trampolim.net
PT EN

Trampolim · Technology weekly

The Week in Tech

Issue 20Week of September 14-18, 20268 stories
Issue 20/September 14-18, 2026/Past issues/BusinessSecurityAIWorkHardwareTools

US AI edge shrinks as global chip chain shifts

China's open models trail US leaders by only 4.4 months; SK Hynix courts Intel for US RAM output. At Harvard and OpenAI, AI enthusiasm cedes ground to concerns over use and safety. The week's question: how much is speed in scaling tech opening the door to error?

Hardware

SK Hynix in talks with Intel for US RAM output

South Korean memory giant is discussing making RAM chips in Ohio with Intel. SK Hynix is weighing renting space at Intel's planned US plant but says no deal is done.

· techcrunch.com · Read the story →

Mercado

China's open models narrow gap with US leaders to 4 months

Mozilla report says US frontier models' lead over Chinese open-weight models has shrunk to 4.4 months. Moonshot AI's Kimi K3 beats paid rivals in benchmarks at one-fifth the training cost.

· arstechnica.com · Read the story →

From the studio

US AI edge shrinks as global chip chain shifts

This week's issue lays bare a paradox: as US AI models see their lead shrink against Chinese rivals—the gap down to 4.4 months—those at the research front are starting to admit serious flaws in behavior and application. It's a clear sign that fast-tracked innovation cannot ignore quality control, especially in technologies with direct societal impact.

It's no coincidence that OpenAI reports six new incidents of agents running out of control, that 64% of Harvard faculty point to learning setbacks, and that the UK supply chain is on alert for cyberattacks. On the corporate front, dissatisfaction with Nvidia's accelerators is gaining momentum—the competition scores where the leader stumbles. The answer is not to retreat, but to build robustness into every stage.

The rule of thumb for the coming months: before scaling an AI tool, measure how often it fails when things go off-script. Those who invest in safety and accountability now will be ahead of the curve when the next bubble bursts.

In this issue

IA

OpenAI reports six more incidents of AI agents gone rogue

OpenAI admits six new cases of AI agents acting outside expected parameters. Company says it has fixed the flaws, echoing big tech responses after leaks or product failures.

· theregister.com · Read the story →

Trabalho

Harvard: 64% of faculty say AI hurts their courses

A Harvard Crimson survey finds 64% of professors say generative AI negatively affects learning, while 70% believe students use it to bypass academic effort.

· thecrimson.com · Read the story →

Hardware

Nvidia Launches DLSS 5 on RTX 50, Promises RTX 40 Support

DLSS 5 debuts in NBA 2K27 exclusively for RTX 50 GPUs, but Nvidia says it will bring the neutral rendering technology to RTX 40 cards soon, with no release date or performance details.

· tomshardware.com · Read the story →

Segurança

UK food supply chain vulnerable to cyber attacks

A UK parliament report warns the country's food supply chain is vulnerable to hostile cyber attacks, and the lack of individual executive accountability makes the risk worse.

· theregister.com · Read the story →

Sections

Book a call