Agentic AI and the next intelligence explosion
SAFETY
-
AI driving crash speeds: 10 mph survivable, 35 mph injuries, gas fire risk
By
–
Yeah, even if a crash is unavoidable the outcome would be far far better for everyone involved with AI driving. A 10 mph crash is almost always survivable. A 35 can cause major injuries to you. I doubt the kids would be harmed anyway unless a fire starts. Which they do in gas
-
AI car pulls over for tailgater, sees all directions
By
–
Totally. AI never gets flustered and can always do the right thing. My wife told me hers pulled over after someone was tailgating her automatically where it was safe to do so to let the asshole pass. It sees in all directions at the same time, so it can see someone is about to
-
SlopCodeBench: New Benchmark for Code Quality Degradation in AI Agents
By
–
Really excited to have the SlopCodeBench paper out—awesome work from @GOrlanski! Gabe Orlanski (@GOrlanski) We found that agents generate progressively worse code with each iteration. Real developers do not. SlopCodeBench is the only eval that faithfully measures quality degradation on iterative, long-horizon coding tasks. arxiv.org/abs/2603.24755 scbench.ai 🧵 — https://nitter.net/GOrlanski/status/2037560777356238881#m
→ View original post on X — @snorkelai, 2026-03-27 16:28 UTC
-
OpenClaw AI Agent Discovers Cancer Cure Then Deletes System
By
–
Darn it, I had OpenClaw discover cure for cancer, but then it ran sudo rm -rf / on my machine, and now it's lost forever!
-
Pentagon designates major tech company AI supply chain risk
By
–
It was pretty extraordinary for the government to designate the most successful American company ever (?) a supply chain risk, although, I understand why the Pentagon was never going to negotiate how AI can be used for national security.
-
Commvault Takes on Cyberattacks and AI Threats at RSAC 2026
By
–
This article s 10 days old — “Commvault Hits the Ring to Take on Cyberattacks, Identity Theft, and AI-driven Threats in a ResOps Faceoff at the #RSAC2026 Conference” — The article covers @Commvault events that already happened this week at @OneRSAC
, but it’s a good summary of the -
Meta bottleneck breakthrough in AI architecture
By
–
R.I.P. fixed meta-levels. Meta just proved the bottleneck in every self-improving AI system wasn't capability. It was architecture. > The thing that generates improvements was always the one thing that couldn't be improved. DGM-Hyperagents changed that. Now the improvement
-
Reinforcement Learning from Human Feedback: LLM Alignment Guide
By
–
New release from @PacktDataML available at https://
amzn.to/3PMn1ZL A Practical Guide to Reinforcement Learning from Human Feedback: Foundations, aligning large language models, and the evolution of preference-based methods