Assertion Supported AI assessment confidence: 95% certainty 4/5 debate potential 2/5

Eck: As of 2017, AI cannot generate a single coherent text paragraph

Doug Eck · Making Music and Art Through Machine Learning - Doug Eck of Magenta · Y Combinator · Jul 21, 2017 · at 25:51

Doug Eck, research scientist leading Google's Project Magenta, clarifies the technical limitations of state-of-the-art language models in 2017.

0:00 / 0:10exact quote · 10.7s

▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark

“So, so everybody understands that's listening or watching, you know, we can, we can't generate a coherent paragraph, right? So, I don't mean we, magenta. I mean, kind of we, humanity.”

quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →

The claim, plainly

As of July 2017, AI researchers across humanity cannot generate a coherent paragraph of text using generative models.

a quote alone can be ambiguous, so this restates it with the references and context of the conversation around it resolved. AI-written; the tape below the player is the source.

Assessment

As of July 2017, natural language generation literature and state-of-the-art neural models were incapable of reliably generating coherent multi-sentence paragraphs that preserved global semantic context and discourse structure. Contemporaneous benchmarks—such as Andrej Karpathy's 2015 RNN analyses, Krause et al.'s CVPR 2017 hierarchical image-paragraph experiments, and Wiseman et al.'s July 2017 data-to-document evaluation—consistently documented that while neural models produced locally fluent sentences, they suffered from repetitive phrases, hallucinations, and rapid loss of topic consistency across paragraphs. The Transformer architecture had only just been introduced in June 2017 for machine translation and parsing, and reliable multi-sentence coherence was not demonstrated until larger autoregressive models such as GPT-2 in 2019.

Andrej Karpathy's May 2015 benchmark demonstrated that character-level LSTMs/RNNs maintain syntactic plausibility only over short sequences and lose macro-level semantic coherence over paragraphs.
Krause et al. (CVPR 2017) demonstrated that even hierarchical sentence-and-word RNN architectures suffered from hallucinations and poor inter-sentence coherence.
Wiseman, Shieber, and Rush (July 2017 / EMNLP 2017) demonstrated that neural seq2seq models failed to produce coherent multi-sentence document summaries, trailing rule-based baselines in macro-structure and semantic consistency.
The Transformer paper (Vaswani et al., June 2017) addressed sentence-level sequence transduction tasks (machine translation) rather than open-ended coherent multi-sentence paragraph generation.

Checked against

[1] github.io [2] manning.com [3] postshift.com [4] aclanthology.org [5] arxiv.org [6] github.com [7] aclanthology.org [8] arxiv.org

"Challenges in Data-to-Document Generation" Wiseman 2017 "A Hierarchical Approach for Generating Descriptive Image Paragraphs" Krause 2017 "The Unreasonable Effectiveness of Recurrent Neural Networks" coherence text generation "coherent paragraph" RNN LSTM 2016 2017 generative models text coherence "paragraph" state of the art 2017 "Attention is All You Need" Vaswani June 2017 "Language Models are Unsupervised Multitask Learners" "coherent" "paragraphs"

assessed as of 2026-09-05

This assessment is an AI reading of a public statement, made against the cited references with the confidence shown. It can be wrong. The quote, the tape and the references are all on this page; if something is off, use report an error below. full disclaimer →

Mentioned

generative language models humanity Magenta

⚑ Spot an error? Report it

More from Doug Eck

Opinion

Eck: Alex Graves advanced LSTMs more than anyone, including its creator

“Among the three of us, by far, Alex Graves has done the most with LSTM. So he continued, after he finished his PhD, and he continued doggedly to try to understand how recurrent neural networks worked, how to train them, and how to make them useful for sequence…”

Doug Eck Jul 21, 2017 ▶ 17:39 Making Music and Art Through Machine Learning - Doug Eck of Magenta · Y Combinator

Assertion Supported

Eck: Untrained listeners rated Magenta compositions as more Bach-like than Bach

“When we put these tunes out for, like, untrained listeners to listen to, they sometimes voted them as sounding more Bach-y”

Doug Eck Jul 21, 2017 ▶ 22:14 Making Music and Art Through Machine Learning - Doug Eck of Magenta · Y Combinator

Disclosure

Eck: Magenta aims to build creative tools, not act as artists

“The reason that I put that quote there, I think, is to be honest with the division between engineering and research and artistry, and to not think that what I'm doing is being a machine learning artist, but we're trying to build interesting ways to make new ki…”

Doug Eck Jul 21, 2017 ▶ 1:20 Making Music and Art Through Machine Learning - Doug Eck of Magenta · Y Combinator

Insight

Eck: Users' first instinct with AI art models is breaking them

“The first thing you're gonna do, if you think, if someone comes to you and says, here's this really smart model that you can make art with, what are you gonna do? You're gonna try to show the world that it's a stupid model, right? But maybe the way that, maybe…”

Doug Eck Jul 21, 2017 ▶ 2:03 Making Music and Art Through Machine Learning - Doug Eck of Magenta · Y Combinator

Opinion

Eck: Magenta's generated AI music is not yet good enough for testing

“I, at least for Magenta, I haven't felt like the quality of what we've been generating has been good enough to bother, so to speak. Like you find it, you cherry pick, you find some good things, you're like, okay, this model trains, and it's interesting, and no…”

Doug Eck Jul 21, 2017 ▶ 5:08 Making Music and Art Through Machine Learning - Doug Eck of Magenta · Y Combinator

Insight

Eck: Audio model generation errors sound musical because of learned training variance

“I think they're interesting because the model has been forced to capture some of the important sources of variance in real music audio, and even when it Fails to reproduce all of them when it fills in with confusion, so to speak. Even that confusion is, is som…”

Doug Eck Jul 21, 2017 ▶ 12:05 Making Music and Art Through Machine Learning - Doug Eck of Magenta · Y Combinator

Eck: As of 2017, AI cannot generate a single coherent text paragraph

More from Doug Eck

Turn any episode into a week of clips.