Zofiliaporno May 2026

Are LLMs following the correct reasoning paths?


University of California, Davis University of Pennsylvania   ▶ University of Southern California

We propose a novel probing method and benchmark called EUREQA. EUREQA is an entity-searching task where a model finds a missing entity based on described multi-hop relations with other entities. These deliberately designed multi-hop relations create deceptive semantic associations, and models must stick to the correct reasoning path instead of incorrect shortcuts to find the correct answer. Experiments show that existing LLMs cannot follow correct reasoning paths and resist the attempt of greedy shortcuts. Analyses provide further evidence that LLMs rely on semantic biases to solve the task instead of proper reasoning, questioning the validity and generalizability of current LLMs’ high performances.

zofiliaporno
LLMs make errors when correct surface-level semantic cues-entities are recursively replaced with descriptions, and the errors are likely related to token similarity. GPT-3.5-turbo is used for this example.

zofiliaporno The EUREQA dataset

Download the dataset from [Dataset]

In EUREQA, every question is constructed through an implicit reasoning chain. The chain is constructed by parsing DBPedia. Each layer comprises three components: an entity, a fact about the entity, and a relation between the entity and its counterpart from the next layer. The layers stack up to create chains with different depths of reasoning. We verbalize reasoning chains into natural sentences and anonymize the entity of each layer to create the question. Questions can be solved layer by layer and each layer is guaranteed a unique answer. EUREQA is not a knowledge game: we adopt a knowledge filtering process that ensures that most LLMs have sufficient world knowledge to answer our questions.
EUREQA comprises a total of 2,991 questions of different reasoning depths and difficulties. The entities encompass a broad spectrum of topics, effectively reducing any potential bias arising from specific entity categories. These data are great for analyzing the reasoning processes of LLMs

Image 1
Categories of entities in EUREQA
Image 2
Splits of questions in EUREQA.

Zofiliaporno May 2026

Gaming is no longer a subculture; it is the dominant form of media. Platforms like Fortnite and Roblox act as social squares where users attend virtual concerts and socialize, proving that media is now a space you inhabit, not just a screen you watch.

However, the real disruption lies in . Platforms like YouTube and TikTok have democratized media production. An independent creator in their bedroom now competes for the same "eyeball time" as a multi-million dollar television production. In this new era, the algorithm is the new programmer, surfacing content based on individual psyche rather than broad demographics. The Rise of Immersive Experiences

The future of entertainment and media content is . As technology like AI begins to assist in content creation—from writing scripts to generating photorealistic visuals—the volume of content will only explode. The challenge for the future isn't finding something to watch; it’s finding the signal within the noise.

Virtual and Augmented Reality are beginning to move beyond novelty, offering "presence"—the feeling of actually being inside a news story or a fictional world. The Personalization Paradox

zofiliaporno Analyses and discussion

Gaming is no longer a subculture; it is the dominant form of media. Platforms like Fortnite and Roblox act as social squares where users attend virtual concerts and socialize, proving that media is now a space you inhabit, not just a screen you watch.

However, the real disruption lies in . Platforms like YouTube and TikTok have democratized media production. An independent creator in their bedroom now competes for the same "eyeball time" as a multi-million dollar television production. In this new era, the algorithm is the new programmer, surfacing content based on individual psyche rather than broad demographics. The Rise of Immersive Experiences

The future of entertainment and media content is . As technology like AI begins to assist in content creation—from writing scripts to generating photorealistic visuals—the volume of content will only explode. The challenge for the future isn't finding something to watch; it’s finding the signal within the noise.

Virtual and Augmented Reality are beginning to move beyond novelty, offering "presence"—the feeling of actually being inside a news story or a fictional world. The Personalization Paradox

Acknowledgement

This website is adapted from Nerfies, UniversalNER and LLaVA, licensed under a Creative Commons Attribution-ShareAlike 4.0 International License. We thank the LLaMA team for giving us access to their models.

Usage and License Notices: The data abd code is intended and licensed for research use only. They are also restricted to uses that follow the license agreement of LLaMA, ChatGPT, and the original dataset used in the benchmark. The dataset is CC BY NC 4.0 (allowing only non-commercial use) and models trained using the dataset should not be used outside of research purposes.