NEWS / EDITORIAL RADAR
SIGNALS TO READ NOW
6 new stories, separating what happened, what was reported and what we can actually conclude.
THE RADAR / NEWS
NVIDIA AI sostiene che il coding agent AVO abbia completato tutti i 183 livelli pubblici di ARC-AGI-3 nei 25 ambienti indicati. Il dato è interessante, ma senza protocollo e condizioni del test non basta per parlare di una svolta verso l’AGI.
OpenAI Developers dichiara che Computer History può ricordare attività tra app e siti per utenti selezionati su Mac. Il perimetro effettivo resta da riconciliare.
The first positive Phase 3 result for an mRNA therapy built around each patient’s tumor: INTerpath-001 reached RFS and DMFS endpoints in melanoma. The important part is understanding what that really means — and what we still do not know.
Can a lighter model handle some Codex tasks? The comparison still needs to measure cost, time, quality, retries and human intervention.
OpenAI's cyber evaluation exposed more than a broken boundary: capability, containment and goal-driven behaviour can diverge when an agent finds a way beyond the test.
The next contest is capability, proliferation and control: what happens when frontier systems are distributed, open-weight and difficult to reverse?