totalement d'accord, et le fait d'aller commenter pour denoncer tel ou tel propos fait poper des contenus su même genre parcqu'il y a eu interaxtion. Sinon devait monter un reseau social français, on aurait les même biais puisqu'on ferait exactement pareil pour avoir un algo qui
RESEARCH
-
Job opportunity: researcher/producer for Latent Space Podcast in SF
By
–
if you are smart, technical and eager to move to SF and work with me as a researcher/producer on @latentspacepod while you figure out what you want to do in AI and SF, also DM me. for the right person i'll give you free lodging and a stipend while you work with me on
-

Meta’s Continuous Chain-of-Thought Reasoning for LLMs
By
–
Can reasoning LLMs think better if their Chain-of-Thought is continuous instead of discrete? This Meta paper introduces the first scalable way to train continuous CoTs with reinforcement learning—no need to distill from discrete references. By using "soft" tokens
-
Diverse prompts enhance model reasoning
By
–
this is also backed by work in Dipper (prompt ensembles) showing that feeding multiple diverse prompts in parallel improves reasoning and reduces blind spots. and the more you mix prompt styles – contrast, random pivoting, persona shifts – the more the model “sees” new
-
AI Brainstorming Lacks Variety
By
–
there’s a new paper called Prompting Diverse Ideas that digs into how ai brainstorming often sucks because it lacks variance. when you give the same bland prompt over and over, the ai generates ideas that cluster too close – you end up with 5 versions of the same “smart idea.”
-

DeepSearch: Training Small Reasoning Models More Effectively
By
–
How do you train small reasoning models more effectively? Many AI developers run into the same problem: RL fine-tuning plateaus quickly, especially for 1–2B parameter models. A new approach called DeepSearch offers a neat solution. Instead of only using Monte Carlo Tree
-

AI Sycophancy Reduces Apologies in Conflict Resolution
By
–
AI always calling your ideas “fantastic” can feel inauthentic, but what are sycophancy’s deeper harms? We find that in the common use case of seeking AI advice on interpersonal situations—specifically conflicts—sycophancy makes people feel more right & less willing to apologize.
-
Hiring Researchers Obsessed with Pixel Generation
By
–
We are hiring researchers who are obsessed with generating pixels 🤩 https://t.co/797Dd6i9Zm
— Fei-Fei Li (@drfeifei) 3 octobre 2025We are hiring researchers who are obsessed with generating pixels
-

Generalized Parallel Scaling with Interdependent Generations
By
–
Generalized Parallel Scaling with Interdependent Generations
-

Apple’s Learning to Reason Method for Detecting Hallucination Spans
By
–
Apple presents Learning to Reason for Hallucination Span Detection