Signals for 2026-06-27

What happened after 2,000 people tried to hack my AI assistant

Simon Willison

What happened after 2,000 people tried to hack my AI assistant. Dit is relevant omdat agentwaarde steeds meer in workflowontwerp en taakafbakening zit, niet alleen in een slimmer model.

#agent #agentic-workflows #evals

OpenAI limits GPT-5.6 rollout after government request, says restrictions shouldn’t be the norm

TechCrunch AI

OpenAI limits GPT-5.6 rollout after government request, says restrictions shouldn’t be the norm. Dit is relevant omdat de builderlaag rond AI concreter wordt: tools, runtimes en ontwikkelworkflows bepalen steeds vaker de echte hefboom.

#builder #implementation #systems-framing #tooling-runtime

Ask, Don't Judge: Binary Questions for Interpretable LLM Evaluation and Self-Improvement

arXiv reasoning / agents / evals

Ask, Don't Judge: Binary Questions for Interpretable LLM Evaluation and Self-Improvement. Dit is relevant omdat serieuze AI-implementatie valt of staat met evaluatie, betrouwbaarheid en begrip van nieuwe failure modes.

#builder #evals #research-evals

Empowering GUI Agents via Autonomous Experience Exploration and Hindsight Experience Utilization for Task Planning

arXiv reasoning / agents / evals

Empowering GUI Agents via Autonomous Experience Exploration and Hindsight Experience Utilization for Task Planning. Dit is relevant omdat agentwaarde steeds meer in workflowontwerp en taakafbakening zit, niet alleen in een slimmer model.

#agent #agentic-workflows

Beyond Surface Forms: A Comprehensive, Mechanism-Oriented Taxonomy of Indirect Linguistic Encoding for LLM-Based Coded Language Detection

arXiv reasoning / agents / evals

Beyond Surface Forms: A Comprehensive, Mechanism-Oriented Taxonomy of Indirect Linguistic Encoding for LLM-Based Coded Language Detection. Dit is relevant omdat serieuze AI-implementatie valt of staat met evaluatie, betrouwbaarheid en begrip van nieuwe failure modes.

#evals #implementation #research-evals

OpenAI's GPT-5.6 Sol launches to rival Claude Mythos under government access rules it calls unsustainable

The Decoder

OpenAI's GPT-5.6 Sol launches to rival Claude Mythos under government access rules it calls unsustainable. Dit is relevant omdat het laat zien waar duurzame waarde in de AI-stack kan blijven hangen na de hype.

#evals #implementation #market-strategy

An AI model programmed nonstop for 19 days on a single MirrorCode task that cost $2,600 to run

The Decoder

An AI model programmed nonstop for 19 days on a single MirrorCode task that cost $2,600 to run. Dit is relevant omdat modelkeuze steeds meer een architectuurvraag wordt rond kosten, context, latency en controle.

#evals #models-architecture

OpenAI's GPT 5.6 rollout now requires US government approval on a "customer by customer basis"

The Decoder

OpenAI's GPT 5.6 rollout now requires US government approval on a "customer by customer basis". Dit is relevant omdat modelkeuze steeds meer een architectuurvraag wordt rond kosten, context, latency en controle.

#implementation #models-architecture

Databricks’ former AI chief thinks he can cut AI’s power bill by 1,000x

TechCrunch AI

Databricks’ former AI chief thinks he can cut AI’s power bill by 1,000x. Dit is relevant omdat de builderlaag rond AI concreter wordt: tools, runtimes en ontwikkelworkflows bepalen steeds vaker de echte hefboom.

#tooling-runtime

Adobe acquires image and video enhancement tool maker Topaz Labs

TechCrunch AI

Adobe acquires image and video enhancement tool maker Topaz Labs. Dit is relevant omdat de builderlaag rond AI concreter wordt: tools, runtimes en ontwikkelworkflows bepalen steeds vaker de echte hefboom.

#tooling-runtime