dejanseo
Joined 12 years ago
Karma
331
Submissions
24
Member since
Oct 22
I use machine learning and mechanistic interpretability to optimize brand visibility in AI search: https://dejan.ai/
View on Hacker NewsRecent submissions
Comment · 2 months ago
Bulls eye!
https://z.ai/blog/glm-5.3-flash
Comment · 2 months ago
Comment · 2 months ago
This looks spot on, where is it from?
Comment · 2 months ago
Hey, it may be trivial, I'm pretty proud of it :)
Comment · 2 months ago
Hello! I wrote the above article and did the NCD on model outputs. The same thought crossed my mind when I saw Gemma misclassified as Gemini quite frequently. And GLM almost as Claude and not as Gemini at all. Gave me the feeling as if GLM didn't train on Gemini generated synthetic data at all but mainly on Claude and GPT.
Comment · 2 months ago
Author here. Tell me more. How do you use NCD and what is your concern about its application here? Should I have use da larger reference sample? You can test it in action here: https://dejan.ai/tools/ai/ (e.g. drop a claude article or GLM article in and see what it says, it's not perfect but reasonably good).
Comment · 2 months ago
> "Claude models launched on or after August 2, 2026 support marking at launch."
No Anthropic model has been launched in August.
Comment · 8 years ago
Google? When I brought a serious issue up in 2012 https://dejanseo.com.au/hijack/ Google never fixed it:
In summary, I can take any of your (or anyone else's content) pass more pagerank to it than the original page and then I become the original page. Not only that but all your inbound links now count towards my site and I can see your links in Search Console of my domain.
This is something link graph theory refers to as "link inversion" and is very harmful to smaller publishers.
Comment · 8 years ago
The irony...
Comment · 8 years ago
Trying to be objective and understand my own motivations here. Obviously I didn't do anything out of malice. But yes, I could have told Google directly about the problem, but then I'd have no cool story to publish on my blog. At the end of the day, that's what it boils down to. Now that I got too much attention from it, I regret all of it.
Comment · 8 years ago
I hope so, and I also hope Chrome gets a fix for this.
Comment · 8 years ago
Thank you! :)
Comment · 8 years ago
Thank you. I'm not having a good time at the moment. Anyway, the basis of my test hypothesis is that people are easily fooled by URL both by HTTPS and brand recognition (e.g. subdomain) so I conducted a survey which revealed the very real problem: https://dejanseo.com.au/trust/
Raw data: https://dejanseo.com.au/wp-content/uploads/2017/04/survey-te...
Comment · 8 years ago
Thank you, I did mess up and wish I could take it back. To everyone bashing on me, I'm truly sorry to offend so many people. That was not the intention. This was purely as you describe it, intellectual curiosity.
I really appreciate your comment and hope it's OK that I added it here: https://dejanseo.com.au/competitor-hack/#shawn
Comment · 8 years ago
Not the first time either. Every now and again I get an interesting idea, test it and share it with the world. The test that was left forgotten had no commercial impact on anyone and very low traffic.