Good morning. Hello. How are you? #848
Man the Philipines campaign was bad. My bank is trying to buy SVB gulp. Way too much about AI. Teaching Jane scenario planning.
Good morning! Hello, there! How are you? Tuesday. Running late. Slept 15 minutes too late, even though I went to sleep an hour early. Seriously. What is going on. Why am I always tired. And it’s soo cold. I knew this deep-freeze-before-spring was coming, it didn’t fool me one bit nope nope, but that doesn’t make it any more pleasant. it is 26 degrees out there. On Friday it will be 84. Ridonkadonk, as I used to say at some point, I think maybe around ‘92, ‘93.
Still reading Howard Zinn’s People’s History of the United States and it’s still bringing me down. Got through our disastrous, genocidal campaign in the Philipines, which I knew about as a vague historical bad thing we did but my god, the scale, the violence, the atrocity, just awful. We sure have done a lot of terrible things, America. Then got through the labor strife and strikes, and WWI, and the Wobblies, boy the Wobblies were great, and all the people we kept throwing in jail — thousand upon thousands of people — for just talking. For just saying things that got people upset, like that the US should maybe not attack its own citizens or maybe not get involved in stupid evil wars. You just went to jail. I mean, shit is messed up these days but reading a history constantly reminds me that it’s always been awful, this country’s been so much worse in the past than it’s being right now. That’s not nothing.
So it seems that my beloved local bank, the favorite of all our banks, First Citizens, with a branch just down the street here in rural North Carolina by the evil Harris Teeter, the bank that is run by the grandson of the dude that took over the small bank in the 1920’s, well it seems that they are verrry interested in buying Silicon Valley Bank. This is so gloriously ridiculous I can’t even tell you. What fantastic ambition. As a man who routinely tries to buy absurdly large, somewhast tarnished companies for very little money, I can deeply appreciate this. Very into the 37th largest bank in the United States trying to buy the 16th largest bank.
However as a customer of First Citizens I cannot say this excites me too too much. I mean, I guess it would be kind of hilarious if I could go down the road to my podunk little local bank branch and, like, finance a line of credit against some shares I have in a series D startup or something that would be kind of funny. One thing that’s pretty hilarious to me is that First Citizens does not even give its normal customers the ability to wire money out. What a brilliant way to contain a modern bank run! Don’t let people wire money! You have to go into a branch. Of course, I doubt this is the case for the business customers, but still. Clever.
AI AI AI man people sure are freaking out about AI. ChatGPT 4 is going to change the world. Okay, I have questions: it seems pretty clear that large language models, such as ChatGPT 4 are flawed. The make mistakes all the time. And yet we seem, as a species, compelled to ignore this and talk about how they’re going to change the world and replace a ton of grunt labor and… I am just confused. Are we all going to just accept that they make mistakes and not do anything about it? I mean… I don’t think I’ve even seen any theories about how to fix this. All LLMs are flawed because humans are flawed.
It makes sense to me that these chatbots are better with code than other forms of communication, since code generally has rigid structure and was entirely invented by humans, devoid of interaction with the physical world. And I suppose you could argue that we made up all words and language too, but we did not make up the things they represent, and we have not perfected language. There is more to it. Language is not intelligence.
I mean maybe you could obviate this with scholarly journals or actual books instead of the garbage dump that is the internet, but even then I am dubious. I do wonder if Project Gutenberg has been scraped, wittingly or no, by any LLMs. I wonder if that would make it better. I kind of suspect not because humans made all these words and humans get smarter and kinder by reading a bunch of books, but only to a point.
But still, quality of the LLM itself aside, the thing that makes these AIs what they are is the association within the model, the hopping to-and-fro and spitting out language that is plausibly related, even though it’s not always related. The errors seem inherent in the approach! No one is denying this! No one seems to have a plan to fix it! Yet people just ignore it.
It seems to me that even if ChatGPT 4 and its ilk succeeded as wildly well as the most fevered dreams, it wouldn’t replace anyone. It would just turn us all into proofreaders correcting its mistakes. And it seems to me that it is inevitable that some of those mistakes will not be caught, because some are buried deep and just seem true. It seems inevitable to me that some hubristic PE firm is going to hand over the keys to a single fund to ChatGPT 4 or some equivalent and accidentally lose a billion dollars from it. Or worse.
Hopefully the freqency of the small mistakes takes the bloom off the rose quickly enough that it keeps us from handing over more and more responsibility to AIs that are, by design, guaranteed to be wrong with some regularity.
Hopefully.
A good thing I read about ChatGPT and AI yesterday. Been subscribed to this AI newsletter for like five years since I did that Superintelligence and Public Opinion piece. It’s a great newsletter but mostly focused on the technical side of things. But it had a nice little bit yesterday about ChatGPT being a political act:
Why GPT-4 matters - GPT-4 is political power: GPT-4 is more interesting to me as a political artifact than a technical artifact. By this I mean that GPT-4 is basically hard power politics rendered via computation; it's a vastly capable knowledge worker and data transformation engine whose weights are controlled by a single private sector actor and shared (with a bunch of controls) via an API. GPT-4 is going to have a bearing on economic life and also cause societal changes (obvious case: chatGPT has already led to irrevocable changes in how education works).
GPT-4 should be thought of more like a large-scale oil refinery operated by one of the ancient vast oil corporations at the dawn of the oil era than a typical SaaS product. And in the same way the old oil refineries eventually gave rise to significant political blowback (antitrust, the formation of the intelligence services), I expect that as the world wakes up to the true power of GPT-4 and what it represents, we'll see similar societal changes and political snapbacks.
The times, they are a changing, but history sure does love to rhyme!
There was also another good section this week about how AI represents “hard political power” that we are ceding to the private sector when it should be being undertaken by the public sector — I’ll add the caveat “if at all.”
Why this matters - twilight of democracy: The ability to train large-scale, capital intensive AI models represents political 'hard power', especially given that these models encode their own political ideologies and can become powerful forces in driving economic and organizational efficiencies. It perplexes me that governments are seemingly standing by as a small set of private sector companies are developing hard political power via increasingly powerful models.
History shows that when forces outside of government develop hard political power you either get a) messy revolutions, or b) a wild overreaction by the state to reclaim power. I am not sure why in the Western world we are rolling the dice here, but we are rolling them!
I do not completely agree with either of these — I don’t believe ChatGPT is going to overcome the fundamental flaws of LLMs, and thus I don’t think it will ever truly supercede a knowledge worker in anything other than speed, which will be vastly mitigated by the consistent insertion of potentially undetectactable flaws. But I am happy to see this newsletter thinking about these things. And I do think there is some relevance to the oil refinery metaphor.
And I absolutely agree with “whose weights are controlled by a single private sector actor” and it’s worse than that, because they’re controlled “in secret.” It really is fucked up these private companies are digesting giant chunks of humanity’s work, getting different results because of which chunks they ingest, adjusting it accordingly, and then calling all of this digestion of the work of humanity proprietary.
God bless Google Books for being so hubristic early on and spawning a lawsuit that has successfully kept most book contents off of the internet. The internet is garbage compared to books. These models would still suck but they’d be better if they used books. And then we’d all be fucked. Of course, I have zero doubt that these companies have no intention of following the guidelines of the Google Books lawsuit, but at least they have a bunch of shitty labor in front of them to slow them down. Unless of course they brute force the Kindle DRM which, again, wouldn’t put it past them.
Okay just gonna get all my AI shit out here in one issue so I don’t have to talk about it for another week or two. I think people are freaking out about ChatGPT because they weren’t freaking out enough before, not because ChatGPT is so good. But I guess the net result might be the same. If ChatGPT fools enough people into thinking that the end is near, and gets them off their ass to support regulation, well, then, maybe it will have done some good in the end.
For the love of god just ban this shit. Go have a SALT II-type conference.
One thing I do like is that AI is going to be yet another brain drain away from social media. Fingers crossed. Get all these VCs and bankers and libertarians and society-killers out of internet tech please, that’d be great. If I am going to keep working in internet advertising I want to be able to say so to someone on the plane and it’s as boring of an answer as saying I work in ball bearings or container coatings. I want it to be a non-society killing, non-exciting, boring-ass industrial industry that no one cares about but that we all need. And the people who get rich in it just get boringly rich, where they’re bigwigs in their state politics and maybe buy Porches and make slightly too big politicial donations but otherwise no one has ever heard of them. Like the Uhleins or Louis Joy or something, except the ones we’ve never heard of. That’s what I want the social ad-supported web to be, if it’s gonna exist at all: boring as fuck, devoid of hubris or showboaters. They can all go work in AI and crypto.
Jane bedtime was mostly lovely last night. We are working on scenario planning now, weirdly. Like making best case, worst case, okay case, good case scenarios for the next day. Figuring out how to plan for them, how to mitigate the worst case scenarios. She is very into it. Will it help, I don’t know, but Napier Collyns would be proud.
Daddy Jane Dance Party consisted of Pulp (liked them), the Dandy Warhols (indifferent), early Depeche Mode where she was very fascinated with how young they looked, The Stone Roses (didn’t care), Donna Summer and Giorgio Moroder (liked it), the Slits (hard pass), the Raincoats (moderately into it but it wasn’t dancy enough), the Knife (liked the video more than the song), Santigold (very into the video and appreciated its dancability) and ended with Wet Leg and Robyn. Sad old man wind down was the Afghan Whigs, which she said wasn’t really sad. She doesn’t believe Greg Dulli is really sad and, you know, fair point.
Covers playlist today. Doing a little Dulli bookend here, forgive me. Ties in with Daddy-Jane Dance Party and all. Plus I’m running behind schedule here crap it’s already 8:20. Man.
OK bye sorry I won’t talk about AI again for a while.
i’m loving Jane’s lavendar haze-esque wall color. bravo Emma!!