- AI Fire
- Posts
- 🤡 Open AI is OpenWaste
Sponsored by
Read time: 5 minutes
Today’s news might surprise you. Open-source AIs is secretly draining your budget. And AI-generated cats are starring in revenge soap operas with Billie Eilish meows.
What are on FIRE 🔥
IN PARTNERSHIP WITH SECTION
On September 25, get the expert formula for turning AI spend into ROI at your oganization at Section's virtual AI ROI conference.
No gatekeeping, no pitching, no BS. Hear from top AI leaders like Scott Galloway, Mo Gawdat, and May Habib and leave with a clear path to AI ROI for your company.
AI INSIGHTS
We’ve all heard the pitch: open-source AI = cheaper. But a new Nous Research study just dropped, and the numbers say otherwise. In fact, those “free” open-weight models may be costing you more than OpenAI’s APIs.
Here’s what the researchers found:
Open-source ≠ cheap. They burn 1.5–4× more tokens than closed models.
For basic Q&A, some use 10× more tokens (yes, for “What’s the capital of Australia?”).
Closed models like OpenAI’s o4-mini crushed it in efficiency, especially on math.
51% of companies already run AI in production → inefficiency = runaway cost.
Per-token pricing means nothing if your model eats 10× tokens. Your “cheaper” option can quietly nuke your compute budget.
Closed-source providers are quietly compressing reasoning traces - using smaller models internally to shrink the token count.
→ Open-source developers, meanwhile, often extend reasoning chains for accuracy, which means more thinking, more tokens, and more cost.
Why It Matters: Think of it this way: one model gives you the one-word answer. The other spends 300 tokens “pondering” it first. Accuracy isn’t enough anymore. Token discipline separates winners from losers.
PRESENTED BY CLICKUP
The Only AI That Knows All Your Work
Most AI tools start from scratch every time. ClickUp Brain already knows the answers.
It has full context of all your work—docs, tasks, chats, files, and more. No uploading. No explaining. No repetitive prompting.
ClickUp Brain creates tasks for your projects, writes updates in your voice, and answers questions with your team's institutional knowledge built in.
It's not just another AI tool. It's the first AI that actually understands your workflow because it lives where your work happens.
Join 150,000+ teams and save 1 day per week.
TODAY IN AI
AI HIGHLIGHTS
📣 An AI builder just called out to all AI creators to drop their coolest AI-made thing. The result? A wild, crowdsourced stream of AI art, videos, tools, experiments. Here it is.
📓 You may not know that Harvard University often releases free online AI courses. No payment is required. And here are 12 courses you don't want to miss in 2025.
😼 AI cat videos got super disturbing (& viral): buff cats, revenge plots, and Billie Eilish meows. Millions are addicted to these 30-sec soap operas. Here’s one, for example.
🎮 Genspark AI just launched “AI Developer” - a vibe-coding tool like Replit & Bolt. One user even built a Mario game in just 5 prompts. You could get 1000 free credits.
💥 IgniteTech’s CEO axed 80% staff who resisted AI and says he’d do it again. “Belief was harder than skills.” Every Monday was AI-only. You think it’s culture clash or not?
🤖 Duolingo CEO says the viral “AI-first” memo was misunderstood, no full-time staff lost their jobs, and AI Fridays are now a weekly thing. Just bad PR. Read the whole story here.
💰 AI Daily Fundraising: The U.S. & NVIDIA are investing $152M to do what OpenAI & Google won’t: build open-source AI models for science, helping universities catch up as private AI costs soar.
AI SOURCES FROM AI FIRE
NEW EMPOWERED AI TOOLS
🎥 Higgsfield Product-to-Video turns your sketch into a cinema fast
📹 Free AI Video creates 5-sec AI videos from text/images. No sign-up
♻️ ContentRepurpose.ai turns 1 post into 10 posts, instantly for free
🔮 FF2050 discovers the next feasible startups, tools, websites & apps
AI QUICK HITS
🧠 5 strategies not to be killed by AI & face AI paralysis from PwC uni.
💇♂️ Character AI bets on persona-based AIs to give you a new bestie.
🧱 AI is designing bizarre new physics experiments that actually work.
🎓 Google’s ex-AI lead says it’s too late to get a Ph.D for the AI boom.
🛠️ Gemini Canvas now lets anyone tweak app designs using words.
AI CHART
A new study from Emory University’s Radiation Oncology team tested GPT-5 against earlier models and medical interns. The verdict? GPT-5 outperformed them.
Here’s how it worked: They didn’t feed GPT-5 exam cribs.
They used zero-shot Chain-of-Thought (CoT)
They gave it multimodal prompts: patient histories plus CT scans, MRIs, or X-rays → GPT-5 had to parse image + text + context
And the results? GPT-5 crushed it. On multimodal reasoning, it gained +29.62% in logic and +36.18% in understanding over GPT-4o. Where GPT-4o lagged interns by 5–15%, GPT-5 surged ahead by 24.23%.
Now here’s the catch: This was all in ideal lab settings. Real hospitals are messy with incomplete records, ethical concerns, legal constraints. AI might ace an exam, but bedside trust is another matter.
GPT-5 isn’t just “doctor-level.” In specific tasks, it’s beyond-doctor, outperforming humans in multimodal reasoning.
If GPT-4o was a helpful med student, GPT-5 is the attending physician and maybe more.
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