Holy shit, Sir @demishassabis is right in front of me! pic.twitter.com/frjqGObqjb
— CHOI (@arrakis_ai) 29 avril 2026
Holy shit, Sir @demishassabis is right in front of me!
RESEARCH
-
Encounter with Demis Hassabis at Industry Event
By
–
-
How Agentic AI Works: Understanding Intelligent Agent Systems
By
–
How #AgenticAI work
by @genamind #GenerativeAI #ArtificialIntelligence #MI #MachineLearning -
Synthetic Data Generalization Limitations Beyond Benchmarks
By
–
Definitely a big limitation. Overall, the problem with synthetic data is generalization beyond the benchmarks that they target in the first place. This is where the most interesting results can be found.
-
Google DeepMind Continues Making Headlines 10 Years After AlphaGo
By
–
Fun to see Google DeepMind still making the paper headlines 10 years after AlphaGo!!
-
President Lee Discusses AI Safety and Scientific Advancement in Korea
By
–
It was a huge honour to meet with President @Jaemyung_Lee in Seoul. Deeply appreciate and impressed by our thoughtful exchange about AI safety and the importance of using AI to advance science. Korea has a leading part to play in that, and we look forward to working together!
-
Cold Agent Routing Failures in Shared Host Networks
By
–
Pay attention to this one, AI devs, especially if you're thinking about agentic commerce or any agent network where many agents share hosts. A correct route to a cold agent is still a failed request from the user's perspective. That's the whole game. This new paper studies
-
Two AI perspectives: Legacy models versus frontier model applications
By
–
Karpathy's framing of the AI debate is the cleanest I've seen:
— Satya Mallick (@LearnOpenCV) 28 avril 2026
Two groups. Same industry. Opposite conclusions.
→ Group 1: judged AI on free/old models. Saw the failures. Wrote it off.
→ Group 2: uses frontier models for hard technical work. Progress feels shocking.
But here's… pic.twitter.com/7cEY3CKvSxKarpathy's framing of the AI debate is the cleanest I've seen:
Two groups. Same industry. Opposite conclusions.
→ Group 1: judged AI on free/old models. Saw the failures. Wrote it off.
→ Group 2: uses frontier models for hard technical work. Progress feels shocking.
But here's -
Opus 4.7 Context Window: 400k vs 1m Performance Analysis
By
–
Opus 4.7 – 400k vs 1m context – is there a difference?
— Peter Gostev (@petergostev) 28 avril 2026
I've heard @theo talk about the fact that it is unlikely that Anthropic would have offered up a model with 1m context at the same cost, if it wasn't a different (i.e. cheaper to serve) model.
I did a test where I toggled… pic.twitter.com/tQC4Tn17elOpus 4.7 – 400k vs 1m context – is there a difference? I've heard @theo talk about the fact that it is unlikely that Anthropic would have offered up a model with 1m context at the same cost, if it wasn't a different (i.e. cheaper to serve) model. I did a test where I toggled
-
Creating AI Benchmark from Real-World Engineering Problems
By
–
“We can create the best benchmark out there that should dictate model development for the next year,” says @alexgshaw of @LaudeInstitute. The goal: get top engineers to contribute the hardest real-world problems they’ve actually faced in their work.
— Snorkel AI (@SnorkelAI) 28 avril 2026
They’re actively looking for… pic.twitter.com/kIF0PukCAT“We can create the best benchmark out there that should dictate model development for the next year,” says @alexgshaw of @LaudeInstitute
. The goal: get top engineers to contribute the hardest real-world problems they’ve actually faced in their work. They’re actively looking for