend usingyou can also this Google Docs add-on to convert your Google Docs drafts (formatted as you like) into Markdown syntax.Markdown.
"Noticing confusion" means noticing the "tiny note of discord" (the quiet sense of "something is off") that our minds produce when what we're seeing differs from what we would have expected to see.
There is much rationality skill to be found in learning to raise this subtle signal to full consciousness, and to pay attention to it.
Sequences on Shard Theory:
Related tags and wikis: Disagreement, Modesty, Modesty argument, Aumann agreement, The Aumann Game, ⟨M,N,ε,δ⟩-agreement
Are we sure that the gene doesn't just cause people who have it to believe in Functional Decision Theory?