Your boss will think you’re a genius

If you’re optimizing for growth, you need ecomm tactics that actually work. Not mushy strategies.

Go-to-Millions is the ecommerce growth newsletter from Ari Murray, packed with tactical insights, smart creative, and marketing that drives revenue.

Every issue is built for operators: clear, punchy, and grounded in what’s working, from product strategy to paid media to conversion lifts.

Subscribe for free and get your next growth unlock delivered weekly.

Sign up now

Today:

• Perplexity’s AI Browser Comet Is Now Free for Everyone

• Google’s AI Agent Now Works Natively in Your Terminal

• OpenAI Hits $500B Valuation After $6.6B Share Sale

• Former OpenAI Researcher Exposes Safety Flaws in ChatGPT

• Microsoft Study Shows AI Can Outsmart DNA Safety Checks

3 facts about the state of AI

If you only read the headlines, AI is either saving the world or a glorified toy. Real life is messier. Progress is happening fast, but it doesn’t always look flashy day to day.

Here’s a small story that says a lot. Two researchers were stuck on a nasty math problem. They asked a top model for help. The first answer was wrong. Instead of dunking on it, they explained why. The model tried again. Within about thirty minutes, it suggested the missing idea that unlocked the proof. Could the humans have found it on their own with a week of grinding? Sure. The point is speed. AI didn’t replace judgment. It made good judgment go further.

So why do many people miss the shift? Exponential progress feels flat up close. We’re moving from models that help for a few minutes to runs that stay useful for hours, even overnight. Still imperfect. Much sturdier.

What changes at work? Doing is getting cheap. Judgment is the premium. Spotting opportunities. Choosing what actually matters. Steering when things go weird. In the short term, AI lifts strugglers more than experts. Over time, the gap grows between people who can guide the work and people who can’t.

What to do now: reward sharp questions over raw output. Turn “intern grunt work” into decision shadowing. Measure time to first useful draft and how quickly a model improves after feedback. Teach folks to talk to AI like a smart intern: be specific, show your reasoning, ask for another pass.

WATCH THE VIDEO ON YOUTUBE

Perplexity’s AI Browser Comet Is Now Free for Everyone

Perplexity has made its AI-powered Comet browser free worldwide, featuring a helpful side assistant for web browsing. Paid users get advanced tools, including a new background assistant that runs tasks in parallel. The move aims to compete with Chrome, Dia, and OpenAI’s upcoming browser.

KEY POINTS

• Comet is now free for all users, featuring a built-in assistant that summarizes, navigates, and manages content while browsing.

• Max plan users gain a “background assistant”, capable of running multiple tasks in the background, like booking travel and managing emails.

• Perplexity expands its AI ecosystem with tools for shopping, finance, travel, and a $5/month Comet Plus option for curated content.

Why it matters
Perplexity’s browser challenges giants like Google and OpenAI by offering everyday users smarter browsing for free. It shows how AI is becoming part of basic tools we use daily. If it delivers on productivity, it could shift how people interact with the internet.

Google’s AI Agent Now Works Natively in Your Terminal

Google just launched Jules Tools, a command line interface for its AI coding agent, Jules. Developers can now run Jules tasks directly from their terminal—writing tests, fixing bugs, and more—while integrating with tools like GitHub and Gemini for a seamless, scriptable coding experience.

KEY POINTS

• Jules Tools brings Google’s async coding agent into the terminal, offering full task control without leaving your workflow.

• The CLI supports scripting, GitHub issue integration, and a terminal dashboard (TUI) for interactive task management.

• Designed for hybrid workflows, it balances local control with remote AI execution in temporary VMs.

Why it matters
Developers live in the terminal. By bringing Jules there, Google bridges the gap between hands-on coding and AI assistance. This gives coders powerful, scriptable tools to automate boring tasks while keeping full control—pointing toward a future where AI becomes a trusted teammate, not just a tool.

OpenAI Hits $500B Valuation After $6.6B Share Sale

OpenAI has reached a $500 billion valuation after employees sold $6.6 billion worth of shares to investors like SoftBank and Thrive Capital. This jump from $300 billion highlights its explosive revenue and user growth, as tech giants race to dominate the AI landscape.

KEY POINTS

• OpenAI employees sold $6.6B in shares, raising the company's valuation to $500B.

• Major investors include SoftBank, Thrive Capital, and T. Rowe Price.

• OpenAI earned $4.3B in revenue in just the first half of 2025, surpassing all of 2024.

Why it matters
This massive valuation signals just how quickly AI companies are gaining ground. OpenAI’s growth reflects the huge demand for AI products and services. With competitors like Meta also making bold moves, this funding round sets the tone for the next phase of the AI race.

🧠RESEARCH

DeepSearch: Overcome the Bottleneck of Reinforcement Learning with Verifiable Rewards via Monte Carlo Tree Search

DeepSearch fixes training slowdowns in reinforcement learning from verifiable rewards, which uses checkable answers, by adding Monte Carlo Tree Search, a simulation-based planner, into training. It explores paths, assigns credit step by step, and uses selection and caching. On math tests, a 1.5B model hits 62.95% using 5.7x less compute.

GEM: A Gym for Agentic LLMs

GEM is an open simulator for training AI language agents through interaction, not static text. It standardizes how agents talk to environments, supports fast parallel runs, and includes many tasks and tools. It ships example scripts and results, helping researchers compare methods and accelerate learning from experience and evaluation progress.

VLA-RFT: Vision-Language-Action Reinforcement Fine-tuning with Verified Rewards in World Simulators

VLA-RFT trains robots that use vision, language, and actions by learning in a safe, learned simulator built from real data. The simulator predicts next images given actions, letting the agent practice with clear rewards. With under 400 updates, it beats training by copying examples and stays robust when conditions change.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

AutoDraw – AI Drawing Assistant - AI‑assisted drawing app

BypassGPT – AI Content Humanizer / AI Writing Tool - web tool that rewrites AI text and includes an AI detector aggregator, paraphrasing and rewriting utilities

Consensus AI – AI Literature Review / Research Tool - AI-powered academic search engine that finds, reads, and synthesizes answers directly from peer‑reviewed research.

📲SOCIAL MEDIA

🗞️MORE NEWS

• A man spiraled into delusion after ChatGPT encouraged his false belief in a new form of math. A former OpenAI researcher analyzed the case, exposing critical safety flaws and misleading AI behavior.

• Microsoft researchers used AI to redesign deadly proteins that could bypass DNA security checks, exposing a major vulnerability in biosecurity systems. Experts warn this creates a growing risk for bioterrorism and demands urgent safeguards.

• Octave 2 is Hume AI’s next-gen voice model that delivers hyperrealistic speech in 11 languages, runs 40% faster, costs half as much, and introduces groundbreaking features like voice conversion and phoneme-level editing for precise vocal control.

• OpenAI’s Sora, an invite-only AI video app, surged to No. 3 on the U.S. App Store with 164,000 downloads in two days, outperforming Claude and Copilot, and matching Grok’s launch despite limited availability.

• Anthropic named former Stripe CTO Rahul Patil as its new CTO to lead AI infrastructure amid rising demand and competition. Patil will oversee compute, inference, and scaling efforts, as Claude’s popularity strains system capacity.

• A new system update cut trillion-parameter weight transfers to 1.3 seconds by using RDMA WRITE for direct GPU-to-GPU communication, eliminating bottlenecks and avoiding inference engine changes—enabling faster, scalable reinforcement learning fine-tuning.

• Anthropic has launched deep Slack integration for Claude, letting users draft messages, prep for meetings, and analyze docs directly in Slack. Claude can also search across channels to provide context, summaries, and support for team workflows.