- AI Fire
- Posts
- 👮♂️ Open(ish)AI with No freedom ChatC.O.Police
👮♂️ Open(ish)AI with No freedom ChatC.O.Police
GPT-Skip-In-5... if you’re lucky
Wendy
October 29, 2025
In partnership with
Robots failed to pass butter. OpenAI dropped new safety models. And Meta’s AI ad tool swapped a model for a grandma. It’s funny, weird, and kind of important today.
What are on FIRE 🔥
IN PARTNERSHIP WITH SECTION
Danilo McGarry took a €900M company and turned it into an AI-first organization worth €4.9B. You'll learn how he did it, the essential strategy, and the hang-ups to avoid.
Join Section on 11/20 from 12-1pm ET as they pick the brain of a consultant who used AI to 5.5X the value of his company.
AI INSIGHTS
OpenAI’s been under pressure to prove it still cares about safety and now it’s finally doing something. They dropped 2 new reasoning models:
gpt-oss-safeguard-120b(big brain)gpt-oss-safeguard-20b(slightly smaller brain)
They’re built on the gpt-oss models from August, but tuned specifically for online safety (detecting fake reviews, toxic comments, cheating talk)
These models show their reasoning, not just “yes/no,” but why something was flagged for transparency. You can download and try them now as part of a research preview.
→ A Reddit mod forum might use them to detect hate speech. A review site could catch fakes. A Discord server could block cheats. It’s all about setting your own safety policy.
For us, it’s kind of ironic, everyone knows they’ve been accused in that “Parents sue OpenAI, alleging ChatGPT assisted their son’s suicide” case, and of skimming on safety.
Why it matters: They're still early though, research preview only. You’ll need to provide feedback to help grow them. And honestly? That’s a step in the right direction, for now.
PRESENTED BY SYNTHFLOW
Introducing Synthflow Voice AI Agents for WhatsApp Calls
Until now, answering WhatsApp calls with the same speed, consistency, and automation as phones was nearly impossible.
With Synthflow, that changes. Connect your WhatsApp Business account instantly and have every call answered by Voice AI Agents that resolve issues, book appointments, track orders, send follow-ups, and escalate when needed.
24/7 coverage, 200+ integrations, 30+ languages, and enterprise-grade security — all in one platform.
AI SOURCES FROM AI FIRE
1. The simple 'DIG' method to make AI work like a smart assistant. You’ll learn the 'DIG' framework, a simple way to get accurate data analysis from AI.
2. Sora 2 is secretly 6x cheaper (using this n8n hack). The complete no-code guide to combining Sora 2 and n8n that creates high-quality video content.
3. Make money with AI songs. 5 easy ways you can start today. This article shows 5 real ways to earn, even if you don’t have any skills, from YouTube to selling sounds.
TODAY IN AI
AI HIGHLIGHTS
😮 Google released a new web that designs full AI marketing campaigns just based on our website. We took a look at the demo & had to say “wow”. It literally gets your DNA. Demo here.
🚨 OpenAI may add ads to ChatGPT. People aren’t happy, we’ve seen so many saying “ads? okay bye.” You can see more reactions & also write your own opinion here.
👵 Then, Meta’s AI ad tool went rogue too, replacing a fashion model with an AI grandma. See the ad fail here. It’s better to test these ad tools by yourself & pick one.
🔥 Yesterday, did we mention you’ll also eventually be able to edit with Photoshop & Express directly inside ChatGPT? That demo was sick (here’s a video from Darren)
📺️ We’ve been into Sora, Veo or Nano Banana, AI clip from SpongeBob is also viral on social media, using LTX-2 (just a short prompt is enough). You can give it a try here.
💰 AI Daily Fundraising: Fireworks AI raised $250M in its latest funding round. It’s supported by investors like Lightspeed, Index Ventures, and NVIDIA.
AI TUTORIAL
Mark your calendar: Agorapulse is hosting a free live webinar, showing how to use AI for social media without losing the human touch.
In just 60 minutes, you'll learn:
How to turn briefs, captions, and repurposing into minutes, not hours
Ways to triage engagement and never miss priority replies
Simple prompts and guardrails your team can adopt today
How to connect results to pipeline, not just vanity metrics
What to automate vs. what to keep human
You will get a Free practical PDF workflow to use instantly when you sign up for the on-demand replay called: “AI for Social Media That Actually Delivers”.
NEW EMPOWERED AI TOOLS
AI QUICK HITS
🌿 Eli Lilly has opened the world's largest AI factory with NVIDIA
🤖 GitHub unified AI coding agents from OpenAI, Anthropic, Google
🚀 KakaoTalk now features AI tools using GPT-5 for smarter chats
⚡️ Amazon’s biggest Indiana data center is out, boosting Claude
💪 OpenAI offers ChatGPT Go to one more country, today’s Brazil
AI CHART
A new robotics benchmark called Butter-Bench tested how well today’s top AI models, like GPT-5, Claude, and Gemini, could control simple home robots to do tasks like “pass the butter” or:
Finding a butter package & figuring out which one is the butter
Navigating to the user & confirming pickup, then returning to the dock
Sounds easy, right? We do this all the time without thinking. But here's how it went:
Humans crushed it with a 95% success rate (obviously)
Top AI (Gemini 2.5 Pro)? Just 40%
Others like Claude, GPT-5, Grok, and LLaMA… also struggled
They spun in circles. Missed obvious visual cues. Lost track of where they were. And when their battery started dying? The bots lost their minds. Claude Sonnet 3.5 hit low battery and couldn’t redock.
It feels like watching a toddler try to figure out the world, with a PhD in literature. You can almost feel it, the physical AI might start with a robot trying (& failing) to pass the butter.
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