@bartekwlodarczyk
PL/EN • Assistant prof. UW | LIS | knowledge organization & AI | Chinese social media. Adiunkt UW | INiB | organizacja wiedzy & AI | chińskie media społecznościowe. https://orcid.org/0000-0001-9229-4656 Private opinions | Prywatne opinie.
The 2025 COS Impact Report is now live! From open infrastructure to research initiatives, policy engagement, and global partnerships, see what COS accomplished over the past year to advance transparent, rigorous, and accessible research.
🔍 Check out the report: www.cos.io/impact
Chińska bezpieka i ChatGPT
https://wp.me/p3fv0T-i7s #Japonia #Chiny #ChRL #dezinformacja #bezpieka #OpenAi #AI #Qwen #DeepSeek #ChatGPT #POLECANE
🔥 Generative AI is reshaping scientific research. But do these systems actually understand science? In our new paper, we peek inside LLMs to understand how they encode scientific quality, revealing several monosemantic features that predict research quality.
📄 arxiv.org/pdf/2602.19115
Second, the author shows that the addition of synthetic data considerably improves the values of metrics such as F1 score or recall@5. It can be leveraged in other studies and in practice.
I’ve noticed two important elements in the paper. First, in the organic-only setting (Table 5), the classical-model ensemble scores higher than the transformer models across F1@5, precision@5, recall@5, and NDCG@5. This suggests classical ensembles are still useful in different scenarios (p. 709).
What is the performance of ML and ensemble models in automating LCSH assignment?
doi.org/10.1108/EL-0...
Why read? It compares classical algorithms with transformers & assesses synthetic data use.
What’s next? Research on the best combination of ML models for indexing.
#ShortReview #AI #LCSH #KO
At the same time, China is also using AI to improve many sectors, such as healthcare.
doi.org/10.1007/s135...
#AI #China
AI can yield many positive results (e.g., learning how to code), but there are also significant risks of unacceptable, unethical, and harmful use. Here’s how China is looking at its potential surveillance applications.
#AI #China
Maciej Kawecki (This is IT) przeprowadził interesujący wywiad z Anthonym Aguirre na temat rozwoju autonomicznej sztucznej inteligencji i związanych z tym zagrożeń. Warto posłuchać wypowiedzi amerykańskiego fizyka o jednym z promowanych kierunków rozwoju AI.
youtu.be/himTPjX098w
#AI #AGI #Wywiad
It is interesting that such a small 0.5B model achieved strong performance (first-level category). This supports the potential of small language models as a lower-cost option for LIS researchers. This approach may be relevant for classification of various resources.
#AI #LLM #Classification #KO
How can AI classify multilingual research datasets?
doi.org/10.1108/EL-0...
Why read? It shows a practical pipeline using a fine-tuned Qwen2 to assign CLC codes to multilingual datasets.
Next step: More detailed cross-language evaluation (authors).
#ShortReview #AI #LLM #Classification #Datasets
@hankhplee.bsky.social and colleagues show shifts in critical thinking such as a greater emphasis on information verification. It’s an important contribution. The question remains: what are we really losing?
#AI #CriticalThinking
What is the impact of generative AI on knowledge workers' critical thinking?
doi.org/10.1145/3706...
Why read? It shows what shapes critical thinking when using AI and how AI is changing it.
What's next? Research on changes over a longer period of time.
#ShortReview #AI #CriticalThinking
Concepts and terms are at the core of knowledge organization. The increasing role of AI in KO processes and systems requires us to use the consistent and correct terminology.
#AI #KO #Terminology
The most dangerous situation is when we use this terminology without knowing about different meanings. Our interlocutors, listeners at the conference, or students at a lecture can mean something completely different.
#AI #Terminology
The author analyzes only two terms (AI ethics, AI peer review), but it is a good point of departure for a broader discussion on AI terminology.
#AI #Terminology
Is the AI-related terminology precise and correct?
doi.org/10.1108/JD-0...
Why read? It shows inconsistencies in AI terminology and their consequences.
What’s next? A need for systematic work on AI terminology in LIS and KO.
#ShortReview #AI #LIS #KO #Terminology
Finansowanie polskiej nauki to tylko jeden z jej problemów, chociaż istotny. Potrzebna jest systemowa zmiana podejścia do roli nauki w Polsce.
#Nauka #Polska
Artykuł Karoliny Nowakowskiej z "Gazety Prawnej" dobrze pokazuje problem niskich wynagrodzeń w polskiej nauce na przykładzie fizyki, a więc dyscypliny bardzo ważnej dla rozwoju technologii.
serwisy.gazetaprawna.pl/edukacja/art...
#Nauka #Fizyka #Polska #PAN
However, in the top 50 ranks: US 24, mainland China 10, Hong Kong (China) 4.
www.shanghairanking.com/rankings/gra...
#LIS #GRAS #USA #China
In the 2025 ShanghaiRanking GRAS (Library & Information Science), 7 of the top 10 universities are in the US. The remaining three are from mainland China, Hong Kong (China), and Canada.
www.shanghairanking.com/rankings/gra...
#LIS #GRAS #USA #China
The Wuhan Lockdown
A diary entry
Six years ago today, Wuhan started its 76-day lockdown. So close yet so far
Polecam wywiad Patrycjusza Wyżgi (didaskalia) z Remigiuszem Kinasem na temat modelu Bielik i szerzej o AI.
youtu.be/gDJclWCw1qY
#AI #LLM #Bielik #Wywiad
DCMI 2026 will be in Seoul (August 3-7, 2026). The conference will explore the relationship between AI and metadata. The submissions are due by 30 March (Papers/Panel/Workshop) and 1 May (other contributions).
www.dublincore.org/conferences/...
#AI #Metadata #KO #Conference #DCMI #Korea
As China pushes a national reading initiative and AI reshapes information, one Shanghai shop offering 1,000 publications represents a vanishing era of relative editorial variety. chinamediaproject.org/2026/01/16/s...
Preprint: Regulatory Gray Areas of LLM Terms (via arXiv): arxiv.org/abs/2601.08415 #AI #LLMs #GenAI #regulations #researchers
Anthropic's New Cowork Feature Brings Claude Code's Agent Capabilities to People Who Don't Write Code (via @thedecoder.bsky.social) the-decoder.com/anthropics-n... #AI #GenAI #LLMs
"Are You Dead Yet?" just became China's #1 app. The name sparked heated debate—but 125 million people living alone might need it anyway. lingua-sinica.org/are-you-dead...
Google, Apple enter into multi-year AI deal for Gemini models reut.rs/4bse4NZ
The submission for the 19th International ISKO Conference, which will be held in São Paulo (August 17–19, 2026), is open until March 1, 2026. There are three thematic axes: epistemological, applied, and cultural & social.
www.isko2026.com.br/isko-2026/pa...
#KO #ISKO #Conference #Brazil