- AI Fire
- Posts
- 🧠 GPT-5.3 Instant Goes Live!
ChatGPT was getting a little weird and awkward. So OpenAI’s "un-cringing" update: How GPT-5.3 Instant is cutting the lectures and getting straight to the point!
What's on FIRE 🔥
IN PARTNERSHIP WITH RUBRIK
Attackers are no longer just locking files; they are weaponizing compromised identities to dismantle security from the inside out.
Join us virtually on March 10th for a critical discussion on the convergence of cyber threats.
AI INSIGHTS
For a long time, the biggest complaint about ChatGPT wasn’t speed or intelligence. It was the tone. Too many weird lines like “Stop. Take a breath.”
→ So OpenAI just released GPT‑5.3 Instant, an update designed to make ChatGPT sound more natural and less awkward. Instead of long disclaimers or moralizing preambles, the model now tries to:
answer more directly & reduce unnecessary caveats
avoid overly dramatic language & keep conversations naturally
It also reduces hallucinations, it tries to surface the key information immediately.
26.8% fewer hallucinations when using web results
19.7% improvement in reliability with internal knowledge
22.5% fewer hallucinations based on user feedback
Another major change: fewer unnecessary refusals. OpenAI says this change came directly from user feedback. The previous model, GPT-5.2 Instant, will remain available until June 3, when it will be retired. And GPT-5.4….
PRESENTED BY MASTERWORKS
Wall Street Just Named the Most Crowded Trades of 2026
AI stocks. Metals. Crypto.
Surprise, surprise; gold crashed 16%. Silver plunged 34%. Bitcoin dropped to 1 year lows.
All supposedly "uncorrelated" assets moving in lockstep largely because of overleveraged margin.
JPM strategists warn that the same leverage is still a risk.
Those markets may be recovering now, but cascading liquidations could trigger quickly across several asset classes simultaneously.
So much for diversifying away risk, right?
But get this–
70,819 everyday investors have allocated $1.3 billion fractionally across 500+ exclusive investments.
Not real estate or PE… Blue-chip art. Sounds crazy, right?
Now it’s easy to invest in art featuring legends like Banksy, Basquiat, and Picasso, thanks to Masterworks.
They do the heavy lifting from acquisition to sale, so you can diversify with the strategy typically limited to the ultra-wealthy.
(Past sales delivered net returns like 14.6%, 17.6%, and 17.8% on works held longer than a year.)*
*Investing involves risk. Past performance is not indicative of future returns. Important Reg A disclosures: masterworks.com/cd
AI SOURCES FROM AI FIRE
1. Google Antigravity Mastery Guide: Learn 100 Hours of Agent Workflows in Under 1 Hour. The exact workflow to stay coherent, cut costs, and ship faster without writing messy prompts.
2. 7 Secret Chrome Extensions for Power Users to Make NotebookLM 10x Faster for Research! If you rely on NotebookLM for research, you’re probably missing key shortcuts. These 7 extensions remove friction and cut your workflow time in half.
3. Copy This Setup: Google AI Studio + Supabase to Bring Any Tough Ideas to Real Life. From zero to a working app with authentication, database, and file uploads. You’ll see the full flow, the hidden traps, and the shortcuts most people miss.
AI AGENTS & VIBE-CODING LIVE WORKSHOP
Live Session • 9:30 PM (21:30) EST • Mar 4, 2026 (Today)
Today we (AI Fire) are hosting a live workshop where we will build a real AI system from scratch. Imagine turning an idea into a real AI tool that can store data, remember users, and run tasks automatically. No coding!
In this first live workshop, you’ll learn how to build a real web app and a custom AI agent using powerful tools like Google AI Studio, Supabase, and n8n.
What You’ll Experience:
💡 Turn an idea into a real AI tool (live demo)
🌐 Build a real website using a free Google tool
🤖 Create your own AI agent with n8n
⚡ See how these AI builders work together
No technical background. No complexity. Seats are limited to keep it interactive.
🎯 Join the workshop and learn how to build real AI tools.
TODAY IN AI
AI HIGHLIGHTS
🎙️ Anthropic rolled out Voice Mode for Claude Code. It lets you speak commands instead of typing. Say something and it'll write the code. See how to use it here.
🔻 A new test asked ChatGPT, Gemini, and Claude about the Iran war. One AI hallucinated details, another stayed cautious. Here’s the result & what it revealed.
⚠️ X will suspend anyone for 90 days if they post AI war videos without disclosure. And they’ll be permanently out of its revenue-sharing program. Be careful, guys!
👀 Did OpenAI just accidentally confirm GPT-5.4? Some devs spotted GPT-5.4 inside OpenAI Codex code then the references disappeared hours later. "Oops" leak!?
⚡ Google introduced Gemini 3.1 Flash-Lite in preview - a cheaper, faster version. It’s 45% faster than Gemini Flash and can handle 1M-token prompts. See model card.
🤔 Right after Alibaba launched new Qwen 3.5 models, the project’s key tech lead suddenly stepped down. Colleagues say the move feels like “the end of an era.”
💰 Big AI Fundraising: Ayar Labs raised $500M at a $3.75B valuation. Backed by Nvidia & Advanced Micro Devices, it uses light to connect AI chips faster.
NEW EMPOWERED AI TOOLS
🎧 Krisp Conversion understands accented speech in real time. Fully on-device with near-zero latency & works across Zoom, Teams, Meet.
🧠 Qwen3.5 Small is a native multimodal with improved architecture and scaled RL. 0.8B, and 2B are tiny and fast for edge devices.
💬 getviktor.com lives in Slack, connects to 3,000+ tools across your entire stack, and acts all on its own.
🧬 Deep Personality offers science-backed personality insights for you. It understands yourself better than a therapist would after 10 sessions.
AI BREAKTHROUGH
Alibaba just released Qwen3.5 Small, a new family of lightweight AI models that can run on devices like laptops and phones. And the most powerful version outscored an OpenAI model more than 13× its size on reasoning benchmarks.
The series includes 4 different models, designed for different hardware levels:
0.8B parameters → small enough for phones
2B parameters → lightweight edge AI tasks
4B parameters → stronger multimodal reasoning
9B parameters → laptop-level AI performance
The 9B model managed something unusual. It outscored OpenAI’s GPT-OSS-120B on graduate-level reasoning and multilingual knowledge tests.
The 4B model even matched visual benchmark scores that previously required models about 20× larger.
Even Elon Musk reacted to the release, calling the models’ performance “impressive intelligence density.” And because these models are open-weight, developers can experiment freely.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply