I didn't say p(doom) was zero. I said:
1. All estimates are pulled out of thin air
2. It makes little sense to attribute a probability to an event on which we have agency. If you do so, it's more of a statement about our collective ability to do the right thing than about the
RESEARCH
-
Probability of AI doom: agency and collective responsibility
By
–
-
FrankenMoEs and Mergekit Experiments in AI Model Fusion
By
–
That's so nice, it reminds me of frankenMoEs like Phixtral and Beyonder experiments with mergekit! (cc @chargoddard
) -
Boston Dynamics Spot Masters Whole-Body Manipulation with AI
By
–
Boston Dynamics’ Spot Masters Whole-Body Manipulation — Using AI to Drag, Roll, and Stack 15 kg Tires with Precision
— Ronald van Loon (@Ronald_vanLoon) 21 avril 2026
by @rai_inst#Robotics #Engineering #ArtificialIntelligence #Innovation #Technology pic.twitter.com/yDuIm45GmDBoston Dynamics’ Spot Masters Whole-Body Manipulation — Using AI to Drag, Roll, and Stack 15 kg Tires with Precision
by @rai_inst #Robotics #Engineering #ArtificialIntelligence #Innovation #Technology -
Beyond ChatGPT: The Full Spectrum of AI Technologies Explained
By
–
People see ChatGPT.
But AI is much bigger: ML, DL, Neural Networks, CV, NLP, Predictive Analytics, Speech Recognition, Agentic AI — the full iceberg beneath the surface. Great visual. Credit in image. #AI #ML #DeepLearning #Tech #Innovation -
Harnesses gaining equal recognition with AI models in surveys
By
–
Harnesses finally getting equal billing with models in survey form, this was overdue.
-
Prompting-Only Fix Improves Distribution Faithfulness Beyond Temperature
By
–
A prompting-only fix for distribution faithfulness is a really clean result, temperature knobs were never enough haha.
-
From 4o Vision Demo to Usable Computer Use: A Two-Year Journey
By
–
Took 2 years from the 4o Vision demo to reach usable computer use, way harder than the Twitter hype implied.
-
Breaking Down Scaling Laws in AI Research
By
–
The scaling-laws-as-one-thing conflation is everywhere, good to see someone finally breaking it out properly.
-
Hugging Face Ecosystem Powers SOTA ML Training Agents
By
–
Love this work from Aksel and the post-training team at Hugging Face!
— Thomas Wolf (@Thom_Wolf) 21 avril 2026
Turns out the HF ecosystem (papers, datasets, models all accessible through CLI, skills and md files) is perfect for running SOTA ML agents: agents that can train any type of AI model to top performance.
A… https://t.co/uakFTU5oD0Love this work from Aksel and the post-training team at Hugging Face! Turns out the HF ecosystem (papers, datasets, models all accessible through CLI, skills and md files) is perfect for running SOTA ML agents: agents that can train any type of AI model to top performance. A
-
Reconstructing AI timelines for informed audience understanding
By
–
One thing I am trying to do differently with these videos: not just explain who is right or wrong, but reconstruct the full timeline so people can make up their own mind with all the cards in hand. That part matters a lot in AI right now.