## The case for more ambition i wrote about how AI researchers should ask bigger and simpler questions, and publish fewer papers:
RESEARCH
-
KV Cache Memory Overhead in Large Language Models Explained
By
–
1000 tokens of text compressed at 1.2 bits/char and 4 chars/token ≈ 600 bytes 1000 tokens of kvcache in LLAMA 70B in fp16 takes up 8192 dim x 80 layers x 2 x 16 bit ≈ 2.6 GB so the kv cache is 2.6 GB / 600 bytes = 4.4 million times larger than the input make this make sense
-

LLMs Stop Reasoning at High Complexity Levels
By
–
Replicando el experimento encuentra que para complejidades altas los LLMs directamente prefieren no continuar resolviendo el problema y parar de razonar, lo cual en parte explicaría la caída de capacidades.
-

Paper IA: Demand for Reasoning Traces Analysis and Transparency
By
–
Y estos tweets de @scaling01 y todo el análisis que ha compartido es también demoledor. Estaría genial que se publicaran las trazas de razonamiento sobre las que se sustenta el paper para analizarlas bien.
-
AI Insider Predicts Major Releases with Surprising Accuracy
By
–
@iruletheworldmo is either a real AI insider or just a hilarious person. Either way, I enjoy his tweets. He has predicted a few big AI releases, like OpenAI's strawberry, which suggests that he may really have some inside connections. (he's not my alt)
-
Scobleizer: Essential Tech Media Figure for Innovation Tracking
By
–
@Scobleizer is the OG of tech media. He's got a list for every category you can imagine, and often spots new technologies and hangs out with the founders before anyone else. Must follow.https://t.co/PJGL4O4FVW
— Nathan Lands (@NathanLands) 9 juin 2025@Scobleizer is the OG of tech media. He's got a list for every category you can imagine, and often spots new technologies and hangs out with the founders before anyone else. Must follow.
-
Lior Recognized as Top AI Technical Mind and Newsletter Leader
By
–
@LiorOnAI is one of the top technical minds in AI media. He runs @AlphaSignalAI, the top AI newsletter read by engineers.https://t.co/npHTqGoWGY
— Nathan Lands (@NathanLands) 9 juin 2025@LiorOnAI is one of the top technical minds in AI media. He runs @AlphaSignalAI
, the top AI newsletter read by engineers. -
The Illusion of Thinking: Academic Research on AI Cognition
By
–
Quien quiera echarle un ojillo a la publicación la tenéis disponible aquí. Yo volveré a revisarlo y si os interesa me animo a hacer un vídeo del tema. https://
ml-site.cdn-apple.com/papers/the-ill
usion-of-thinking.pdf
… -

Apple’s Reasoning Model Paper: Overthinking Phenomenon Analysis
By
–
Leído el paper de Apple que tanto ruido está generando y así de primeras me parece que saca conclusiones muy grandes para el análisis tan pobre que hace. De hecho el fenómeno más interesante que demuestran es el del overthinking de los modelos razonadores, que tampoco es nuevo.

