In partnership with
Free email without sacrificing your privacy
Gmail is free, but you pay with your data. Proton Mail is different.
We don’t scan your messages. We don’t sell your behavior. We don’t follow you across the internet.
Proton Mail gives you full-featured, private email without surveillance or creepy profiling. It’s email that respects your time, your attention, and your boundaries.
Email doesn’t have to cost your privacy.
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
OpenAI’s ChatGPT Cuts Hallucinations by 27% with new GPT5.3 Model
TLDR: OpenAI released GPT-5.3 Instant, a ChatGPT update that cuts hallucinations by up to 27%, stops refusing safe questions, and tries to sound less preachy.
The Story:
OpenAI launched GPT-5.3 Instant on Tuesday, replacing the default model that powers ChatGPT for hundreds of millions of users. The update targets a problem the company openly acknowledged: GPT-5.2 Instant was, in their own words, "cringe." It would make assumptions about your emotions, refuse to answer perfectly safe questions, and say things like "Stop. Take a breath." when nobody asked for therapy. GPT-5.3 cuts hallucinations (when the AI makes stuff up) by 26.8% when using web search and 19.7% when relying on its own knowledge. It also does a better job pulling useful answers from the web instead of dumping a list of loosely related links. The previous model, GPT-5.2 Instant, will stay available to paid users until June 3, 2026. And OpenAI is already teasing what's next, posting "5.4 sooner than you think" on X shortly after the launch.
Its Significance:
Hallucinations are not relegated to just one company. If you use ChatGPT, this update hits your daily experience directly. Fewer made-up answers means you can trust what it tells you a bit more, whether you're researching something for work or checking facts for school. The reduction in unnecessary refusals is a big deal too. Lots of people have complained about ChatGPT acting like a hall monitor, refusing to answer basic questions because it thought they might be risky. ChatGPT, Claude, and Gemini are all have mid to high level moderation built in, unlike Elon Musk’s Grok which has lower moderation. Higher moderation will be less likely to answer safe questions that appear unsafe to the AI.
QUICK TAKES
The story: Researchers at Binghamton University and startup Cauth AI built a tool called My Music My Choice (MMMC) that protects songs from AI voice cloning. It adds tiny, invisible changes to a song's audio that sound normal to human ears but turn into distorted noise when an AI tries to copy the voice.
Your takeaway: AI can now clone any singer's voice with just a few seconds of audio, and fake songs are flooding the internet. This tool gives artists a way to fight back before their voices get hijacked. The research was presented at the NeurIPS 2025 AI for Music workshop.
The story: Google released Gemini 3.1 Flash-Lite, a new AI model built for speed and low cost. Priced at $0.25 per million input tokens, it's 2.5 times faster than the previous Gemini 2.5 Flash and 45% faster at generating output, while still outperforming it on reasoning tests.
Your takeaway: This isn't for everyday ChatGPT-style chatting. It's for developers building apps that make millions of AI calls per day, like translation services or content moderation. The price war between Google, OpenAI, and Anthropic keeps pushing costs down, which eventually means cheaper AI tools for everyone.
The story: The U.S. Supreme Court refused to take up the question of whether AI-created art can be copyrighted. The case involved computer scientist Stephen Thaler, who wanted copyright protection for an image his AI system "DABUS" made entirely on its own. Lower courts said no, and the Supreme Court let those rulings stand.
Your takeaway: This settles it for now: if an AI creates something without a human guiding the process, it can't be copyrighted in the U.S. But if you use AI as a tool and direct the creative process yourself, you may still be able to get protection. This matters for anyone using AI to create art, music, or written content for their business.
TOOLS ON OUR RADAR
🐧 Immich Free and Open Source: A highly performant self hosted photo and video backup application that seamlessly syncs media from your mobile phone ensuring your personal memories remain completely private. (Alternative to Google Photos)
📝 Tally Freemium: An incredibly intuitive form builder that works exactly like a simple text document allowing you to create beautiful custom surveys and collect user data without knowing how to code.
Andi Free: A privacy focused conversational search engine that actually reads and summarizes the best web pages to give you direct answers instead of a long list of blue links. (Category: Conversational Search)
📌 Milanote Freemium: A highly flexible visual workspace designed specifically for creative professionals to organize their mood boards project briefs and design inspiration in one beautiful place.
TRENDING
Grok 4.20 Beta 2 Gets an Update - xAI pushed out a Beta 2 update to its Grok 4.20 model, improving instruction following, reducing capability hallucinations, and fixing image search and rendering reliability.
Lunit to Present 21 AI Imaging Studies at ECR 2026 - South Korean medical AI company Lunit will showcase 21 studies on AI-powered breast cancer and lung disease detection at Europe's biggest radiology conference, building on research showing its AI can match or replace a human radiologist in screening.
OpenAI Will Let Users Add Trusted Contacts for Mental Health Alerts - Facing multiple wrongful death lawsuits, OpenAI announced a new feature that lets adult ChatGPT users name a trusted person to be notified if the system detects signs of a mental health crisis during a conversation.
Michael Pollan's New Book Explores Whether AI Can Be Conscious - In "A World Appears," bestselling author Michael Pollan explores what consciousness really is, visiting scientists trying to engineer feelings into AI and questioning whether machines could ever truly experience awareness.
Harvard Professor Says AI Users Are Losing Cognitive Abilities - Harvard astronomer Avi Loeb wrote that he's watching people around him lose thinking skills from overusing AI chatbots, comparing it to losing muscle from never walking. Research backs him up: a 2025 study found frequent AI use can cause critical thinking to weaken over time.
Xiaomi Deploys Humanoid Robot at Its EV Factory - Chinese tech giant Xiaomi put a humanoid robot to work on its electric vehicle assembly line, where it ran autonomously for 3 straight hours installing nuts with a 90.2% success rate. Xiaomi's CEO predicted the company will have large numbers of humanoid robots working in its factories within five years.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
🙏 Build a gratitude journal with mood tracking, three-item daily entries, and a past entries history view
Build me a single-file HTML app called "Gratitude Journal" using React 18
(CDN) + Babel standal
#f0a840 accent). No external CSS or imports.
Include
- Today's Entry tab: mood selector (5 emoji buttons), 3 grateful-for text
inputs, optional reflection textarea, and a S
- Past Entries tab: scrollable list of saved entries
the 3 items, and any reflection note
- Stats panel:
- Journal Prompt card with rotating prompts and a "New prompt" button
- Search Best Practices button cycling through 4 gratitude journaling tips
Use only React.useState, i
Make it fully functional in a single HTML file.What this does:
Gives you a simple daily gratitude practice with just enough structure. Pick your mood, write three things you’re grateful for, and add an optional reflection. Past entries are saved in a scrollable history so you can look back. The journal prompt rotates so you never stare at a blank page wondering what to write.
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Protect a musician's voice from being cloned by adding invisible changes to audio files that break AI copycats.
❌ Still Can't: Receive copyright protection in the U.S. for anything it creates without human involvement.
✅ AI Can Now: Detect signs of breast cancer up to six years before a formal diagnosis by analyzing routine mammograms.
❌ Still Can't: Reliably detect when a user is heading into a mental health crisis any more than search engines can.
FROM THE WEB
Launches Tuesday, March 10th
RECOMMENDED LISTENING/READING/WATCHING
A two-player card game where one person plays a megacorporation trying to push through secret agendas, and the other plays a hacker trying to steal them before they can. The asymmetry is the whole point: the corp builds walls and traps on the network while the runner probes for weaknesses, and both sides are playing a completely different game with different cards, mechanics, and win conditions. Originally designed by Richard Garfield in 1996 and revived by Fantasy Flight Games in 2012, it remains one of the most elegantly designed games about AI, surveillance capitalism, and corporate power ever made.
“AI is Going to Fundamentally Change…Everything”
That’s what NVIDIA’s CEO said, calling AI “the largest infrastructure buildout in history.” Their chips helped make it happen. Now they’re collaborating with Miso Robotics for key robotics advances. Miso’s restaurant-kitchen-AI robots logged 200K+ hours for brands like White Castle. And NVIDIA helps unlock up to 35% faster performance. 100k+ US fast-food locations are in need, a $4B/year opportunity for Miso.
This is a paid advertisement for Miso Robotics’ Regulation A offering. Please read the offering circular at invest.misorobotics.com.
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.
