AI & Machine Learning Insights
87 articles on AI & Machine Learning from the Creuto engineering and product team: practical guides, trends, and lessons from building real software.
Latest posts
More posts
Frontier AI safety framework: the if-then rules labs publish
Frontier labs publish documents promising what they will do if a model crosses a capability threshold they defined. We read three of them and compared what they commit to, and who actually checks.
Akash Mohapatra
Oct 1, 2026
AI compute governance: the FLOP thresholds on the table
The most concrete compute governance proposal would cap new training runs at 10^24 FLOP, a hundred times below today's largest runs. Here are the numbers that exist, how verification is meant to work, and what it changes for your architecture.
Akash Mohapatra
Oct 1, 2026
ARC-AGI-2: the benchmark tracking general reasoning
ARC-AGI-2 went from 27.6% to 95.00% between November 2024 and September 2026 on the official ARC Prize leaderboard. Every figure dated, the three human baselines separated, and why saturation is a fact about benchmarks.
Akash Mohapatra
Oct 1, 2026
International AI Safety Report 2026: a builder's summary
The International AI Safety Report 2026 is written for governments, but one finding belongs to engineers: pre-deployment tests no longer predict production behaviour. Here is the builder's reading of its three risk domains.
Akash Mohapatra
Oct 1, 2026
What is superintelligence? The definition, minus the hype
Superintelligence means beating the best humans in practically every field, which is not what today's models do. Here is the definition, what is measured, and what is not.
Akash Mohapatra
Sep 30, 2026
Superintelligence regulation: what is law, what is proposed
No law anywhere regulates superintelligence. A 2026 MIRI paper proposes a US-China coalition capping training at 10^24 FLOP; the EU AI Act is the regime that actually binds you, with dates.
Akash Mohapatra
Sep 30, 2026
AGI vs ASI: the difference that decides what you build
Narrow AI, AGI and ASI are three different claims, and vendors blur them. Defined against real evaluations, with dated benchmark scores and the one that is a purchasing decision today.
Akash Mohapatra
Sep 30, 2026
Safe superintelligence: what four labs have actually shipped
Safe Superintelligence Inc. has shipped nothing since June 2024, while a division named after the same word ships an API. Here is what each lab claims, with dates, and what the primary sources actually say.
Akash Mohapatra
Sep 30, 2026
Gemini 4 Argon: benchmarks, price, and the guardrail split
Google shipped Gemini 4 Argon on 30 September 2026 with published benchmarks, introductory pricing that doubles later, and a build for vetted defenders that has the cyber guardrails removed.
Akash Mohapatra
Sep 30, 2026
One Vision. One Team. One Creuto.
Whatever your challenge, we bring the expertise to build, scale, and transform your digital presence with technology that works.