One-Line Summary

Chat-based creative tools are entering everyday workflows while AI-tuned PC storage and an AI-solved classic math problem signal faster movement from novelty to utility.

Big Tech

AI reportedly solves an 80-year-old Erdős problem

The Wall Street Journal reports an AI system has solved a famous problem associated with mathematician Paul Erdős after 80 years, highlighting AI’s reach beyond text generation into advanced reasoning. While details are gated, the headline result underscores AI’s growing role in formal problem-solving. ¹

For business teams, this points to AI being used to search for patterns, test hypotheses, or draft structured arguments in complex domains where exhaustive human search is costly. Expect more R&D and knowledge work to probe where AI can complement experts on tightly scoped reasoning tasks. ¹

Industry & Biz

Silicon Motion unveils Gen5 DRAMless SSD controller for AI PCs

Silicon Motion, a major supplier of SSD controllers, announces the SM2524XT, a PCIe Gen5 x4 DRAMless SSD controller aimed at AI PCs and on-device inference. It targets workloads like key-value cache with sequential reads up to 14 GB/s and up to 2.5 million IOPS random performance, built on TSMC’s 6nm process with up to 25% better performance per watt than the prior generation. The controller supports NAND interface speeds up to 4800 MT/s and integrates Separated Command Address, advanced FTL scheduling, and NANDXtend LDPC ECC. (Announced May 29, 2026.) ²

Why it matters: as local agents and on-device models become common, random I/O and latency stability can bottleneck responsiveness. Silicon Motion positions the SM2524XT to sustain high random throughput under thermal and power constraints, a sign that “AI PC” value will hinge as much on storage architecture as on CPU/GPU branding. ²

New Tools

Adobe Firefly Assistant offers chat-based edits across Adobe apps

Adobe’s Firefly AI Assistant is a chatbot that performs multi-step edits by invoking tools from apps like Photoshop and Illustrator when you describe what you want. In The Verge’s beta test, the assistant explains the scene, details which tools it plans to use, and narrates steps as it executes them, turning AI edits into a teachable, conversational workflow. ³

In practice, it can adjust lighting, change backgrounds, and add objects, and it enforces guardrails by declining illegal requests or explicit body alterations. The Verge notes it sometimes allowed additions like “hand-rolled cigarettes” and guns in photos but refused to generate anything outright illegal, reflecting evolving content policies and safety filters. ³

Output quality is mixed: many edits look convincing at a glance, but color can skew vivid and blends can be imperfect, making results resemble novice-level work rather than a professional editor’s. The assistant is transparent about limitations, proposes alternatives, and can self-correct mid-task, but still requires human oversight. ³

Compared with Canva’s conversational agent, The Verge says Adobe’s stands out by clearly explaining its process, which can help users learn editing basics while delegating busywork. Adobe pitches Firefly as a time-saver for creatives, but pros may feel it’s more like supervising a junior intern until quality improves. ³

What This Means for You

Creative and marketing teams can treat Firefly as a quick draft generator and tutor: use it for first-pass crops, lighting tweaks, and variant exploration, then finalize in your usual tools. The value comes from faster iteration and its step-by-step explanations, but plan for human QA before publishing. ³

Set guardrails and a review checklist for AI-edited visuals (brand colors, edges/halos, skin tones, typography alignment). Assign a human approver and document when the assistant declines a request so you can adjust prompts or switch to manual edits. ³

If you are evaluating “AI PCs,” look beyond CPU/GPU labels: random I/O for key-value caches and latency stability will shape how responsive local agents feel. Ask vendors for NVMe specs (e.g., IOPS, PCIe Gen5) and test with your actual on-device workloads when possible; controllers like SM2524XT emphasize this shift. ²

The WSJ headline suggests AI is stepping into harder reasoning domains. For knowledge work, try using AI to draft structured arguments or explore alternative approaches on a bounded problem, then have a domain expert review the logic. ¹

Action Items

1. Trial Firefly Assistant on a real asset: If you have Creative Cloud or Firefly (beta) access, upload one photo and ask for a multi-step edit (crop, relight, background change), then compare with a manual edit.
2. Create a visual QA checklist: Define 6–8 checks (brand color, haloing, shadows, text kerning) and require sign-off before any AI-edited creative ships.
3. Ask vendors for storage IOPS, not just capacity: When buying or refreshing laptops/desktops, request random read/write IOPS and PCIe generation details; prioritize higher IOPS for local agents.
4. Run a reasoning drill: Give your AI tool a small, real decision (e.g., promo mechanics) and request step-by-step reasoning; compare with your team’s approach and note gaps.