Thanks for the hat tip- fittingly a nicety for which LLMs are not architecturally suited (and in the case of buffing off different kinds of software licenses and spitting out plausibly deniable plagiarism for students, are essentially the business model).
I wonder how many details in that book review are simply fabricated- it's like to be at least a quarter and potentially quite more- and trending towards the lower end of that list inherently hinges on a pretty similar document existing in its training corpus, which I suspect our credulous erstwhile book review didn't expend any effort to find.
I think your point about what good this piece of writing would actually be- who would actually publish it, who would find its output useful, whose labor could it actually replace or augment- is really the core economic problem of generative AI that they desperately need the hype machine to drown out. None of these companies are profitable. They continue to lose money per generation even with their most eye-watering subscription services. Each milestone performance (which time and time again are shown to be far more the product of leaked answers than some kind of synthetic intellect) is sitting off to one side buried under the useless legal briefs without actual citations. Skills they perform well are skills for which there already exists an eye-watering volume of successful material that doesn't infer an GPU-melting cost. Meme generation is not a trillion dollar industry. Puffed up student papers that adjuncts are too exhausted to troubleshoot for their obvious faults are not a trillion dollar industry. Spam and content farming might be a trillion-dollar industry- but it's also the focus of billions of dollars of effort to circumvent and avoid. This is almost by definition a book review the sort of people who care about book reviews will *actively avoid reading.*
Is there a world in which LLMs write books that are then reviewed by LLMs and summarized by LLMs by people who don’t actually want to read? It seems bleak.
Trying to remember if there were any successful strategies for quieting down the Characters on message boards and can’t really remember any (beyond the timeout followed by a permaban followed by Character creating new accounts and the cycle repeating itself). «Lol» was always more of a parry and attempt at getting other regulars to join the fight on the mods behalf but rarely worked to divert an inspired Character.
Yeah I think the only real long-term strategy is to just intently ignore until they either give and leave or settle in and become an affectionately regarded but mostly glossed over poster. Too bad that doesn't really work in politics!!
I imagine ChatGPT would be somewhat less successful at summarising a new book that hasn't been widely analysed. OTOH I reckon it could probably muscle out the lower end of pop history
Well, it'd be less successful at summarizing it accurately, but it could definitely produce 800 plausible and graceful words. One thing about F.O.B. voice is that the main goal is "filling space on a page" rather than any of the other things we prize writing for!
Slight disagreement with your point about accuracy. These chatbots are exceptionally good at summarizing text. It's probably what they are best at. I'm dumbfounded every time I ask them to do it.
If you uploaded a new book or article and requested a summary, I'd bet your reaction to the output would be, "It's amazing a computer can do that."
Interesting that trump "won" a forum native debate. It's fascinating how he's such a good poster, considering "forumisms" are more of a Gen x debate/discourse style, and Trump is a geriatric. Has anyone ever done a write up of how "don't feed the trolls" and other unwritten rules of flame war decorum came to be? Is it all downstream of gawker, snarky BBS boards? Interesting how small cohorts of antisocial, anonymous, disagreeable tech nerds from 30 years ago unintentionally of seeded public discourse style for this century.
Is it really that tragic? You don't like being entertained? There's a good chance we'll eventually get to experience a "normal" politician in the form of JD Vance or one of those abundance people anyway and we'll see how THAT goes.
Such a great piece. I love how you actually expose the real problem with book reviews and magazines and newspapers and maybe all digital media: It's not that Chat-GPT is so good; it's that the quality in digital media is so low now, unlike what it was 30 years ago. Thanks for this!
Thanks for the hat tip- fittingly a nicety for which LLMs are not architecturally suited (and in the case of buffing off different kinds of software licenses and spitting out plausibly deniable plagiarism for students, are essentially the business model).
I wonder how many details in that book review are simply fabricated- it's like to be at least a quarter and potentially quite more- and trending towards the lower end of that list inherently hinges on a pretty similar document existing in its training corpus, which I suspect our credulous erstwhile book review didn't expend any effort to find.
I think your point about what good this piece of writing would actually be- who would actually publish it, who would find its output useful, whose labor could it actually replace or augment- is really the core economic problem of generative AI that they desperately need the hype machine to drown out. None of these companies are profitable. They continue to lose money per generation even with their most eye-watering subscription services. Each milestone performance (which time and time again are shown to be far more the product of leaked answers than some kind of synthetic intellect) is sitting off to one side buried under the useless legal briefs without actual citations. Skills they perform well are skills for which there already exists an eye-watering volume of successful material that doesn't infer an GPU-melting cost. Meme generation is not a trillion dollar industry. Puffed up student papers that adjuncts are too exhausted to troubleshoot for their obvious faults are not a trillion dollar industry. Spam and content farming might be a trillion-dollar industry- but it's also the focus of billions of dollars of effort to circumvent and avoid. This is almost by definition a book review the sort of people who care about book reviews will *actively avoid reading.*
Is there a world in which LLMs write books that are then reviewed by LLMs and summarized by LLMs by people who don’t actually want to read? It seems bleak.
Trying to remember if there were any successful strategies for quieting down the Characters on message boards and can’t really remember any (beyond the timeout followed by a permaban followed by Character creating new accounts and the cycle repeating itself). «Lol» was always more of a parry and attempt at getting other regulars to join the fight on the mods behalf but rarely worked to divert an inspired Character.
Yeah I think the only real long-term strategy is to just intently ignore until they either give and leave or settle in and become an affectionately regarded but mostly glossed over poster. Too bad that doesn't really work in politics!!
I imagine ChatGPT would be somewhat less successful at summarising a new book that hasn't been widely analysed. OTOH I reckon it could probably muscle out the lower end of pop history
Well, it'd be less successful at summarizing it accurately, but it could definitely produce 800 plausible and graceful words. One thing about F.O.B. voice is that the main goal is "filling space on a page" rather than any of the other things we prize writing for!
Slight disagreement with your point about accuracy. These chatbots are exceptionally good at summarizing text. It's probably what they are best at. I'm dumbfounded every time I ask them to do it.
If you uploaded a new book or article and requested a summary, I'd bet your reaction to the output would be, "It's amazing a computer can do that."
Only one way to find out...
Your description of forum culture is so perfectly spot-on as to become poetic. Bravo!
"fewer than half a dozen genuinely good general-interest magazines still publishing in the U.S."
Which ones? I'm just curious
High Times
Cracked
Guns & Ammo
2600
Maggot Brain
Teen Vogue
Also extremely curious for this list
forget lowtax, elon's lowmaxxing
Lol
Looks like the Trump-Epstein stuff will continue to be studiously ignored. Oh, but everyone already knew about that, right?
"amazing but not very good" is such a perfect description of LLM output, imo!
Interesting that trump "won" a forum native debate. It's fascinating how he's such a good poster, considering "forumisms" are more of a Gen x debate/discourse style, and Trump is a geriatric. Has anyone ever done a write up of how "don't feed the trolls" and other unwritten rules of flame war decorum came to be? Is it all downstream of gawker, snarky BBS boards? Interesting how small cohorts of antisocial, anonymous, disagreeable tech nerds from 30 years ago unintentionally of seeded public discourse style for this century.
Is it really that tragic? You don't like being entertained? There's a good chance we'll eventually get to experience a "normal" politician in the form of JD Vance or one of those abundance people anyway and we'll see how THAT goes.
Don't AI reviews just read like supercharged MadLibs?
Such a great piece. I love how you actually expose the real problem with book reviews and magazines and newspapers and maybe all digital media: It's not that Chat-GPT is so good; it's that the quality in digital media is so low now, unlike what it was 30 years ago. Thanks for this!
Is the second part of this post a critique of AI’s prowess or more a eulogy for a declining industry that once thrived? Or both?
Loved this perspective on Trump-Elon.