RT @ClementDelangue: We keep saying we want open-source frontier agents. Fine. Then let’s build the dataset. @badlogicgames, creator of P…
→ View original post on X — @huggingface, 2026-04-06 15:37 UTC
AI
-
Automatic cars eliminate need for physical buttons
By
–
My car is completely automatic. Buttons make very little sense in that world.
-
Gary Marcus: His Warnings About Sam Altman Confirmed by Investigation
By
–
Marcus, August 2024 @guardian: "OpenAI's Sam Altman is becoming one of the most powerful people on Earth. We should be very afraid." @RonanFarrow, 2026, @NewYorker: "Sam Altman May Control Our Future—Can He Be Trusted?
New interviews and closely guarded documents shed light on the persistent doubts about the head of OpenAI." Maybe people will listen now? — Ronan Farrow (@RonanFarrow) The reporting on OpenAI and Sam Altman that I've been working on for the past year and a half, for @NewYorker, with @andrewmarantz: newyorker.com/magazine/2026/… [Translated from EN to English]→ View original post on X — @garymarcus, 2026-04-06 15:32 UTC
-
AGIALPHA Tends Towards Infinity
By
–
#AGIALPHA => ♾️ [Translated from EN to English]
→ View original post on X — @montreal_ai, 2026-04-06 15:29 UTC
-
Audit reveals Claude Code’s token waste
By
–
Este usuario auditó 926 sesiones de Claude Code y descubrió que la mayor parte del desperdicio de tokens provenía de su plataforma.
— SONIA (@S0N_IA_) 6 avril 2026
Todos culpan a Anthropic por las limitaciones, así que decidió analizar los datos.
858 sesiones, 18 903 turnos y un gasto estimado de $1619 en 33… pic.twitter.com/eulMkfgK8vThis user audited 926 sessions of Claude Code and discovered that most of the token waste came from its platform. Everyone blames Anthropic for the limitations, so he decided to analyze the data. 858 sessions, 18,903 turns, and an estimated spend of $1,619 in 33 days. This is
-
Serverless Spark Jobs: Simplified Data Processing Without Cluster Management
By
–
Running Scala and Java Spark jobs often requires managing clusters and handling runtime upgrades. Serverless JARs remove that overhead. Jobs start in seconds and run on the latest supported runtime without cluster management. With Databricks Connect, you can build and test from
-
What permitting requirements apply to AI systems?
By
–
What kind of permitting is involved with these?
-
Business Leaders Spreading AI Misinformation, Economic Bubble Risk
By
–
This quote from @EricRWeinstein — “ our business people become our designed public thinkers by default” — looms heavily now, as business people with enormous economic interests busy themselves redefining and opining on terms like “AGI” and “superintelligence”. Their lies and distortions around the alleged imminence of these things are likely to cause a crippling blow to the world’s economy. And the business media’s common “A CEO said a thing” journalistic style, in which they parrot absurd claims—without serious skepticism—is only exacerbating all this When the bubble pops, and taxpayers are forced to bail out this mess, the cultlike adoration of business people who know little about how intelligence actually works will have been a major factor. Gary Marcus (@GaryMarcus) Allowing “our business people become our designated public thinkers [b]y default” has indeed been an absolute catastrophe. — https://nitter.net/GaryMarcus/status/2028859980896751660#m
→ View original post on X — @garymarcus, 2026-04-06 15:26 UTC
-
OSGym: Infrastructure for Training Computer Use Agents
By
–
So why did the researchers develop OSGym? Training computer use agents is both a modeling problem & an infrastructure challenge: • Each task needs a full OS (GUI, apps, browser, etc.) • Scaling → massive CPU/RAM/disk costs • Random crashes from real software environments • Thousands of parallel runs required It's a key bottleneck, though it's not often discussed. [Translated from EN to English]
→ View original post on X — @mit_csail, 2026-04-06 15:25 UTC
-
MIT CSAIL Introduces OSGym for Training Computer Agents
By
–
How do you train AI agents that can use computers like humans? 🧵 MIT CSAIL researchers introduce "OSGym": scalable OS infrastructure to improve the capabilities of computer use agents. It introduces large-scale training made possible by extensive infrastructure optimization: bit.ly/47JprPd [Translated from EN to English]
→ View original post on X — @mit_csail, 2026-04-06 15:25 UTC