Tech Explainers

A do-nothing AI manager scores 26, not 0 — and one breach of trust caps everything. Inside Firmulate, the benchmark that distrusts perfect 100s.

• The New Handset Team
• September 19, 2026

Scientists have documented the first direct, real-time observation of a quantum jump in sound waves, marking a breakthrough in quantum acoustics research.

• The New Handset Team
• September 18, 2026

Search interest in AI skills training with Matt Pocock is spiking, driven by rising demand for practical AI knowledge amid growing industry focus.

• The New Handset Team
• September 17, 2026

University of Pennsylvania uses ChatGPT, Codex, and deep-learning models to cut early antimicrobial candidate discovery from years to hours, boosting fight against resistance.

• The New Handset Team
• September 16, 2026

Opus 4.8 had the deepest analysis and 80 learned rules — and still finished last. Diligence, it turns out, doesn’t equal impact.

• The New Handset Team
• September 13, 2026

Anthropic has announced a project to formalize Fermat’s Last Theorem, but details on scope, completion, and verification remain unavailable.

• The New Handset Team
• September 9, 2026

Liquid AI publicly releases a low-cost fine-tuning method for 350M models, boosting schema compliance on the IFStruct benchmark from 22.6% to 29.7% with minimal resources.

• The New Handset Team
• September 8, 2026

Four AI models ran the same fake company. All spotted the crisis — only the ones that read files two layers deep closed the €55,000 deal at full price.

• The New Handset Team
• September 5, 2026

Hugging Face releases @huggingface/kernels, a JavaScript library with 207 WebGPU kernels, and Fleet, a crowdsourced GPU benchmarking tool for browser AI.

• The New Handset Team
• September 3, 2026

Anthropic announces that automated AI systems can reliably mitigate alignment failures, raising hopes for scalable AI safety solutions but with uncertainties remaining.

• The New Handset Team
• September 3, 2026