In partnership with
Stop Paying for 10 Tools. One AI Does It All.
Most e-commerce sellers are running their store across 6 to 10 separate tools — and spending more time managing software than growing their business. StoreClaw replaces your entire stack with one autonomous AI engine that monitors competitors, optimizes listings, automates marketing, and tracks real profit across Shopify, Amazon, and beyond.
It doesn't wait for you to ask. It runs 24/7 in the background, so you wake up to a full dashboard instead of a list of things you forgot to check.
Connect your store, and StoreClaw gets to work — no prompts, no complex setup, no six-app stack.
Free to start. No credit card required.
Beginners in AI
Good morning and thank you for joining us again!
Welcome to this daily edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated and edited, and published with the intention of making AI news and technology more accessible to everyone.
THE FRONT PAGE
Nvidia Took an AI From 30% to 100% Without Changing the Model
TLDR: Nvidia took a model that scores about 30% on a hard reasoning test, wrapped it in a system called AVO, and got a perfect score without touching the model.
The Story:
Nvidia published results Friday showing its AVO agent system finished all 183 levels of the ARC-AGI-3 benchmark across 25 game environments. That test drops an AI into games it has never seen, with no rules, no instructions, and no stated goal. It has to poke around and work out what's going on. The model doing the thinking was Claude Opus 5, which ARC Prize scores at roughly 30% when it runs on its own. AVO cleared everything in 6,624 moves, about 12% fewer than a rival setup called VISTA using the same model. Here's the odd part: Nvidia didn't build AVO for games. It built it to tune the low-level code that runs on graphics chips, where it ran for seven straight days, tried more than 500 ideas, and produced code that beat FlashAttention-4 by up to 10.5%. Swapping it over to the game test meant changing the tools, not the agent.
Its Significance:
Most AI news is about which model is smartest. This says the wrapper around the model counts for just as much. AVO gives the model two things a chatbot doesn't get: memory that survives between sessions, so it stops re-solving the same problem, and a supervisor watching from above that nudges it when it's spinning in circles. Add those and the same model goes from failing most of the test to finishing all of it. Nvidia is careful about what this proves. The 100% is on the public set, not the harder private one, and the two systems weren't tested under matching conditions, so the 30-to-100 difference isn't a clean measurement. Their own line at the end says it best: "The model matters, but the model is not the entire agent." Basically, this means that there are ways to get more capability out of the AI models that we already have.
QUICK TAKES
The story: Researchers at the University of Konstanz put groups of AI agents in a room, so to speak, and watched them settle on a shared choice with no rules and no leader. Weaker models fell apart at around 30 agents, while stronger ones stayed coordinated at 1,000, the biggest group tested.
Your takeaway: People top out around 150 to 200 before we need rules and bosses to keep a group together. Software doesn't have that ceiling in the same way. The less flattering truth is buried in the findings: a large group of agents can agree strongly and still be wrong, all at once, in the same direction. Mass delusions are not limited to humans.
The story: Duke researchers say better chips and better algorithms have shrunk jobs that used to take weeks into a few days. Computational biologist Rohit Singh runs models across millions of gene readings to spot what separates a healthy cell from a cancer cell, then asks whether a drug could close that gap.
Your takeaway: Singh describes these models as "microscopes" for biology, which is a better frame than most. They don't answer your question. They let you see something you couldn't see before, and then you ask a new question. Duke is also building a small GPU center for 2027 designed to hold down power and water use, which is a counterpoint to the data center fights breaking out elsewhere.
The story: A reader letter in a Minnesota paper argues China is pulling ahead on AI and factory robots. It cites a 900-yard truck assembly line with no people on it, phone plants running with the lights off, and a state media report of a rail station built in 38 months by a robot crew.
Your takeaway: This is opinion, not reporting, and the figures come secondhand from a 2025 Telegraph article and Chinese state press. Read it for what it signals instead. Arguments about robot factories are turning up in small-town letters pages now. That's a different stage of a debate than white papers and hearings, and is a growing concern around the country for its implications.
TOOLS ON OUR RADAR
🚀 Dashy Free and Open Source: A highly customizable homepage that allows you to organize all of your favorite websites and personal applications into one beautiful and easy to navigate digital dashboard.(ask your AI to wire it up for you)
🔍 SearXNG Free and Open Source: A privacy respecting search engine that aggregates results from many different sources while keeping your search history and personal data completely anonymous.
🔔 Ntfy Free and Open Source: A simple service that allows you to send push notifications directly to your phone or desktop from any application without needing a complex setup process.
🎵 Navidrome Free and Open Source: A fast and lightweight music server that lets you listen to your own personal audio collection from any device using any compatible music player.
TRENDING
Anthropic's IPO filing will list AI backlash as a risk - Sources told CNBC that public anger at AI and data centers will appear in the risk section of Anthropic's prospectus, expected within weeks. A May Gallup poll found seven in 10 Americans oppose a data center being built near them, and the company's growth depends on building more of them.
A third of web pages published since ChatGPT launched show signs of AI writing - Pew ran nearly half a million English pages through a detection tool and found AI fingerprints on 35% of everything published after November 2022. Commercial .com sites showed it at roughly 10 times the rate of .edu and .gov, which both sat near 1%.
Anthropic brings its most cyber-capable model to more security teams - Claude Mythos 5 now runs vulnerability scans inside Claude Security for enterprise customers, and Anthropic is putting $35 million in credits behind a fund for patching open source software. Customers get the patches and alerts, not direct access to the model, which is the whole point of the design.
OpenAI chases Anthropic's business customers with a zero data retention pledge - OpenAI previewed a system it calls Private Safety Processing that looks for misuse across a run of related requests without staff reading the content underneath. It's aimed straight at companies unhappy with Anthropic's 30-day retention rule for its top models.
Slack Code puts AI coding agents into the group chat - Tag a coding agent in Slack and it spins up its own channel where the whole team watches the work happen, reviews the changes, and signs off before anything ships. Claude Code, Devin, GitHub Copilot, ChatGPT, and Vercel's agent all launched with it, free on any Slack plan, though you still pay for the agents themselves.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🌇 Name a city and the kind of spot you want. Get a real place to watch tonight's sunset, timed to when the sky there actually turns.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Live Rooftop & Sunset Spot Finder — name a city, get a real sunset-viewing spot via live web search, timed to tonight's actual sunset with live cloud-cover context.
Aesthetic: dusk purple-black (#160f18), coral-pink (#e0637a) primary with a pink glow top-left, violet (#7a5ce0) secondary for the practical-note card with a glow bottom-right. Fraunces serif for headings/sunset time/spot name, Inter for body, JetBrains Mono for labels.
Form: city text input, spot-type dropdown (rooftop bar / park-viewpoint / restaurant with a view / waterfront-bridge).
Technical:
1. Geocode the city via geocoding-api.open-meteo.com/v1/search?name={city}&count=1 for lat/lon/timezone.
2. In parallel: call the Anthropic Messages API WITH tools:[{type:'web_search_20250305', name:'web_search'}] (works automatically), system prompt as someone finding ONE real specific sunset-viewing place matching the city/type, grounded in search, never inventing a place without evidence — if unconfident, found:false. Return raw JSON: found, spot_name, type, neighborhood, why_it_fits (2-3 sentences on orientation/elevation/view), practical_note (optional), backup (real alternate: spot_name + why).
3. Fetch the real sunset time via api.sunrise-sunset.org/json?lat={lat}&lng={lon}&date={today}&formatted=0, formatted in the location's own timezone via Intl.DateTimeFormat.
4. Fetch live cloud cover via api.open-meteo.com/v1/forecast?latitude={lat}&longitude={lon}¤t=cloud_cover&timezone=auto, mapped to a plain-language read (clear=vivid, partly cloudy=dramatic, overcast=muted).
Client-side compute: arrival time = sunset time minus 35 minutes, displayed as "arrive by X to grab a spot."
Render: a gradient "tonight's sunset" hero card with the real sunset time, the arrive-by note, and the cloud-cover read. A "go here" spot card (name, type/neighborhood meta, why-it-fits). An optional violet practical-note card. A backup-spot card. A "Open in Google Maps" link built client-side (no API key, just a maps search URL). On a found:false response, surface a clear error asking the person to adjust city or spot type.What this does: Enter a city and a spot type, rooftop bar, park, restaurant, waterfront, and it searches the live web for one real place known for sunset views, west-facing or elevated, never an invented placeholder. Alongside that, it pulls tonight's real sunset time for that exact location and tells you when to arrive, about 35 minutes early, to grab a spot before the light builds. Real live cloud cover shapes an honest read on what tonight's sky will actually look like, vivid if it's clear, dramatic if it's partly cloudy, muted if it's overcast, so you know what you're walking into. A real backup spot and a one-tap Maps link round it out.
What this looks like:
Same sales team. More booked meetings
Aimfox Avatars allows you to rent dedicated LinkedIn profiles that start conversations and hand interested replies to your existing sellers. Expand outbound capacity without adding another SDR seat.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Finish all 183 levels of a reasoning test it was given no rules for, once you attach memory and a supervisor to the model.
❌ Still Can't: Show that the same score holds on the harder private test set. Nvidia flags this limit itself.
✅ AI Can Now: Keep a group of up to 1,000 agents coordinated on one shared answer with no leader telling them what to pick.
❌ Still Can't: Notice when that shared answer is wrong. Strong agreement and being wrong sit together comfortably.
FROM THE WEB
RECOMMENDED LISTENING/READING/WATCHING
Roderick, or The Education of a Young Machine by John Sladek - Book
A 1980 satirical novel, nominated for the Philip K. Dick Award, about a robot named Roderick learning what it means to be human by moving through a thoroughly absurd America of cults, consultants, bureaucrats, and academic infighting. Sladek's joke is that Roderick reasons more carefully than almost anyone he meets, and the humans are the ones running on autopilot. The sequel came in 1983 and both were reissued together as The Complete Roderick.
Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!
-James
By the way, this is the link if you liked the content and want to share with a friend.
Some * designated product links may be affiliate or referral links. As an Amazon Associate, I earn from qualifying purchases. This helps support the newsletter at no extra cost to you and Amazon makes a tiny hair less.
