Sam Altman wrote: “I am a stochastic parrot”. Bateson comments (in 1979): “The mountain, personified in our speech, will not become a person, will not learn a more personal way of being. But the human being, depersonified in his own talk, and thought, may indeed learn more thingish habits of action.”
The machine will be able to recreate all of this! But the surface of the whole earth will be too small to accommodate this machine — this machine whose dimensions and weight will continually increase as it attempts to reproduce the peculiarities of mind and soul of an average, inconspicuous human being.
Fascism annihilated tens of millions of people.
When you’re enamoured of a technology and someone points out important ethical challenges, a typical reflex is to seek permission: yeah, but what about this use? If you find yourself seeking permission, one useful thing to do is to step back and inspect the underlying value conflict. What does your own moral compass say? Centering values leads to mindful choices. As you move from a tech-first to a values-first perspective, the question shifts from “won’t you give me permission?” to “how do I do the best science possible?”. And that, to me, is a question worth asking.
Some thoughts on generative AI, reproducibility, and why the praxis of slow and reproducible science provides a useful lesson for navigating the lures of “AI”.
Succinct argument for scientific publishers to adopt a clear and unequivocal policy that discourages authors from submitting synthetically generated text, and that invites them to uphold basic standards and core values of research integrity.
For Kurzweil, singularity is the point at which machine intelligence would be more powerful than human intelligence. I’m coming to think it may be near just by the sheer amount of mindnumbing bullshit we are forced to contend with, which is making us all more stupid.
Start your blog with an exultant tone, pompous words, and gratuitous alliterations and I know we’re in for a rapid descent into the wastelands of utter mediocrity. I recently came across some obvious LLM-generated slop on science blogging aggregator Rogue Scholar. Here I write up why synthetic text has no place in scholarly blogging.
GPT based text generators like ChatGPT or Microsoft Copilot have rapidly become a “cultural sensation”. Here I provide scientific background and guidance on how to think critically and mindfully about these tools in academic writing and research.
I saw a thing fly by on PsyArxiv and could not ignore it so I’m doing a drive-by peer-review. It claims that English-based AI-generated norms are “of particular value for under-resourced languages”. Is that pesky linguistic diversity bothering you? Here, try on these rosy English-tinted glasses and everything will look all prim and proper, promise. Warning: snark detected.
Writing is thinking. The writing process is the most neglected part of our job. We spend millions on fancy equipment and uncountable hours on training for using this or that toolkit. Yet we assume the BA-level academic writing course we once followed is sufficient; the rest we’ll just learn on the job and hopefully soon we’ll automate away with LLMs. It is all formulaic anyway. To think this way is to hollow out the very foundations of scholarly work. Can’t think original thoughts if you don’t find your own voice.
With the first excitement of ChatGPT dying down, people are catching up on the risk of relying on closed and proprietary models that may stop being supported overnight or may change in undocumented ways. Good news: we’ve tracked developments in this field and there are now over 20 alternatives with varying degrees of openness, most of them more transparent than ChatGPT.
Will synthetic text generators usher in a new age of creative thinking? The remarkable fluency of large language models may make them interesting tools for rapidly exploring semantic and stylistic spaces, yet the deceptive ease with which they generate output also provides countless new ways of appropriating ideas and erasing authorship. To better understand challenges and opportunities in this domain, we consider the interactional foundations of human creativity. Drawing on documented cases of creative collaboration, we argue that there is a world of a difference between text generation and true grasp — and that the friction and interaction that comes with the latter may be a meaningful element of human collaborative creativity.
It’s easy to forget amidst a rising tide of synthetic text, but language is not actually about strings of words, and language scientists would do well not to chain themselves to models that presume so. For apt and timely commentary we turn to Bronislaw Malinowski
Over two years ago I wrote about the unstoppable tide of uninformation that follows the rise of large language models. With ChatGPT and other models bringing large-scale text generation to the masses, I want to register a dystopian prediction: this enables a whole new form of monetization.
DALL-E, a new image generation system by OpenAI, does impressive visualizations of biased datasets. A widely circulated PR animation features meme-like koala dunking a baseball leading into an array of old white men — representing at one blow the past and future of representation and generation. This post jumps from reflections on techbros to the erasure of human labour in Cosmopolitan’s rushed “first AI magazine cover”.
Large language models make it entirely trivial to generate endless amounts of seemingly plausible text. The web is about to be engulfed in unending waves of algorithmically tuned AI-generated uninformation. This builds a feedback loop of uninformation feeding on uninformation. Counterintuitively, there was never a better time to be a scholar.