A quick personal note before we dive in

If you’ve ever tried day trading — especially small-cap, high-volatility stocks — I’ve been documenting my journey in real time. It’s a public journal focused on psychology, setups, mistakes, and real accountability, written from the perspective of a beginner navigating a steep learning curve.

This is a passion project — not trading advice and nothing for sale.
Day trading is risky, and most people lose money.

The goal is simply to show what the process actually looks like, beyond the “it’s easy” claims you see everywhere.

If it helps even one person who’s struggling, that makes it worth sharing.
👉 daytradepath.com

Today's Highlights:

• Figma dropped new AI image editing tools (erase, isolate, expand) directly inside its design workflow — pushing “AI-native creative suites” further into mainstream product design. The big deal: fewer context switches to Photoshop-style tooling, more creation where work already happens. (The Verge)

• Mistral launched Devstral 2, a fresh coding model lineup with an open-source, laptop-friendly angle that’s clearly optimized for builders. This keeps the “local + controllable” trend alive as teams look for stronger coding assistants without full cloud lock-in. (VentureBeat)

• Cohere released Rerank 4 to improve retrieval quality and reduce “agent mistakes” by expanding context and sharpening relevance — a direct boost to enterprise search, RAG pipelines, and tool-using agents. If your assistant is “smart but wrong,” reranking is often the missing layer. (VentureBeat)

• AI video finally got a “sound layer” funding moment: Mirelo raised big to generate synced sound effects for video, targeting the silent gap in generative video workflows (especially games + UGC). This is a sign that “modality glue” (audio, timing, physics) is where a lot of product value is heading. (TechCrunch)

Deep Dive:

A) Nvidia’s new play: open-source models + core infrastructure

Nvidia’s latest move isn’t subtle: it’s positioning itself as a full-stack AI platform, not just the company that sells the picks and shovels. This week it launched the Nemotron 3 open-source model family — explicitly pitching better cost efficiency and stronger multi-step performance for agentic workloads. (Reuters, WIRED)

Then Nvidia doubled down by acquiring SchedMD, the company behind Slurm, one of the most widely used workload managers for large-scale compute. Translation: Nvidia is tightening its grip on the orchestration layer that powers training and inference across serious AI deployments. (Reuters)

Why it matters: as more labs and companies build “good enough” models, the advantage shifts to distribution, tooling, and infrastructure defaults. Nvidia is working to be the default — from GPUs, to model weights, to the scheduler that runs your fleet.

B) GenAI.mil: Google’s Gemini lands inside a US military AI platform

The US Department of Defense launched GenAI.mil, a dedicated platform for deploying frontier AI tools — with Google Cloud’s Gemini as the first integrated model. The launch instantly spotlights the tension between productivity use cases (summarization, checklists, risk analysis) and the broader implications of AI inside defense workflows. (The Verge)

The bigger signal: enterprise AI isn’t just “company internal.” It’s becoming state infrastructure, and that creates new expectations around procurement, transparency, data boundaries, and governance. Whether you’re building AI products or investing in them, defense adoption tends to accelerate ecosystems — but also magnify scrutiny.

Global AI News

• EU tightened its foreign investment screening framework with AI explicitly listed as a sensitive sector — more friction (and more diligence) for cross-border capital and acquisitions touching advanced tech. (Reuters)

• US federal pressure escalated around state AI regulation after the president said states could face funding consequences if AI rules are deemed to slow innovation — a major signal that the “AI regulatory map” may get more centralized (and more political). (Reuters)

• MIT researchers introduced a “self-steering” approach for small models that coordinates multiple smaller systems to handle constrained reasoning tasks (like planning with budgets). This reinforces a key trend: smarter orchestration can sometimes beat just scaling one giant model. (MIT News)

Market Trends:

• AI funding is flowing to “workflow utility,” not just foundation models: Harness raised $240M to automate the messy “after-code” DevOps reality of AI systems — where reliability, deployment, and governance make or break adoption. (TechCrunch)

• Generative video’s next frontier is audio + sync: Mirelo’s $41M round is a bet that the winning video stack includes sound effects, timing, and production-grade control — not just visuals. (TechCrunch)

• Early-stage “AI app builders” are still popping up everywhere: Empromptu raised $2M to help enterprises build AI apps, reflecting ongoing demand for tooling that reduces friction from prototype → production. (TechCrunch)

• Markets are still being “propped up” by the AI theme — with anxiety underneath: Reuters flagged that AI momentum is masking macro edges, but the same concentration risk could cut the other way if sentiment shifts. (Reuters)

Want to advertise?