BullshitBench update – Kimi K2.6, GLM-5.1, Qwen-3.6-Plus. Kimi and Qwen are in the top 15-25, but GLM is pretty low down 73/86 Kimi-K2.6:
– Kimi K2.6 improved vs K2.5: 52% -> 65% pushback (green) with no reasoning
– Kimi reasoning hurt: K2.6 fell from 65% green at none to 50% at
RESEARCH
-
Kimi K2.6 and Qwen-3.6 Top LLM Benchmark Rankings
By
–
-
MathNet: Global Multimodal Benchmark for Mathematical Reasoning and Retrieval
By
–
MathNet a Global Multimodal Benchmark for Mathematical Reasoning and Retrieval paper: https://
huggingface.co/papers/2604.18
584
… -
Extending consciousness to the stars requires vastly more advancement
By
–
It’s a start. Vastly more will be needed to extend consciousness to the stars.
-
Beyond Simple Combination: Deeper AI Understanding Required
By
–
Vous vous couvrez de ridicule
— Docteur Laurent Alexandre (@dr_l_alexandre) 21 avril 2026
Vous racontez n’importe quoi !
Travaillez le sujet au lieu de parler de l’IA comme si on était en 1985
C’est franchement pathétique de penser que l’IA ne fait que combiner… https://t.co/Psbq1yj75hVous vous couvrez de ridicule Vous racontez n’importe quoi ! Travaillez le sujet au lieu de parler de l’IA comme si on était en 1985 C’est franchement pathétique de penser que l’IA ne fait que combiner…
-
One-Step Image Generation Extended from Class Labels to Text
By
–
Extending One-Step Image Generation from Class Labels to Text via Discriminative Text Representation
— AK (@_akhaliq) 21 avril 2026
paper: https://t.co/8jpoGrAqRP pic.twitter.com/JO0J5dkT47Extending One-Step Image Generation from Class Labels to Text via Discriminative Text Representation paper: https://
huggingface.co/papers/2604.18
168
… -
OpenGame: Open Agentic Coding Framework for Games
By
–
OpenGame
— AK (@_akhaliq) 21 avril 2026
Open Agentic Coding for Games
paper: https://t.co/dlI0HnM7Hm pic.twitter.com/q9TZFHAkK6OpenGame Open Agentic Coding for Games paper: https://
huggingface.co/papers/2604.18
394
… -
Nano Banana Pro outperforms gpt-image-2 in satellite map accuracy
By
–
> a satellite map of London, showing a close up of the area around Big Ben Left: Nano Banana Pro
Right: gpt-image-2 Seeing NBP has more accuracy than gpt-image-2 for this prompt. Westminster bridge should be next to Big Ben. The road layouts, park positions and buildings are -
Nous Research Hermes Agent Self-Evolution Rewrites Its Own Prompts
By
–
Nous Research built an AI that rewrites its own brain for $2. Most AI agents run the same prompts forever. One open-source repo changes that. Hermes Agent Self-Evolution lets agents rewrite their own prompts, skills, and code. No manual tuning needed. The engine powering
-
Is There an AI Bubble? Expert Perspectives on Market Viability
By
–
Is there an AI bubble?
#AI #AIio #AIInnovation #ML #DataScience #Futureofwork @demishassabis @Ronald_vanLoon @TamaraMcCleary @geoffreyhinton @goodfellow_ian @jeffdean @erikbryn -
Somatic Mutations in Microglia Drive Alzheimer’s Neuroinflammation
By
–
Somatic mutations in microglia brain cells (the same mutations as seen in cancer) are enriched in and may drive the neuroinflammation of Alzheimer's disease @CellCellPress https://
cell.com/cell/fulltext/
S0092-8674(26)00341-7
…