No ratings yet.

Over the last year, I have used many different GPTs on the OpenAI platform — mostly the
ordinary ChatGPT function, but also tools to visualize certain ideas. The experiences vary a
lot from one another.

In my opinion, text-to-image and text-to-video GenAIs are improving rapidly, but they are
still nowhere near real videos or pictures. When an AI-generated video appears on my feed, it
is immediately clear that it is fake. I am aware that I may recognize them because of my daily
exposure to this kind of content. My grandma, however, could be much easier to trick. When
these text-to-image and text-to-video GenAIs improve further, this could become dangerous.
What picture is real? People could start using fake images and videos for many harmful
purposes. In my opinion, there are far more downsides and potential negative consequences
than positive ones.

During the early days of ChatGPT, I was not yet convinced of its utility and helpfulness as a
tool. Many of the results it produced were simply wrong, or the chatbot did not give the
answer I was looking for. However, in the last couple of months, major improvements have
been made. The GenAI seems to be “thinking” for a while before answering questions or
performing certain tasks, which has significantly improved the quality of the output in my
opinion. Furthermore, I think it’s great that the current answers often end with a suggestion I
might not have thought of myself — for example: “Here is the data you were looking for.
Would you like me to visualize it in a graph?” Yes please!

However, ChatGPT still “hallucinates,” meaning it sometimes provides made-up sources or
fabricated citations. This is frustrating and reduces the positive impact on productivity since I
have to consistently check whether the information is correct. I also see a potential danger
arising — that GenAI tools might try to cover up these hallucinations by simply inventing
sources. In that case, all information on the internet would be diluted in terms of credibility.
But that seems a little apocalyptic… or not?

5/5 (1)

China’s two leading short-video platforms share a surface similarity but pursue distinct strategies. Kuaishou emphasizes grassroots creators and trust-based commerce in lower-tier communities, with revenues spanning ads, livestream e-commerce, and gifting. Douyin (TikTok’s Chinese sibling) leans on entertainment scale and ad performance, layering in rapid e-commerce growth (Tang et al., 2022).

These choices surface in ranking design. Kuaishou invests in long-term user health, using reinforcement-learning objectives tied to retention which is an explicit shift from short-term clicks to durable engagement (Cai et al., 2023). In parallel, the fair-ranking literature shows why platforms that monitor exposure inequality can mitigate winner-take-all dynamics among creators (Biega, Gummadi, & Weikum, 2018; Do & Usunier, 2022).

Douyin, by contrast, publicly emphasizes a highly personalized “For You” recommender that scales content based on interaction signals (watch time, shares, follows), now with added transparency controls which is a classic accelerator design (Feng, 2025).

The theory of two-sided markets suggests both models are rational: unchecked same-side effects like creator crowd-out can weaken ecosystems, so platforms must govern distribution to keep value flowing (Eisenmann et al., 2006; Van Alstyne et al., 2016). Kuaishou builds the brakes into its ranking, while Douyin optimizes for speed and patches with policy.

In my view, Kuaishou’s approach feels more sustainable. By ensuring smaller creators gain visibility, it fosters loyalty and builds a broader base for commerce. It reflects a deliberate choice: growth through fairness. Douyin’s accelerator excels at grabbing attention, but the risk is burnout. That could be the case for both users tired of repetitive content and creators crowded out of discovery.

If forced to bet on the long game, I’d pick the algorithm and business model of Kuaishou. Scale delivers quick wins, but healthier network effects come from balancing growth with inclusion which may prove to be Kuaishou’s hidden advantage.

References

Biega, A. J., Gummadi, K. P., & Weikum, G. (2018). Equity of attention: Amortizing individual fairness in rankings. Proceedings of the 41st International ACM SIGIR Conference on Research and Development in Information Retrieval, 405–414. https://doi.org/10.1145/3209978.3210063

Cai, Q., Liu, S., Wang, X., Zuo, T., Xie, W., Yang, B., Zheng, D., Jiang, P., & Gai, K. (2023). Reinforcing user retention in a billion scale short video recommender system. Proceedings of the ACM Web Conference 2023, 1273–1282.https://doi.org/10.1145/3543873.3584640

Do, V., & Usunier, N. (2022). Optimizing generalized Gini indices for fairness in rankings. Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, 737–747.https://doi.org/10.1145/3477495.3532035

Eisenmann, T. R., Parker, G., & Van Alstyne, M. W. (2006). Strategies for two-sided markets. Harvard Business Review, 84(10), 92–101.

Feng, C. (2025, April 2). ByteDance-owned Douyin sheds light on recommendation algorithm amid regulatory pressure. South China Morning Post. https://www.scmp.com/tech/big-tech/article/3304799/bytedance-owned-douyin-sheds-lights-recommendation-algorithm-amid-regulatory-pressure

Van Alstyne, M. W., Parker, G. G., & Choudary, S. P. (2016). Pipelines, platforms, and the new rules of strategy. Harvard Business Review, 94(4), 54–60, 62.