11 Comments
User's avatar
Laura Ferraz Baick's avatar

My ChatGPT usage decreased A LOT since GPT5 launch. Great post mapping out hallucination.

Expand full comment
Elena Calvillo's avatar

GPT5 is much much better!

Expand full comment
Antonio Avalos's avatar

“If you put bad or poor inputs it gets you bad or poor outputs”

This is something that echoes my mind every time “hallucinations” is mentioned about AI. We need to refocus our work in “prompt/context engineering” to get the best out of AI.

But your post highlights something so important, the training! It will shape the output to not make look bad an idea, question or person.

So I guess this will change when we break our uses to agents instead of the model directly.

Expand full comment
Joel Salinas's avatar

Elena, excellent post! And thanks for the mention :)

Expand full comment
Elena Calvillo's avatar

Thanks, Joel!

Expand full comment
Dario Radecic's avatar

Honored to be on this list, thank you!

Expand full comment
Elena Calvillo's avatar

Thanks to you, keep going! 🚀

Expand full comment
Alex Pawlowski's avatar

Fair points, AI is ultimately tue most heterogeneous thing ever 🙏🏻

Expand full comment
Elena Calvillo's avatar

Facts! 💯

Expand full comment
Product Career Hub's avatar

AI is a sharp intern, not the boss. One thing that works great is to run a 10-min pre-mortem where you cite sources, write the counter-case, set the rollback, and name one owner.

Expand full comment
Elena Calvillo's avatar

I love this advice! Thanks for sharing it!

Expand full comment