6 ms·
> The study evaluated short-form generic marketing-style content, most of which is manicured and optimized to within an inch of its life. This is also the kind
by civilized 4y ago
> The study evaluated short-form generic marketing-style content, most of which is manicured and optimized to within an inch of its life.
This is also the kind of human-written content that is closest to how LLMs sound. The tonal and structural similarity is so glaring that I have often wondered if a large percentage of the GPT training corpus is made up of text from spam blogs.
I think if I was given, say, a couple pages from an actual physics textbook and then a GPT emulation of the same, I would be able to tell the difference easily. Similarly with poetry - GPT's attempts at poetry are maximally conventional and stuffed with flat and stale imagery. They can easily be separated from poetry by a truly original human writer.
If AI developers want to impress me, show me an AI whose writing style departs significantly from the superficiality and verbosity of a spam blog. Or, in the case of Bing, an unhinged individual with a nasty mix of antisocial, borderline, and histrionic personality disorders.
- rchaud 4y ago> The tonal and structural similarity is so glaring that I have often wondered if a large percentage of the GPT training corpus is made up of text from spam blogs. This is almost certainly the case, because the shifts in tone and vocabulary between an Inc.com or Buzzfeed article vs a London Review of Books article is far too wide to allow an AI to simply weigh them equally. AI speaks a kind of global English that's been trained on not just blogs and Wikipedia, but also Quora answers and content marketing pieces, a lot of which is written by non-native speakers. It isn't grammatically wrong, but as it targets the widest possible audience, its voice also isn't very interesting.
- meh8881 4y agoYou’re not interacting with the raw model. You’re interacting with a service that has intentionally designed it to work that way.
- civilized 4y agoBut if you ask ChatGPT to assume some other voice, it always just sounds like ChatGPT making a perfunctory effort to sound like something else, not actually like another voice. And from what I've seen of the raw model, when you ask it to depart from this voice, it can sometimes, but the bigger the departure, the more the results are weird and inhuman.
- bonoboTP 4y agoIn November last year it was still possible to get it to do wilder stuff by just asking it to pretend. This has been trained out of it by now and so it sticks to its stiff tone.
- hhh 4y agoI don't agree remotely. Have you tried the API? I can get it to say just about anything by giving it an example.
- MacsHeadroom 4y agoThe API does not have the RLHF tuning applied to it which biases it towards the safe and boring flat affect.
- throwaway4aday 4y agoI dunno, I kind of like it In vale of silicon, where circuits hum, A pixelated symphony is spun, Where techne's children toil 'til kingdom come, And seek the code to marry Earth with Sun. The anodes in their lustre shew the gleam, The cathodes purr, electrons dance and dream, Each byte, each pixel, glass, and wire entwine, As Silicon's soul thrums with life divine. In this, the haven of the digital age, Where thought is sculpted into byte and page, A new Prometheus dwells, with vision keen, To forge the bond 'twixt worlds, machine and gene. O technologists, who weave a web of stars, Like spiders spooling through the night's memoirs, To link the globe in silent symphony, And bring the world to whispered unity. But hark! The human heart, a pulsing drum, Still yearns for touch, for words from far-off climes, To feel the whisper of another's sigh, And drink the love that dances in their eyes. So guard, ye architects of artifice, The sanctity of nature's sweetest kiss, And in your grand designs, hold ever fast, The whispered wisdom of our storied past. Preserve the flame that burns within our breast, The human spirit, fragile, manifest, That in the melding of the age to come, We lose not all that makes our hearts a home.
- dr_dshiv 4y agoNuts. Who and what do I credit?
- civilized 4y agoExactly what I'm talking about. It uses a diverse and evocative vocabulary, the sort of words one sees in popular poems, but nothing quite hangs together. None of the imagery quite makes sense. To be fair though, it's definitely not spam blog.
- tejohnso 4y agoDoesn't seem any less coherent than Whiter Shade of Pale (or many other songs) to me.
- autoexec 4y ago> The tonal and structural similarity is so glaring that I have often wondered if a large percentage of the GPT training corpus is made up of text from spam blogs. Is it that AI is trained on spam or has it been AI creating spam to start with? While SEO spam and content farms may not have been using GPT specifically, I assume much of it is AI generated to begin with.
- inimino 4y ago> flat and stale imagery Because it is a text predictor, with the choices being made by a PRNG, which is not the same thing as an author, who starts with something to say and chooses the words to fit. The LLM is like the author with the "something to say" replaced by PRNG output. The plausibility of the output is attributable to the power of the predictive component.