Good morning. It's Wednesday, April 29, and today we’re covering: MIT’s breakthrough in AI uncertainty, Samsung’s strike hitting memory supply, and OpenAI’s AWS push.

Know someone who'd love this? Forward it their way. First time reading? Sign up here to keep up with the future of tech.

OpenAI Expands Into AWS Enterprise Stack
Rather than pushing enterprises into new platforms, OpenAI is embedding its latest models directly inside Amazon Web Services environments. The partnership brings OpenAI models, Codex, and managed agents to Amazon Bedrock in limited preview. It includes access to GPT-5.5 and expands Codex, already used by over 4 million weekly users, across tools like VS Code and CLI. The move lets enterprises run AI within existing security, billing, and compliance systems, while applying usage toward AWS cloud commitments.

NVIDIA Unifies Modalities To Cut Agent Costs and Latency
Instead of scaling agent systems with more models, NVIDIA collapses vision, audio, and language into a single multimodal system. The company launches Nemotron 3 Nano Omni, an open model designed to power faster, more efficient enterprise AI agents across video, audio, images, and text. Built on a 30B-A3B hybrid MoE architecture with 256K context, it delivers up to 9x higher throughput while topping six multimodal benchmarks. Early adopters include Palantir and Foxconn.

Anthropic Embeds Claude Across Creative Software
Anthropic continues integrating Claude directly into creative tools through a new connector ecosystem. The company launches partnerships with platforms like Adobe, Blender, Autodesk, and Ableton to let Claude operate inside existing creative workflows. The connectors enable tasks like generating 3D models, automating batch edits, and searching royalty-free audio, while also introducing Claude Design for concept exploration and export to tools like Canva. The rollout also includes academic partnerships to shape AI-driven creative curricula.

Neurable’s Mind-Reading Tech Without Implants
Instead of requiring brain implants, Neurable is positioning “mind-reading” as a lightweight feature for everyday devices. The company plans to license its non-invasive BCI technology to consumer wearable makers across gaming, health, and productivity. Built on EEG sensors and AI signal processing, the system analyzes cognitive performance, with $35 million in Series A funding backing its commercialization push. Early partnerships include HP Inc.’s HyperX and iMotions, while the company emphasizes anonymization and HIPAA-aligned data protections as it scales.

nAnswer in tomorrow’s newsletter.

A viral post sparks debate on kids using AI, highlighting risks like bias, hallucinations, and mental health.

Researchers at MIT CSAIL introduced a new training method, Reinforcement Learning with Calibration Rewards (RLCR), that enables AI models to produce accurate confidence estimates alongside their answers. The approach adds a calibration term to standard reinforcement learning, penalizing mismatches between a model’s stated confidence and actual accuracy.

In tests on a 7-billion-parameter model, RLCR reduced calibration error by up to 90% while maintaining or improving performance across multiple benchmarks, including unseen datasets. The work addresses a core cause of hallucinations in reasoning models, where systems tend to express high confidence even when uncertain, and could improve reliability in high-stakes domains like medicine and finance. → Read the full article here.

An 18-day strike at Samsung Electronics’ Pyeongtaek and Hwaseong plants, expected to begin May 21, could cut global DRAM supply by 3–4% and NAND output by 2–3%, according to industry estimates. Early labor actions have already reduced wafer transfers and triggered production drops, including a reported 58.1% decline in foundry output and a 74.3% fall at one key line.

If the strike proceeds, Samsung could face losses of up to 30 trillion won ($20 billion) and a prolonged disruption, with recovery taking weeks after operations resume. The dispute centers on union demands for higher bonuses tied to AI-driven profits, mirroring large payouts at rival SK Hynix. With memory markets already strained by AI demand, the strike risks intensifying shortages and pushing prices higher for electronics manufacturers and consumers alike. → Read the full article here.

UK technology secretary Liz Kendall warned on April 28, 2026, that Britain must take greater control of artificial intelligence development or risk being shaped by forces largely outside its influence. She cited the growing dominance of US tech giants—including Amazon, Google, Meta, Microsoft, and Oracle—which now control about 70% of global AI computing capacity.

The UK government is responding with a new state-backed AI investment fund and plans to expand its role in chip design and manufacturing. The comments come as the country faces setbacks, including OpenAI pausing a major UK data center project and delays to a planned supercomputer. Kendall emphasized that strengthening domestic AI capabilities should complement—not replace—the UK’s close ties with the US, while rejecting calls to slow AI development. → Read the full article here.

Taylor Swift Files AI Voice Trademarks: She seeks protection for voice recordings and likeness using rare “sound marks,” testing trademark law as a tool against AI-generated imitations in music and media.

Mistral Targets AI Bottleneck: Rather than chasing better models, Mistral is going after the infrastructure gap stalling enterprise adoption with Workflows — a Temporal-powered orchestration layer for running multi-step AI systems.

AI Recreates Pompeii Victim’s Face: Researchers used skeletal data and artifacts to model a man killed in the AD 79 eruption, showing how artificial intelligence can deepen archaeological analysis and public engagement.

OpenAI Misses Growth Targets (Paywall): The company fell short on user and revenue goals, prompting CFO Sarah Friar and board members to question whether costly data-center commitments remain financially sustainable.

Red Hat Engineer Launches Safer OpenClaw Tool: Maintainer Sally O’Malley released Tank OS, a container-based system that isolates and manages AI agents, aiming to reduce security risks as enterprise adoption grows.

Lovable Launches Mobile AI App Builder: The startup released its vibe-coding app on iOS and Android, enabling voice- or text-driven app creation while complying with Apple rules by shifting generated app previews to the web.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team