https://archive.ph/gx01C

The core thesis of this paper is that the AI community needs to stop treating autonomous agents as just another text generation problem and start building comprehensive infrastructure to support closed loop learning. The authors argue that achieving reliable agentic behavior requires a full stack ecosystem that unifies data synthesis with sandboxed execution and specialized reinforcement learning. To prove this point they introduce the Agentic Learning Ecosystem which consists of an RL framework called ROLL alongside a sandbox manager named ROCK and an agent interface known as iFlow CLI. They believe that isolating models in static training environments is a dead end for solving complex real world workflows.

The team developed an open source model named ROME using a tightly integrated training pipeline with reproducible execution environments which allowed a relatively small 30 billion parameter model to rival or beat massive proprietary models exceeding 100 billion parameters on difficult software engineering benchmarks.

A big part of their argument rests on the idea that credit assignment in reinforcement learning needs to change. They propose a novel algorithm called Interaction Perceptive Agentic Policy Optimization which shifts the reward focus from individual text tokens to broader semantic interaction chunks. This chunk level optimization stabilizes the training process over long horizons and prevents the policy collapse often seen in complex tool use scenarios.

We're increasingly seeing a shift of priorities away from raw data scale and focus on the systematic infrastructure as the actual bedrock of next generation models.

"Breaking Free: Pathways to a fair technological future" is a new report from Forbrukerrådet. The report itself is a light read: it's in English, and while it is 100 pages long [PDF], it is in fact enjoyable and even amusing – we laughed quite a few times when reading it. For one thing, it contains a surprising number of puns and the occasional starred-out swearword, such as "Do androids dream of electric s***." A stodgy bureaucratic report this is not.

https://youtu.be/T4Upf_B9RLQ

cross-posted from: https://lemmy.ml/post/44059967

for those not familiar with Mark Pilgrim, he is/was a prolific author, blogger, and hacker who abruptly disappeared from the internet in 2011.

cross-posted from: https://lemmy.bestiver.se/post/968527

HN comments

Shocking news, indeed. I had no idea they had a Discord.

An Amazon Web Services data center in the United Arab Emirates suffered a multi-hour outage on Sunday after unidentified “objects” struck the facility and triggered a fire. The incident occurred around 4:30 a.m. local time and affected the availability zone mec1-az2 in the ME-CENTRAL-1 region.

The fire department cut power to combat the flames, resulting in significant disruptions to cloud services. Given the simultaneous Iranian retaliatory attacks on the Gulf states, suspicion arises that the impacting objects may have been missiles or drones. Amazon has not confirmed anything on its part.

I had been using a magic eraser (never on the screen) and have not yet suffered any ill effects as a result (although I also rarely clean my laptop at all), but I gather that this is not recommended any longer (if it ever was). Alcohol wipes are good for the screen, but not as effective for the keyboard and other non-screen parts. Any suggestions?

cross-posted from: https://lemmy.ml/post/43923170

We're happy to announce a long-term partnership with Motorola. We're collaborating on future devices meeting our privacy and security standards with official GrapheneOS support.

https://motorolanews.com/motorola-three-new-b2b-solutions-at-mwc-2026/

The Banality of Artificial Intelligence

What happens when an AI hallucination leads to bombing an elementary school?

By Michael Altfield
License: CC BY-SA 4.0
https://tech.michaelaltfield.net/

It appears likely that the US government is using Anthropic, OpenAI, Google and/or xAI data models for processing signals intelligence (SIGINT), for AI-generated "kill lists" to determine where to drop their bombs.

[right] This AI datacenter is a machinery of war. Its LLM hallucinations decide which children to assassinate [left] This IG Farben (Bayer/BASF) factory in Auschwitz produced Zyklon B for the Nazis, who murdered over a million children

In Apr 2024, +972 (an Israeli news outlet) published a >9,000 word article describing how the Israeli military had been using Artificial Intelligence to decide which (residential) buildings, hospitals, and schools to bomb in Gaza.

In Feb 2026, the US (and Israel) bombed Iran -- killing over 100 schoolchildren (and Ali Khamenei).

In Mar 2026, it appears that the US has likely built a similar system, leveraging US AI companies' tech to decide which (school) buildings to bomb, false-positive hallucinations be damned.

Who targeted the Shajareh Tayyiba girls' elementary school in Minab, Iran? Could it have been an AI hallucination? A false-positive?
...

Read the full article here:

• https://tech.michaelaltfield.net/2026/03/03/ai-venezuela-iran/