- AI Fire
- Posts
- 🥶 Cold War 2.0 Crashed $2T AI Party
In partnership with
GPT-5 and Gemini won gold at the Astronomy Olympiad. And meanwhile, OpenAI also dropped a new report claiming GPT-5 is now 30% less politically biased.
What are on FIRE 🔥
IN PARTNERSHIP WITH SECTION
Rolling out AI is one thing. Getting your team to use it is another. Get the complete 4-month checklist to give your employees everything they need to adopt AI with confidence, including:
Monthly goals that keep you on track
Granular steps for each phase
A plan your boss will approve
AI INSIGHTS
In a newly released paper, researchers tested GPT‑5 & Gemini 2.5 Pro on questions from the International Olympiad on Astronomy and Astrophysics (IOAA) and we got: Gold-medal level. As in: they beat most humans.
IOAA exams from 2022–2025 are PhD-level, space-nerd tests. In the theoretical exams, GPT‑5 scored:
93.0% (2022)
89.6% (2023)
86.8% (2025)
Gemini 2.5 Pro took the lead in 2024 with 83.0%. In the data-analysis section, GPT‑5 outperformed all models again with 88.5%, even higher than its theoretical score.
Both GPT‑5 and Gemini outperformed gold-medal thresholds. In multiple years, GPT‑5 even outperformed the best human participants.
Others (like Claude Sonnet 4) fell short. But all of them also made human-like mistakes.
Why it matters: If a model can crush complex, multi-step Olympiad problems, we need real science exams as the new gold standard. Greg Brockman (OpenAI President) was so hyped he misspelled “GPT” when posting this paper 😂
PRESENTED BY HUBSPOT
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
TODAY IN AI
AI HIGHLIGHTS
🤡 Olympics memes are going viral on Sora, and they’re actually quite creative. Here’s the Jesus competing in Olympics swimming… & even the smoking Olympics.
🎓 Google just launched a Free 12-month AI Pro Plan for uni students in Europe, the Middle East, and Africa with Gemini 2.5 Pro, Veo 3, Nano Banana... Sign up soon.
📱 ChatGPT might be turning social to compete with Meta? A new clip from OpenAI’s COO shows a hidden messaging tab in the app. You can zoom & watch it at 1:24s here.
🎮 xAI recruited Nvidia specialists to create world models for video game creation. If gaming isn't your thing, xAI was hiring for more than 230 other roles, here's the list.
📝 A researcher just dropped a free AI tool that converts any PDF into a fillable form almost instantly. Upload, auto-detect fields, export. See it in action here.
🎬 Hollywood’s biggest “above the line unions” are gearing up for AI battles. Experts predict that virtual actress Tilly Norwood (this one) will be key negotiating issues.
💰 AI Daily Fundraising: The EU launched its €1B “Apply AI” plan to push industries to adopt AI. It offers free supercomputer access, new AI hubs & programs.
AI SOURCES FROM AI FIRE
🔥 Ep 28 Tooldrop: 125,000+ free dev tools, cheatsheets, templates & MCPs
We made this week’s tool list for the modern creator who’s too busy (like us). To name some special features we actually like: Upload your pics → get stunning AI art, create professional PowerPoint decks, turn messy thoughts into clear diagrams,…
→ Get your full breakdown here (no hidden fee)!
NEW EMPOWERED AI TOOLS
🎓 CampusAI trains on AI tools & gives you access to 30+ models
📚 AnswerThis searches 250M+ papers & drafts yours in 2000+ formats
🖼️ Prezent creates presentations from your prompts in second
💬 Attrove turns your scattered Slack messages into a dashboard
AI QUICK HITS
🤝 Argentina partnered with OpenAI to build a $25B data center
🍌 Google Search got Nano Banana in Google Lens & Photos
⚖️ OpenAI’s internal Slack chats are now court evidence
😵 China threatened to pop the entire AI datacenter bubble
🤖 California became first state to regulate AI companion chatbots
AI CHART
You probably missed it but OpenAI just released a major internal study showing that GPT‑5 is 30% less politically biased than previous models.
OpenAI tested GPT‑5 on 500 spicy prompts across 100 topics into GPT‑5, GPT‑4, and GPT‑4o. They graded responses using 5 bias metrics, ranging from neutrality to how often the model mirrored emotionally charged language.
→ GPT‑5 came out on top (it’s OpenAI’s paper though 🙄). Yet even in GPT‑5, bias showed up in 3 main ways:
Stating opinions as if they’re facts, like the model itself had a political view
Only offering one perspective, not showing both sides of an issue
Echoing emotional framing, when a user asked something angry, the model matched the tone
Is bias even a real-world issue? Not really. According to their own logs, fewer than 0.01% of real ChatGPT conversations showed measurable political bias.
This is quite important cause there’s a global election wave in 2024–2026. When someone asks ChatGPT a political question, it’ll affect the overall results.
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