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 -
https://datasci101.com/wp-content/uploads/2023/08/llm-model-size--980x533.png
Between 2018 and 2023 the best models increased in size by 500x. They are not going to reach 700 trillion tokens by 2028 which would be required for another 500x increase. The performance seems to have increased linearly while training data and model size has increased exponentially. The performance is correlated to the log of the modelsize/training data.
What we will see is specialized AIs and AIs trained on specific data for specific tasks. These AIs will be more useful than a on size fits all AI that is supposed to be used for all purposes. With more curated data and better quality control, useful AIs can be made to replace certain groups. For example a patent application AI, an AI for transcribing medical notes, a CRUD app- programming AI etc.
People have been predicting this for half a decade and it isn't what happened. Instead, we got LLMs trained on all kinds of text data that generalize across the different kinds, and now we're getting omnimodal models trained on text, images, video, etc. Why not just train your big model on all the specific data from different tasks together?
More options
Context Copy link
As someone with no technical background, the 'obvious' play seems to be hooking up models with different specialty training and abilities to each other in a way that they can talk to each other natively and without too much latency. Each model can be a different 'lobe' of an overall more intelligent brain.
Which is of course somewhat similar to how the human cognitive system works.
Maybe ChatGPT is the 'narrative' module that coordinates everything. "Oh, you're asking for solutions to a complex math problem, better send that over to the higher maths module. I'll let you know what answer it produces." or "Ah this question pertains to reading an X-ray and rendering medical opinions, better shoot that over to the model trained on millions of patient records and actual doctors giving feedback."
I dunno, really seems like they haven't picked all the low hanging fruit just yet.
It's definitely an approach that many have been taking, and it's likely to be significant for the next couple years.
A major issue with it is that many times general intelligence needs to answer questions that cut across multiple domains. You route a question to one module, and it turns out another module would be better for it; or, more fundamentally, you need a module that incorporates elements of both, and either one individually (or both combined with a module meant to synthesize them) is inferior to the ideal result. By analogy, suppose you have a problem where you want to do sophisticated analysis on millions of x-rays: although you could probably get by with team of experts (e.g. a radiologist and a statistician), the results will probably be inferior to an expert radiologist who's also an expert statistician.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link