{"id":178066,"date":"2025-12-06T15:33:57","date_gmt":"2025-12-06T15:33:57","guid":{"rendered":"https:\/\/www.newsbeep.com\/ie\/178066\/"},"modified":"2025-12-06T15:33:57","modified_gmt":"2025-12-06T15:33:57","slug":"artificial-intelligence-research-has-a-slop-problem-academics-say-its-a-mess-artificial-intelligence-ai","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/ie\/178066\/","title":{"rendered":"Artificial intelligence research has a slop problem, academics say: \u2018It\u2019s a mess\u2019 | Artificial intelligence (AI)"},"content":{"rendered":"

A single person claims to have authored<\/a> 113 academic papers on artificial intelligence<\/a> this year, 89 of which will be presented this week at one of the world\u2019s leading conference on AI and machine learning, which has raised questions among computer scientists about the state of AI research.<\/p>\n

The author, Kevin Zhu, recently finished<\/a> a bachelor\u2019s degree in computer science at the University of California, Berkeley, and now runs Algoverse, an AI research and mentoring company for high schoolers \u2013 many of whom are his co-authors on the papers. Zhu himself graduated from high school in 2018.<\/p>\n

Papers he has put out<\/a> in the past two years cover subjects like using AI to locate nomadic pastoralists<\/a> in sub-Saharan Africa, to evaluate skin lesions<\/a>, and to translate<\/a> Indonesian dialects. On his LinkedIn, he touts publishing \u201c100+ top conference papers in the past year\u201d, which have been \u201ccited by OpenAI, Microsoft, Google, Stanford, MIT, Oxford and more\u201d.<\/p>\n

Zhu\u2019s papers are a \u201cdisaster\u201d, said Hany Farid, a professor of computer science at Berkeley, in an interview. \u201cI\u2019m fairly convinced that the whole thing, top to bottom, is just vibe coding,\u201d he said, referring to the practice of using AI to create software.<\/p>\n

Farid called attention to Zhu\u2019s prolific publications in a recent LinkedIn post<\/a>, which provoked discussion of other, similar cases among AI researchers, who said their newly popular discipline faces a deluge of low-quality research papers, fueled by academic pressures and, in some cases, AI tools.<\/p>\n

In response to a query from the Guardian, Zhu said that he had supervised the 131 papers, which were \u201cteam endeavors\u201d run by his company, Algoverse. The company charges $3,325 to high-school students and undergraduates for a selective 12-week online mentoring experience \u2013 which involves help submitting work to conferences.<\/p>\n

\u201cAt a minimum, I help review methodology and experimental design in proposals, and I read and comment on full paper drafts before submission,\u201d he said, adding that projects on subjects such as linguistics, healthcare or education involved \u201cprincipal investigators or mentors with relevant expertise\u201d.<\/p>\n

The teams used \u201cstandard productivity tools such as reference managers, spellcheck, and sometimes language models for copy-editing or improving clarity\u201d, he said in response to a query about whether the papers were written with AI.<\/p>\n

Bot watchers in turmoil<\/p>\n

The review standards for AI research differ from most other scientific fields. Most work in AI and machine learning does not go undergo the stringent peer-review processes of fields such as chemistry and biology \u2013 instead, papers are often presented less formally at major conferences such as NeurIPS<\/a>, one of the world\u2019s top machine learning and AI gatherings, where Zhu is slated to present.<\/p>\n

Zhu\u2019s case points at a larger issue in AI research, said Farid. Conferences including NeurIPS are being overwhelmed with increasing numbers<\/a> of submissions: NeurIPS fielded 21,575 papers this year, up from under 10,000 in 2020. Another top AI conference, the International Conference on Learning Representations (ICLR), reported a 70% increase in its yearly submissions for 2026\u2019s conference, nearly 20,000 papers, up from just over 11,000<\/a> for the 2025 conference.<\/p>\n

\u201cReviewers are complaining about the poor quality of the papers, even suspecting that some are AI-generated. Why has this academic feast lost its flavor?\u201d asked the Chinese tech blog 36Kr in a November post<\/a> about ICLR, noting that the average score reviewers had awarded papers had declined year-over-year.<\/p>\n

Meanwhile, students and academics are facing mounting pressure to rack up publications and keep up with their peers. It is uncommon to produce a double-digit number \u2013 much less triple \u2013 of high quality academic computer science papers in a year, academics said. Farid says that at times, his students have \u201cvibe coded\u201d papers to up their publication counts.<\/p>\n

\u201cSo many young people want to get into AI. There\u2019s a frenzy right now,\u201d said Farid.<\/p>\n

NeurIPS reviews papers submitted to it, but its process is far quicker and less thorough than standard scientific peer review, said Jeffrey Walling, an associate professor at Virginia Tech. This year, the conference has used<\/a> large numbers of PhD students to vet papers, which a NeurIPS area chair said compromised the process.<\/p>\n

\u201cThe reality is that often times conference referees must review dozens of papers in a short period of time, and there is usually little to no revision,\u201d said Walling.<\/p>\n

Walling agreed with Farid that too many papers are being published right now, saying he\u2019d encountered other authors with over 100 publications in a year. \u201cAcademics are rewarded for publication volume more than quality \u2026 Everyone loves the myth of super productivity,\u201d he said.<\/p>\n

On Zhu\u2019s Algoverse\u2019s FAQ page, answers discusses how the company\u2019s program can help applicants\u2019 future college or career prospects, saying: \u201cThe skills, accomplishments, and publications you achieve here are highly regarded in academic circles and can indeed strengthen your college application or r\u00e9sum\u00e9. This is especially true if your research is admitted to a top conference \u2013 a prestigious feat even for professional researchers.\u201d<\/p>\n

Farid says that he now counsels students to not go into AI research, because of the \u201cfrenzy\u201d in the field and the large volume of low-quality work being put out by people hoping to better their career prospects.<\/p>\n

skip past newsletter promotion<\/a><\/p>\n

A weekly dive in to how technology is shaping our lives<\/p>\n

Privacy Notice: Newsletters may contain information about charities, online ads, and content funded by outside parties. If you do not have an account, we will create a guest account for you on theguardian.com<\/a> to send you this newsletter. You can complete full registration at any time. For more information about how we use your data see our Privacy Policy<\/a>. We use Google reCaptcha to protect our website and the Google Privacy Policy<\/a> and Terms of Service<\/a> apply.<\/p>\n

\u201cIt\u2019s just a mess. You can\u2019t keep up, you can\u2019t publish, you can\u2019t do good work, you can\u2019t be thoughtful,\u201d he said.<\/p>\n

Slop flood<\/p>\n

Much excellent work has still come out of this process. Famously, Google\u2019s paper on transformers, Attention Is All You Need<\/a> \u2013 the theoretical basis for the advances in AI that led to ChatGPT \u2013 was presented at NeurIPS in 2017.<\/p>\n

NeurIPS organisers agree the conference is under pressure. In a comment to the Guardian, a spokesperson said that the growth of AI as a field had brought \u201ca significant increase in paper submissions and heightened value placed on peer-reviewed acceptance at NeurIPS\u201d, putting \u201cconsiderable strain on our review system\u201d.<\/p>\n

Zhu\u2019s submissions were largely to workshops within NeurIPS, which have a different selection process than the main conference and are often where early-career work gets presented, said NeurIPS organisers. Farid said he did not find this a substantive explanation for one person to put his name on more than 100 papers.<\/p>\n

\u201cI don\u2019t find this a compelling argument for putting your name on 100 papers that you could not have possibly meaningfully contributed to,\u201d said Farid.<\/p>\n

The problem is bigger than a flood of papers at NeurIPS. ICLR used<\/a> AI to review a large volume submissions \u2013 resulting in apparently hallucinated citations and feedback that was \u201cvery verbose with lots of bullet points\u201d, according to a recent article in Nature.<\/p>\n

The feeling of decline is so widespread that finding a solution to the crisis has become the subject of papers itself. A May 2025 position paper<\/a> \u2013 an academic, evidence-based version of a newspaper op-ed \u2013 authored by three South Korean computer scientists that proposed a solution to the \u201cunprecedented challenges with the surge of paper submissions, accompanied by growing concerns over review quality and reviewer responsibility\u201d, won an award for outstanding work at the 2025 International Conference on Machine Learning.<\/p>\n

Meanwhile, says Farid, major tech companies and small AI safety organisations now dump their work on arXiv, a site once reserved for little-viewed preprints of math and physics papers, flooding the internet with work that is presented as science \u2013 but is not subject to review standards.<\/p>\n

The cost of this, says Farid, is that it is almost impossible to know what\u2019s actually going on in AI \u2013 for journalists, the public, and even experts in the field: \u201cYou have no chance, no chance as an average reader to try to understand what is going on in the scientific literature. Your signal-to-noise ratio is basically one. I can barely go to these conferences and figure out what the hell is going on.\u201d<\/p>\n

\u201cWhat I tell students is that, if what you\u2019re trying to optimize publishing papers, you know, it\u2019s actually honestly not that hard to do. Just do really crappy low-quality work and bomb conferences with it. But if you want to do really thoughtful, careful work, you\u2019re at a disadvantage because you\u2019re effectively unilaterally disarmed,\u201d he said.<\/p>\n","protected":false},"excerpt":{"rendered":"A single person claims to have authored 113 academic papers on artificial intelligence this year, 89 of which…\n","protected":false},"author":2,"featured_media":178067,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[220,218,219,61,60,80],"class_list":{"0":"post-178066","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-artificial-intelligence","8":"tag-ai","9":"tag-artificial-intelligence","10":"tag-artificialintelligence","11":"tag-ie","12":"tag-ireland","13":"tag-technology"},"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/posts\/178066","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/comments?post=178066"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/posts\/178066\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/media\/178067"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/media?parent=178066"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/categories?post=178066"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/ie\/wp-json\/wp\/v2\/tags?post=178066"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}