Author | Lightnews

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Oct 8

there are scattered copies of papers all over the internet, the training data of AI models aren't up-to-date.

Until companies improve how they filter retracted papers, users of AI tools should take steps to verify the tool's outputs.

www.technologyreview.com/2025/09/23/1...

AI models are using material from retracted scientific papers

Some companies are working to remedy the issue.

www.technologyreview.com

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Oct 8

Groups like @retractionwatch.com maintain a database of retracted papers, and some companies are starting to use these databases now to filter papers, to a certain extent. But, such measures are not foolproof—retraction databases are not comprehensive...

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Oct 8

Especially when people use AI tools to seek medical or health information, use of retracted papers can carry grave consequences.

The rate of retractions has been raising over the years, with more than 10,000 retractions in 2023 alone, according to a Nature analysis. bsky.app/profile/rich...

Richard Van Noorden @richvn.bsky.social · Dec 12

Milestone: 2023 is the first year with more than 10,000 research paper retractions -- smashing previous records. More than 8,000 of these came from Hindawi (mostly from 'special issues'). Total retractions now >50,000. My analysis for Nature.
www.nature.com/articles/d41...

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Oct 8

There’s “kind of an agreement that retracted papers have been struck off the record of science and the people who are outside of science—they should be warned that these are retracted papers," Yuanxi Fu told me.

Yet, AI tools continue to use these papers to answer user questions.

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Oct 8

ChatGPT, Perplexity, Ai2 ScholarQA and other AI tools cite retracted scientific papers without any warning. Read my report for @technologyreview.com

www.technologyreview.com/2025/09/23/1...

AI models are using material from retracted scientific papers

Some companies are working to remedy the issue.

www.technologyreview.com

1

Reposted by Ananya (ಅನನ್ಯ)

Richard Van Noorden @richvn.bsky.social · Aug 20

Researchers argue over whether ‘novel’ AI-generated works use others’ ideas without credit.
By Ananya | @nature.com
www.nature.com/articles/d41...

What counts as plagiarism? AI-generated papers pose new risks

Researchers argue over whether ‘novel’ AI-generated works use others’ ideas without credit.

www.nature.com

1 1 6

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Aug 27

Researchers are developing AI tools to generate novel research ideas and papers. But, there's a concern that these tools might just be reusing existing ideas without credit. I dug into the debate for @nature.com
www.nature.com/articles/d41...

What counts as plagiarism? AI-generated papers pose new risks

Researchers argue over whether ‘novel’ AI-generated works use others’ ideas without credit.

www.nature.com

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Aug 27

For @nature.com, I write about the book Amphibious Anthropologies, which explores what it means to live in wet environments — where the boundaries between land and water are constantly redrawn — all over the world.
www.nature.com/articles/d41...

Why amphibious, wet environments hold the key to climate adaptation

As the planet braces for climate-change-induced fluctuations, the wisdom of communities living at the intersection of land and water could offer valuable lessons.

www.nature.com

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Oct 11

Indian researchers suspect they were passed over for major awards after criticizing government policies. New by me for
Science Magazine

science.org/content/arti...

Indian government accused of political meddling in science prizes

Researchers suspect they were passed over for major awards after criticizing government policies

science.org

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Jul 16

Today's tests cannot provide any meaningful assessment of AI's ability to reason or understand, and yet there are so many claims that AI systems have humanlike cognitive abilities. I report on the current state of evaluation practices for @sciencenews.bsky.social
www.sciencenews.org/article/ai-u...

AI's understanding and reasoning skills can't be assessed by current tests

Assessing whether large language models — including the one that powers ChatGPT — have humanlike cognitive abilities will require better tests.

www.sciencenews.org

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Apr 26

Researchers found that OpenAI's speech recognition model, Whisper, fabricated about 1.4% of the audio transcriptions tested, about 40% of which were harmful or concerning in some way. It's hard to spot them without listening to the audio again.
My report for Science: www.science.org/content/arti...

AI transcription tools ‘hallucinate,’ too

Study finds surprisingly harmful fabrications in OpenAI’s speech-to-text algorithm

www.science.org

2

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Apr 4

Thanks, Andrew! I will take a look at the paper

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Apr 1

LLM-based multilingual chatbots are becoming common nowadays. But these chatbots might not be the best at answering your healthcare queries, especially if you ask in Hindi, Mandarin and Spanish. My latest for
@sciam.bsky.social exploring the problems.
www.scientificamerican.com/article/chat...

Chatbots Struggle to Answer Medical Questions in Widely Spoken Languages

www.scientificamerican.com

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Mar 26

...catch everything.
The models and what goes into them, behind them, all of it is often hidden. Deemed proprietary knowledge by companies making these models.
But, w/o access to this information, it's hard to say what problems might exist, where exactly the biases arise and how best to squash them.

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Mar 26

Broadly, these models learn to make associations between images and their captions. But the captions themselves can contain incorrect, incomplete, biased, and, as
@abeba.bsky.social and team found, harmful content (which increased as the dataset size increased) . And automated filtering doesn't...

1 1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Mar 26

The models often rely on racist and sexist stereotypes to generate images, sometimes even amplifying bias. Stereotypes related to gender, skin colour, occupations, nationalities, geographies and more.

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Mar 26

...we've sort of decided as a society that that's not what one should do and putting them in immediately forces a decision about stereotypes, she says. Yet, that's what we can ask these models to do.

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Mar 26

One thing that stuck with me from my conversation with Ria is that concepts like poor or kind - words that can't be imaged from a societal perspective - can be fed in as much as any other word, like red or car. It's really harmful to put them into these models because...

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Mar 26

Most solutions - writing better prompts, adding sample images - are band aid fixes, and don't address the underlying problems in how these systems are built. There are many ways for these models to be biased that we haven't figured out yet. And there's certainly no way to automate safety.

1 1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Mar 26

For Nature, I spoke to researchers about tracing sources of bias in image-generating AI, and if they can be fixed.
www.nature.com/articles/d41...

With inputs from @abeba.bsky.social, @rajiinio.bsky.social, Pratyusha Ria Kalluri, Kathleen Fraser and Will Orr.

AI image generators often give racist and sexist results: can they be fixed?

Researchers are tracing sources of racial and gender bias in images generated by artificial intelligence, and making efforts to fix them. Researchers are tracing sources of racial and gender bias in i...

www.nature.com

2 5 10

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Mar 4

For Al Jazeera, I report on how things keep getting worse for India's Urban Company workers and trainees.
www.aljazeera.com/economy/2024...

India’s Urban Company revolutionised gig work for women. Then it bled them

Multiple costs to qualify for work leads, forced product purchases, high rate of blockings have stripped their earnings.

www.aljazeera.com

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Dec 13

Recently, many generative AI tools have been released for public use and more and more companies are looking to integrate these tools into their workflows. What makes them popular and what are the concerns? I talked to
@melaniemitchell.bsky.social to know more. sciencenews.org/article/gene...

Generative AI grabbed headlines this year. Here’s why and what’s next

Prominent artificial intelligence researcher Melanie Mitchell explains why generative AI matters and looks ahead to the technology’s future.

sciencenews.org

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Oct 28

Find the collection here: archive.org/details/Serv...

Internet Archive: Digital Library of Free & Borrowable Books, Movies, Music & Wayback Machine

archive.org

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Oct 28

Free, accessible and well-stocked public libraries are hard to find in India. For @techreview.bsky.social, I write about an effort to digitize books in various Indian languages. technologyreview.com/2023/10/25/1...

The grassroots push to digitize India’s most precious documents

The Servants of Knowledge collection on the Internet Archive is an effort to make up for the lack of library resources in India.

technologyreview.com

1

Ananya (ಅನನ್ಯ) @punarpuli.bsky.social · Sep 8

Algorithms are increasingly being used to make various decisions about our lives, decisions that are often invisible to us. Can we trust those decisions? For Scientific American, I write about how algorithms can go wrong.

www.scientificamerican.com/article/algo...

Algorithms Are Making Important Decisions. What Could Possibly Go Wrong?

Seemingly trivial differences in training data can skew the judgments of AI programs—and that’s not the only problem with automated decision-making

www.scientificamerican.com

1 5