There are always a flood of posts about what AI can or cannot do, so it is worth pausing and paying attention to this one. It is a very hard test, done without tools. It was also viewed as an unlikely goal. Prediction markets had the chance of this happening this year as 20%
GENERATIVE AI
-
GPT-5 spotted in biosec benchmark commit
By
–
BREAKING : GPT-5 has been spotted as “gpt-5-reasoning-alpha-2025-07-13” in the biosec benchmark commit. h/t @swishfever
-
Major Mistakes to Avoid Before Building AI Solutions
By
–
Before You Build with AI, Watch THIS (Big Mistakes to Avoid):
-
Model Stacking Strategy in AI Applications
By
–
4. Model stacking is underrated Genspark doesn’t just use one model it picks the best one for the job: → GPT-4.1, o-3 pro, o4-mini-high
→ Claude
→ Gemini
→ And more ChatGPT? Only OpenAI models. No switching. -
Comparing AI Agent Performance in Presentation Creation
By
–
I tested OpenAI’s new Agent vs Genspark on the same task. One gave me a full presentation in 5 minutes. The other made me open Google Slides. Here’s what happened
-
Marcus wrong again: OpenAI proves AI capabilities beyond predictions
By
–
Y por supuesto, Marcus dándose prisa en declarar que "la IA no puede" antes de que OpenAI le demostrara que, una vez más, se volvía a equivocar. Rápido, a mover de nuevo la portería!
-
New AI Models and Capabilities Discovered Unexpectedly
By
–
Ni idea, hasta hace un par de horas ni sabíamos que estos nuevos modelos existían ni que estas capacidades eran posibles
-
Training Data Curation vs AI Model Reasoning Capability
By
–
For special purpose models, fully curated training data and thought patterns are the obvious way to go, but models that are meant to seriously help you think will either be able to infer dangerous ideas from incomplete data or they will be useless
-
AI Capabilities Growing Exponentially Solving Complex Problems
By
–
Las capacidades de la IA está creciendo en órdenes de magnitud cada año, resolviendo cada vez problemas más y más complejos, que ya pasan de pensar unos pocos segundos a estar pensando horas!
-
Reinforcement Learning Challenges Training Complex AI Reasoners
By
–
El uso de RL para entrenar a los razonadores actuales encontraba el problema de que para problemas muuuuy complejos que requieren mucho tiempo para resolverlos y verificar que estaban correctos (y darle feedback a la IA de que lo ha hecho bien o mal para mejorar su