You are viewing a single comment's thread. View all
1
alele-opathic on scored.co
1 month ago1 point(+0/-0/+1Score on mirror)1 child
> I disagree with your assumption that it will be better in such a short time.
It's tough to predict the pace of outcomes, especially given how everyone who's tried in the last 3 years has seriously underestimated the actual real timeline of LLM or vision model development. That said, I've personally studied the development of LLMs back when they were just government projects, including such 'mostly hidden' secrets such as:
- Twitter, which was purely infinite facebook statuses, was created as a playground for very early LLMs, called 'Influence Bots' in DARPA papers of the time
- These bots were proven to be operating en-masse on Reddit as early as 2017
- Every time the civilian-side development of language models stagnated, google released some 'breakthrough' paper/model that relaunched civilian interest and development - Word2Vec ('king' - 'man' + 'woman' = ? analogies, etc) in 2013, the infamous 'Attention is all you need' paper (which kicked off this current wave), the Gemma3 models from earlier this year (which taught us that our current models were extremely inefficient - these models were so much better than they had any right to be and now researchers scrambled to figure out 'why').
To gloss over what will be a long story, **the reason why everything is proceeding so quickly is because it isn't the pace of 'research', but rather the pace of 'leaking' of already discovered information that is driving this AI wave.**
> Everything is going to get worse from AI due to this
I'd argue it already was. I could make a (relatively weak, evidence-wise) argument that, for a long time, pop music has been made by machine models. I estimate it's been this way for over a decade now.
It's tough to predict the pace of outcomes, especially given how everyone who's tried in the last 3 years has seriously underestimated the actual real timeline of LLM or vision model development. That said, I've personally studied the development of LLMs back when they were just government projects, including such 'mostly hidden' secrets such as:
- Twitter, which was purely infinite facebook statuses, was created as a playground for very early LLMs, called 'Influence Bots' in DARPA papers of the time
- These bots were proven to be operating en-masse on Reddit as early as 2017
- Every time the civilian-side development of language models stagnated, google released some 'breakthrough' paper/model that relaunched civilian interest and development - Word2Vec ('king' - 'man' + 'woman' = ? analogies, etc) in 2013, the infamous 'Attention is all you need' paper (which kicked off this current wave), the Gemma3 models from earlier this year (which taught us that our current models were extremely inefficient - these models were so much better than they had any right to be and now researchers scrambled to figure out 'why').
To gloss over what will be a long story, **the reason why everything is proceeding so quickly is because it isn't the pace of 'research', but rather the pace of 'leaking' of already discovered information that is driving this AI wave.**
> Everything is going to get worse from AI due to this
I'd argue it already was. I could make a (relatively weak, evidence-wise) argument that, for a long time, pop music has been made by machine models. I estimate it's been this way for over a decade now.