😵 Hallucinating AIs

What we’ll cover today:

🤯 OpenAI’s new models love to lie

👀 Meta drops 5 crazy-good AI tools

🇨🇳 Huawei flexes harder than Nvidia

⚙️ 4 tools to boost your AI game

🎥 Must-watch AI guides & hacks

OpenAI’s New Models Are Smart… But Kinda Delusional

OpenAI’s latest AI models o3 and o4-mini—meant to be super smart—are hallucinating way more than older versions like o1.

🚨 Breaking News
OpenAI says o3 hallucinates 33% of the time and o4-mini? A chaotic 48%. That’s double the rate of o1.

⚡ Wire Simplified

• o3 and o4-mini are part of OpenAI’s “reasoning” model fam.

• They’re designed to think deeper—but also make stuff up more.

• o4-mini is basically like that one friend who confidently tells wrong facts.

• Other models like GPT-4.5 and GPT-4o actually hallucinate less.

• HuggingFace’s benchmark shows way lower hallucination rates—so now it's a who-do-you-trust game.

✔️ Straight to the Point
More thinking doesn’t always mean more truth. OpenAI might’ve made smarter AIs that believe their own lies a bit too much.

Meta Drops 5 Wild AI Tools to Mimic Human Smarts

Meta’s AI lab (FAIR) just went all in—dropping FIVE new tools to help machines see, speak, move, and even argue like humans.

🚨 Breaking News
From robot eyes to social agents, Meta’s big push is toward AI that perceives and collaborates just like us.

⚡ Wire Simplified

• Perception Encoder helps AI "see" subtle stuff in images and videos (yes, even stingrays).

• PLM is an open-source vision-language model trained on 2.5M video Q&As.

• Locate 3D lets robots understand “that vase near the TV” kinda instructions.

• Dynamic Byte Latent Transformer ditches tokens, reads raw bytes = more efficient and less dumb.

• Collaborative Reasoner teaches AI how to disagree nicely and solve problems with others.

✔️ Straight to the Point
Meta’s cooking up AI that’s not just smart—it’s socially aware, visually sharp, and potentially your next robot teammate.

Huawei’s AI Beast Could Knock Nvidia Off the Throne

Chinese tech giant Huawei just dropped the CloudMatrix 384 Supernode—an AI chip system said to be stronger than Nvidia’s best.

🚨 Breaking News
Huawei’s new system hits 300 petaflops—crushing Nvidia’s 180—and it's made with zero U.S. tech.

⚡ Wire Simplified

• CloudMatrix 384 is Huawei’s response to Nvidia’s AI dominance.

• Built during U.S. sanctions using only Chinese-made parts.

• Outperforms Nvidia’s NVL72 in compute and token speed.

• Helps run China’s DeepSeek-R1, a high-level reasoning model.

• Ties into China's bigger $52.4B AI infra push led by Alibaba.

✔️ Straight to the Point
Huawei might’ve just proved that even with sanctions, China’s AI game is leveling up—and Nvidia finally has real competition.

🛠️ TOOLS

• Omakase – Talk, sell, close. All in one.

• Voicenotes – Remembers everything so you don’t have to.

• Checklist.gg – AI checklists that actually get done.

• Summarization AI – TL;DR any page in seconds.

📚 RESOURCES

• How I Use AI to take perfect notes...without typing – No typing, perfect notes.

• These 10 AI Uses Will 10x Your Daily Workflow – AI that saves hours.

• If I wanted to build $1M+ AI startup in 2025, I'd do this – Million-dollar AI playbook.

• How To Use Notta AI Note Taker Effectively – Notta tips that slap.

OpenAI’s Step-by-Step Guide to Building Smart AI Agents

This 32-page guide from OpenAI shows how to build AI agents that can do tasks for you—with almost no help. It breaks down tools, models, and safety tips so anyone can get started.
If you want your AI to actually get stuff done for you, this is the cheat sheet you need.

👉 Read the full guide here 🔥