This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
A sound is not a word? (or chunk thereof)
There also is not a 1:1 mapping of sounds to tokens in English. (or most languages really, but English is particularly bad that way)
LLMs are trained on text corpora -- how do you map those text tokens to sound data?
You seem to fundamentally misunderstand what multi modality is.
The LLMs are not trained on text corpora. They are trained on tokens. These tokens are just numbers. The training process is good at finding patterns in the numbers they are fed, and predicting subsequent numbers.
Where do the numbers come from? Well, in text only LLMs, there is a tokenizer stage that takes text input and translates it into string of numbers. These numbers don’t directly correspond to individual letters or words, and so LLMs never really see those. This is why, for example, they had trouble telling how many rs are in strawberry, because what you’re asking them is how many rs are in 2645, 675, 15717 (this is the tokenization of Strawberry). It’s actually pretty magical that they manage to figure out that there are any rs in these numbers at all, that they are able to learn detokenization.
So how about different modalities? Well, these are represented as just different tokens, different numbers. For us humans some numbers represent image fragments, some represent sound snippets, and some represent pieces of text. For a multimodal LLM though, it’s all the same, it’s all just numbers. When you train them, you feed text, images, and sounds directly into the model (after tokenizing), you don’t do anything like “first transcribe sound to text through some different model and then feed it as text”. The sounds are never mapped 1-1 to English tokens, they are mapped 1-1 to sound tokens, which is just a different set of numbers than those representing Unicode text.
Because of this, for an LLM, not only sounds and words are intercompatible, but in fact there is literally no difference between the two on a technical level. There is of course a difference on a semantic level: for example, it learns quickly that text and sound tokens never mix, because it’s never fed training data that’s intermingled sound and text. Instead, after sound comes another sound (eg response to a dialogue) or a bunch of text (a transcription). But, to a model, there is no fundamental difference between text and sound and image.
More options
Context Copy link
Just feed the text into a text-to-speech device or software, then use the output from that for the mapping.
More options
Context Copy link
More options
Context Copy link