People are also being trained/acculturated to LLM speak, so even if they are trained on human output, they could be getting reinforcement for their LLM-tinged crap-speak.
I appreciate this line of questioning. It's really interesting to see how many excuses people need to reach for to avoid the conclusion that the "secret unheard of power" is B.S...
> We are going from the era of manual, line-by-line mental model transcription to one where software engineers can focus on data structures, software architecture and algorithms.
This is what declarative programming gives us, not what LLM-based generation offers.
> I love being able to quickly bring out the program that is already running in my head without having to worry about the grind of typing it into a format that the compiler understands.
Using natural instead of a formal language to get probabilistic results based on token fields is not bypassing arbitrary constraints of the compiler, it is dereliction of the responsibility to know and articulate precisely what you are specifying.
During the crypto bubble, it was common for "crypto insiders" to talk about how it was poised to transform the entire economy, on the cusp of making all finance distributed, about to revolutionize ownership, etc., etc. .
The people in the most inflated parts of bubbles don't tend to have the most clear eyed assessment of the real impact and potential of the dynamics contributing to the bubble: their perception is warped by the bubble, and they cannot help but see everything filtered thru it.
> Such intensive use of AI doesn't come cheap. In a post on X, LisanBench, an LLM benchmark evaluator, estimated that the output tokens alone would cost about $6.5 million at OpenAI's average consumer price. Including the far larger volume of input tokens, the post estimated the total could reach $10 million to $40 million.
This could fund 10 top income mathematicians for 8 years (based on https://careers.usnews.com/best-jobs/mathematician/salary ). Imagine what kinds of results we'd have to transform the foundations of science if we were giving brilliant minds this kind of funding to do nothing but research for most of decade....
Instead, we get slop proofs that are technically correct as PR stunts to enable corrupt kleptocrats, and most likely will drive research into culs-de-sac.
Reading a book is a form of communication between the writers the readers, and literary culture enriches this with the ongoing interactions around texts. The argument voiced by Socrates in the Phaedrus is not primarily that it removes or stifles interaction but that it undermines memory (and thinking) and that it will "create forgetfulness in the learners' souls, because they will not use their memories; they will trust to the external written characters and not remember of themselves". AFAIK, there is well established research that shows that the move from oral to literate cultures does indeed bear this out. More recent studies about how wayfinding deteriorates when we rely on GPS seems a like a similar dynamic.
There is another charge against writing in the Phaedrus, which is that texts cannot answer for themselves (contrary to spoken interlocutors) or adjust their communication to the needs and character of the reader:
> when they have been once written down they are tumbled about anywhere among those who may or may not understand them, and know not to whom they should reply, to whom not: and, if they are maltreated or abused, they have no parent to protect them; and they cannot protect or defend themselves
An interesting thing about computation is that it is a writing and re-writing that can indeed be responsive and, in a non-trivial sense, "answer for itself". LLMs do this in a form that is obviously very successful for engagement (and for hype) and for producing a lot of compelling output.
I think the Phaedrus also has something to say here:
> Soc. Then [they who who knows the just and good and honourable] will not seriously incline to "write" his thoughts "in water" with pen and ink, sowing words which can neither speak for themselves nor teach the truth adequately to others?
>
> Phaedr. No, that is not likely.
>
> Soc. No, that is not likely--in the garden of letters he will sow and plant, but only for the sake of recreation and amusement; he will write them down as memorials to be treasured against the forgetfulness of old age, by himself, or by any other old man who is treading the same path. He will rejoice in beholding their tender growth; and while others are refreshing their souls with banqueting and the like, this will be the pastime in which his days are spent.
Interesting, so based on this, do you think that Socrates would have been more ok with learning from an AI that can speak back to defend its claims (assuming we address the sycophantic tendencies) than with learning from "frozen" books?
I wonder! I do think that that the systematic rewriting realized by computation complicates the nature of writing in ways that radically deepen the problematic. But in fact, the development of literary culture more generally already deepened the problem, because, as we know now, "frozen" books come alive in the proliferation of secondary literature, literary interpretation, derivative works, translations, collage, etc. In my view, none of this solves the dangers of writing Plato recounted, but they make it more complicated and are obviously essential to the way our thinking and world civilization has developed.
Regarding LLMs specifically, my view is that the current state of LLM-based AI is leading us more towards '"write" our thoughts "in water"', because
* what we write to LLMs does not catalyze a transformation in understanding of them as systems, nor drive them to evolve in a sustainable way towards truth,
* and their outputs are the result of potent patterns in the turbulence of humanity's discourse, rather than an understanding or systematization that could "teach the truth adequately to others"
While I think their potency cannot be questioned, their sustainability and correctness certainly can be. IMO, AI in the vein of theorem provers and semantic webs are much more in line with a text that can actually speak back with integrity and genuine responsiveness, "teaching truth adequately". But I also guess
* the LLM profusion will likely catalyze some important developments and advances in those other approaches (and some use of LLMs will persist even after advances on a difference basis take over) and
* probabilistic methods are here to stay and will likely be more integrated into symbolic systems and
reply