Just made a new zotero collection called Actually Interesting AI Work 😂
4
Discussion
Loading...
@grimalkina "actually useful", "actually ethical", ...
@grimalkina This is the thing I worry about — that the bullshit is gonna kill a field that has a stack of useful and interesting stuff in.
The previous AI Winters were un-fun for folk in the field… but none of them came even _vaguely_ close to the damage the upcoming bubble burst is gonna do. Gonna be interested to see what new labels folk will use and how departments will rename to avoid the association.
1
@adrianh hmm, an interesting take that I'm not sure I personally hold! Plenty of fields have mightily transformed methodologically (including my own!) without losing their name or earlier insights. I don't think assistance in coding is going away anytime soon
2
@grimalkina Fair! I'm often wrong — and I wasn't trying to say stuff that works will vanish :)
Expert systems didn’t go away — but you don't hear that label much now.
and… I remember folk writing grant proposals when I was in the AI space back in the 80s/90s who… tweaked… to avoid the reputational damage from the overselling of expert systems, 5th Gen, AI, etc.. Folk slid into adjacent fields, used different words, etc. And this round the overselling feels significantly worse to me ¯\_(ツ)_/¯
1
@grimalkina But either way — I'm happy-happy-joy on the "Actually Interesting" list :-)
1
@adrianh yeah, also fair in the context with which you meant it! I don't think very highly of the machine learning literature in general, since my reading is very much in the behavioral science side, perhaps we have a bit of that protection of always being concerned with human impacts from technology which remain somewhat constant even as there is tons of churn around technology!
@grimalkina @adrianh autocomplete vs. Grand Theft Autocorrect are entirely separate things. Similarly, the stochastic parrot has as much in common with actual machine learning and real AI research as... I'm struggling to find an appropriately vast simile here. Matter and anti-matter perhaps?
And neither these bubble companies nor their "products" will survive. Not a one. The legal liabilities alone would scare anyone rational off their claimed "IP."
1
@grimalkina @adrianh people are not going to line up to buy the carcass of Anthropic where the sales pitch is "here's $10B+ of bills, over $1.5B of judgments, and 50+ ongoing lawsuits over suicides, IP infringement, and more."
Nor is any rational company going to jump at quite literally setting $10B+ in cash on fire every six months just to catch the increasingly wrong parrot up to last year's events.
1
1
@grimalkina @adrianh except fundamentally, that is the discussion. It is not a one-sided rant, it is simply the facts of things. You cannot divorce the method from the money furnace. It's completely unsustainable.
To put it in better perspective: in 2024, Anthropic spent over $6B for $600M in revenue. And "training" costs are *logarithmic*. (2x YoY or greater.)
Imagine building a new LHC for every physics experiment - including repeats - which doubles in cost every run.
1
1 more replies (not shown)
@M I understand that your goal here is simply to perform contempt, but that's not my goal. in fact, there is a lot of interesting work coming out right now
