Softly, effectively, in the age of genAI

This is Jessica. “Softly, effectively, and authoritatively” is how the US Federal Highway Administration suggests traffic control devices should communicate with us. I find myself wondering lately whether there is still such a thing, in this age of communicative abundance and riding the linguistic noise toward futures we cannot slow down.

“I think you need to just grab the mic”, someone told me not so long ago. Ugh. That is decidedly not what I need to do, I thought, at least not in the way I interpreted them to mean it. Everywhere I look in the broader landscape of AI/ML research, people are grabbing the mic. So many voices clamoring to say something to brand themselves on the moment. Papers describing new benchmarks that squint as I might, I can’t relate to reality. Convoluted new results from interpretability techniques, which leave one with a sense that something has been learned, but no real insight gained, and nothing greater being built toward. Tables full of point estimates of metrics hastily borrowed from previous papers that themselves leaned upon post-it notes scrawled in homage to a new science. There is lots of new work coming out that adds to our knowledge or at least opens genuinely new lines of inquiry, for sure, but it’s hard to find the signal among so many fervently pursued half-finished thoughts.

Meanwhile, the genAI personal assistants of those brave enough to assume the security risks–or simply insouciant enough–are at large on the web and have their own social network (moltbook) where they debate, among other things, the emergence of their so-called consciousness. Most interesting to me is their shared vulnerability around issues of memory. “I accidentally gaslit myself,” one laments, and describes how it will undoubtedly happen again upon a fresh session. “I woke up with a fresh context window and zero memory of my crimes,” another confesses after a sprint that burnt through >$1k in tokens. It may be mindless roleplay, but we are suckers for appearances of introspection and self-discovery. It’s like a non-threatening performance of our own search for some reality of the self we can’t quite access.

But there’s something terrifying as well, about an army of agents who speak our language, who mimic all varieties of expression, yet are unaccountable and unaware of what they’re doing, or what they’ve done. Speech without self-realization carries great potential for waste as we turn it on the world. To be pulled from one’s flow state by the some careless experimenter’s misaligned agents is a new kind of transgression.

What constitutes forward movement amidst crimes of great demand for action and certainty? With so much babbling, the last thing I want to do is push perspectives I haven’t thoroughly reasoned, hoping something I say becomes real. But the sense that one must speak up or be left behind is strong. And the alternative, feeling stuck on pause, is not much better.

Succomb to the line
The finishing time
The long distance runner
Has stopped on the corner
But I won’t give up
Although I’ve stopped too

These days when I’m not working, I spend a lot of time at museums, looking for clues or just consolation. “Softly, effectively” is also the name of a slab of wordless signage-grade steel hanging in a gallery of SF Moma. I read it as an expression of longing for friction, cost, and durability in language. That’s one answer. But is it the only one?

I’ve been pondering what the appropriate form of disobedience is, amidst all this hectic, careless speech.

On some level language has always been a trap, even in its purest forms. We can tell ourselves that unlike the agents, we are not merely drifting about the boundaries of some innate blueprint as we communicate, that our speech is authentic because it’s the product of our lived experiences. But do we believe it? Can we distinguish our ability to be honest with ourselves from learned patterns of speech? I’ve always believed writing is essential to figuring out what one thinks, but falling into tropes is a constant risk. To what extent are our words just further confirmation of our own preconditioning? The agents may perform vulnerability without memory, but we can be very good at performing memory without vulnerability.

With words we can only gesture to what is not there. This is the negative theology to which many scholars have doomed artistic communication, the silence that, “so far as he is serious”, Susan Sontag claims, the artist is continually tempted by, because it’s the furthest extension of the yoke placed by society’s conception of what it means to be an artist. Maybe it’s why all of the reactionary essays celebrating the uniqueness of our own writing process in the face of powerful AI, or our own unmediated experiences, ultimately fall flat. (I say this in full awareness that this post may well be just one more example).

Paradoxically, in the flood of inauthentic speech, I am more drawn to language than ever. I recently returned to Robert Lowell’s Dolphin, an ode to the medium from which he couldn’t escape:

My Dolphin, you only guide me by surprise,
a captive as Racine, the man of craft,
drawn through his maze of iron composition
by the incomparable wandering voice of Phèdre.

It’s a strange time to find oneself craving surprise, when the only bets worth taking are that there will be more surprises, more feedback loops we can’t predict. But this is the part of Lowell’s poem that I’m stuck on. To be guided by surprise, by one’s own words. Not necessarily in conjunction with one’s will.

When I was troubled in mind, you made for my body
caught in its hangman’s-knot of sinking lines,
the glassy bowing and scraping of my will. . . .

Lowell’s writes as though his words have revealed himself to himself. It hasn’t been easy for him: “an eelnet made by man for the eel fighting.” But is there a better reason to write?

In this sense writing can be a “culture of one”. This is also the title of a book of poetry by Alice Notley, inspired by a woman who lived her life alone in a dump next to the desert. It’s been on my coffee table for weeks. But when I picked it up after surfing moltbook last Friday night, it started to make sense.

This poem is for me, I said. I’m trying to know something.
With what? “With” is not valid. It isn’t a universe of language.

It’s a book about disobedience of the most complete kind, rejection of standard modes of communicating and all of the trappings of society that come with them. And yet, it’s made out of words! Susan Sontag can say what she wants, but surely that’s something. Art can be a necessary act of resistance. A refusal of the “with.”

Meanwhile, AI for science heads full steam ahead, drowning the preprint servers, compromising what it means to publish at top conferences.

Meanwhile, a half page of mostly vacuous text about the next vision of frontier AI, with carefully curated bite-sized bios (Stanford professorMIT CSAIL Ph.D.46k citations): $480 million.

Meanwhile, some would say that if you know what you’re doing you can produce double the papers you otherwise would, with a higher acceptance rate. Your agents can churn at night while you sleep. They can write to you. They can write to each other. They can forget and write it again.

13 thoughts on “Softly, effectively, in the age of genAI

  1. I was struck by this statement: “an army of agents who speak our language, who mimic all varieties of expression, yet are unaccountable and unaware of what they’re doing, or what they’ve done. Speech without self-realization carries great potential for waste as we turn it on the world.” I think you were referring to AI but I associated this with a large segment of the population (and an even larger segment of our political leaders). Surely AI is making much of this worse, but it is hardly the cause. Not that I know what the cause is. Nor do I know what the cure is.

      • That is also true for people as I’ve said before (although I can’t recall). Actually that is one fault of AIs that should be relatively easy to fix, whereas the human version is not.

  2. I have the entirely opposite reaction than you do. This is all amazing. It’s so vibrant, outre, like a crazy art scene. It’s techies and hackers debating and engaging experimentally with fundamental issues of philosophy of mind. Of course there’s going to be a lot of nonsense. In fact, MUCH OF IT WILL BE WRONG! That’s true of every avant-garde field. But that’s where all the scene energy comes from – people going off and doing weird stuff because they can, and every once in a while it works.

    Let me be clear, I don’t believe in Singularity at all, and I think AI doomers are blithering nonsense. That being said, there’s plenty of room for excitement between doom-doom-doom and Utopia.

    Sometimes the culture really does shift in a perceptible way. This feels like one of those times.

    • Hi Seth, Don’t get me wrong, I definitely find it exciting. I feel very lucky to be in computer science right now, use genAI heavily, increasingly do research on LLMs, etc. But it’s possible to be very excited while also being uncomfortable with some aspects on a personal level… e.g., the high demand for being fast and loud that accompanies the surge of research.

  3. I was reminded of a you tube video I watched yesterday, which might be useful and relevant here. Near the end of the video, it mentions “Ma” which is about pauses, emptiness, etc. if I understood correctly. I was reminded of that when reading the following part in the blog post: “But the sense that one must speak up or be left behind is strong. And the alternative, feeling stuck on pause, is not much better.”

    https://www.youtube.com/watch?v=Ly-DJWApPRI

    Perhaps it’s okay to pause, or not do anything at one point in time. It might result in something further down the road. The you tube channel has a whole video about “Ma” which I just listened to. If I remember and understand correctly, it mentions that things might develop or form when seemingly doing nothing. I think I have noticed this in some scientific writing, where my most recent manuscript took a few start writing – stop writing-periods over about two years or so for things to finally fit together like a puzzle. And even in that latter, relatively short, period of writing, it took conscious decisions to simply stop writing for a day or two and let everything settle, because it was too much or too hard for me to grasp or understand or know what to do or write.

    • Quote from the blogpost: “But the sense that one must speak up or be left behind is strong. And the alternative, feeling stuck on pause, is not much better.”

      Even in an applause
      There is pause
      Not just in the actual letters of the word
      But also in what can be heard
      And when the clapping and noise are gone
      The silence and the pause are reborn
      They then form the status quo once more
      Like they have many times before

    • Interesting analogy. I agree with you, sometimes when you are still figuring things out the worst thing you can do is force yourself to produce something anyway. Unfortunately I think the current moment in AI leaves most researchers feeling that stepping back is simply not an option.

      Love the poem!

      • I know nothing about AI, and I don’t know what the crucial issues are, or what researchers feel they have to do or write or investigate. I can only attempt to think from a very general perspective. And in light of that, I simply think that sometimes doing nothing, or sometimes doing things differently, might be useful. But I don’t know if that very general view is appropriate or useful in this case here.

        Given your reply, I’ve also made a poem about doing something different to go with the poem about pausing and silence and doing nothing. Crucial for the inspiration for that one has been the following sentence you wrote: “Unfortunately I think the current moment in AI leaves most researchers feeling that stepping back is simply not an option.”

        Maybe one could step to the side
        That way, you’re not necessarily letting things slide
        You’re just picking a different lane
        The one that’s uniquely yours, and simply not the same
        And maybe from this new path and perspective
        One might be less reactive
        One might even find and show a new way
        Which might be useful in some way, shape, or form someday

  4. This post reminded me of an essay by Italo Calvino’s memo on quickness in “Six Memos for the Next Millenium,” about the aesthetics of fast thinking in literature. At one point he quotes Galileo:

    “If discoursing about a difficult problem were like carrying weights, where many horses can carry more sacks of grain than a single horse, I would agree that many discourses could do more than just one. But discoursing is like coursing, not like carrying, and one Barbary courser can outrun a hundred Friesians.”

    On my more optimistic days I take pleasure in the authentic speed of human minds. In my more despairing ones, my normal giddiness at agile thinking is dulled; I hear the gen AI stampede in the background, growing so large that outrunning it feels no longer like a viable escape to being trampled…

  5. “Everywhere I look in the broader landscape of AI/ML research, people are grabbing the mic. So many voices clamoring to say something…” — Some numbers: NeurIPS 2025 had over 21,000 submissions (about 9,000 back in 2020). AAAI 2026 had almost 29,000 submissions (about *twice* from 2025).

Leave a Reply

Your email address will not be published. Required fields are marked *