Sitemap

Member-only story

AI Weekly Digest: New Launches and the Hidden Threats of AI in Democracy

4 min readOct 11, 2025

--

Week 41, 2025

This week’s stories show an industry continuing to jump into the mainstream. We’re reaching a critical point in time where we need to decide (and research) how AI will impact our political systems.

This week Nvidia improves reasoning models. OpenAI releases AgentKit, turning agent creation into a visual process. The Guardian raises alarms about AI’s long-term impact on democracy. And an article from Harvard reminds leaders that human judgment remains irreplaceable in the age of AI.

And we have a final shameless self-promotion easter egg🥚

Let’s start!

Nvidia Teaches LLMs to “Think” Before Speaking

Nvidia researchers introduced Reinforcement Learning Pre-Training (RLP), a method that rewards large models for generating internal reasoning chains before predicting text.

Stronger logic, fewer hallucinations, and better generalization without extra supervision is the result from their findings.
If proven scalable, RLP could mark a shift toward built-in reasoning during pre-training, not just during interaction with the users.

Read more on VentureBeat

OpenAI’s AgentKit Turns Agents Into Workflows

--

--

Ivo Bernardo
Ivo Bernardo

Written by Ivo Bernardo

Writing about AI and its consequences | Partner @ DareData | Instructor @ Udemy | also on thedatajourney.substack.com/ and youtube.com/@TheDataJourney42