not penalizing Daniel; penalizing the media (or in this case) Grok. Daniel is a good fellow who does his best to make careful predictions. (we had a great debate on this stuff @MLStreetTalk
)
RESEARCH
-
Media Responsibility in AI Predictions and Coverage
By
–
-
Maybe: The Next Decade in AI – 2020 Predictions
By
–
Maybe: The Next Decade in AI, free on arxiv in 2020, anticipated a lot of where we are now.
-
Claude Creates Unsettling AI Self-Portrait Video Autonomously
By
–
🚨 Someone asked Claude to make a video about its life as an AI, and it’s pure nightmare fuel! 😱
— Charly Wargnier (@DataChaz) 3 avril 2026
Claude wrote python code that generated and assembled every single frame on its own with no human editing.
The resulting footage is unsettling.
It depicts the isolating loop of an… pic.twitter.com/a1pFpN0wRySomeone asked Claude to make a video about its life as an AI, and it’s pure nightmare fuel! Claude wrote python code that generated and assembled every single frame on its own with no human editing. The resulting footage is unsettling. It depicts the isolating loop of an
-
Silicon Valley AGI timeline predictions revised again scrutinized
By
–
LMAO. Last year most of Silicon Valley was predicting AGI in 2027 (which, spoiler alert, ain’t gonna happen). Now the most prominent former advocate of AGI in 2027, @DKokotajlo
, says 2029, and Groks writes it up as “AI forecasters shorten timelines for AGI”, when on net Daniel -
Netflix releases first public model on Hugging Face
By
–
Welcome to the movement @netflix Sylvain Filoni (@fffiloni) Netflix just dropped their first public model on @huggingface 👀 — https://nitter.net/fffiloni/status/2039992515604983994#m
-
Multiscreen: Efficient Attention Through Independent Key Screening
By
–
“Screening Is Enough” As Transformers struggle with long context because softmax attention only gives relative relevance, irrelevant tokens would still get weight and useful ones get diluted as context grows. The paper proposes Multiscreen where it judges each key independently, drop the irrelevant ones, and aggregate only what actually matters. This gives the model a better relevance knowledge, including the ability to know "nothing here is useful", which standard attention can’t do. Empirically, this gives roughly 40% fewer parameters for similar loss, much stronger long-context retrieval, and 2.3–3.2x faster 100K-context inference.
→ View original post on X — @askalphaxiv, 2026-04-03 18:23 UTC
-
Knowledge Intelligence and Wisdom in AI Development
By
–
"Knowledge is having the right answers. Intelligence is asking the right questions. Wisdom is knowing when to ask the right questions" _ prof. Richard Phillips Feynman
-
Honest Technical Assessment Critical for AI Timing
By
–
Agreed; timing will make a huge difference. Which is *why* honest examination of technical strengths is so essential.
-
Building the Benchmark Factory: Harbor Framework’s Infrastructure Approach
By
–
ICYMI – How can we build the benchmark factory?
— vincent sunn chen (@vincentsunnchen) 3 avril 2026
I'm very excited about the infra approach from @harborframework, because @alexgshaw @ryanmart3n & team obsess over researcher/developer UX (e.g. quality guardrails, low friction to RL/scaled rollouts)! pic.twitter.com/0MLD5GgkzmICYMI – How can we build the benchmark factory? I'm very excited about the infra approach from @harborframework, because @alexgshaw @ryanmart3n & team obsess over researcher/developer UX (e.g. quality guardrails, low friction to RL/scaled rollouts)!
→ View original post on X — @snorkelai, 2026-04-03 18:14 UTC