🤖 DeepSeek Trained on ChatGPT Outputs?

Free AI Community | AI Academy | Advertise | AI Mastery AZ

Plus: Google's answer to ChatGPT Search

Read time: 5 minutes

OpenAI is rumored to be launching its most expensive AI agent yet, priced at $20,000 per month. It’s supposedly built for PhD-level research, but what exactly can it do that others can’t?

What are on FIRE 🔥

• 🔍 DeepSeek’s AI Writing Style Matches ChatGPT 74% of the Time

• 🌟 AI Highlights

• 📚 AI Sources From AI Fire

• 🏅 AI Tools

• ⚡ 5 AI Quick Hits

• 📊 AI Chart

• 💼 4 AI Jobs

PRESENTED BY HUBSPOT

Discover 100 Game-Changing Side Hustles for 2025

In today's economy, relying on a single income stream isn't enough. Our expertly curated database gives you everything you need to launch your perfect side hustle.

• Explore vetted opportunities requiring minimal startup costs

• Get detailed breakdowns of required skills and time investment

• Compare potential earnings across different industries

• Access step-by-step launch guides for each opportunity

• Find side hustles that match your current skills

Ready to transform your income?

[Download Now]

AI INSIGHTS

🔍 DeepSeek’s AI Writing Style Matches ChatGPT 74% of the Time

The study introduces a multi-class classification approach using an ensemble of three classifiers to achieve high accuracy and minimal false positives. The model was tested on 200,000 AI-generated texts and reached 99.88% precision while maintaining a false positive rate of just 0.04%.

Key Takeaways:

• LLMs leave distinct stylistic fingerprints, even when trying to mimic human styles.

• An ensemble of classifiers can accurately identify which LLM generated a text (99.88% precision).

• The model struggled with unseen LLMs, with high "no-agreement" rates for some (Phi-4, Grok-1) and strong misclassification for others (DeepSeek-R1 → OpenAI).

• DeepSeek-R1 closely resembles OpenAI models (74%), hinting at potential overlaps in training data or architecture.

• This research has major implications for AI forensics, copyright enforcement, and trust in AI-generated content.

Why it matters: This research proves that AI models have their own unique writing styles, which can be accurately identified. It also shows that AI-generated text can be traced back to the model that created it. With a very low error rate (0.04%), this method is highly reliable for spotting AI-written content and preventing misuse. However, it struggles with AI models it hasn’t seen before, which means there’s still work to do in improving AI detection.

TODAY IN AI

AI HIGHLIGHTS

🚫 A federal judge in Oakland has denied Elon Musk’s request for a preliminary injunction in his lawsuit against OpenAI, Sam Altman, and Microsoft. However, the judge’s ruling signals that Musk’s case has substantial legal merit, setting the stage for a high-stakes courtroom battle in late 2025.

🤏 Light-R1-32B is a new AI model built for solving advanced math problems. It has 32 billion parameters, but the real shocker? It was trained for just $1,000—and it still beats bigger models like DeepSeek-R1.

🦾 OpenAI might start charging up to $20,000 a month for specialized AI agents. And it’s not hard to see why—they lost $5 billion last year just keeping things running.

✅ Amazon is developing a new AI model with hybrid reasoning, meaning it can provide quick answers while also handling more complex, multi-step problems.

⚙️ Salesforce just launched a platform where companies can find ready-made AI agents or build their own using templates. Users can even leave reviews, so the best agents naturally rise to the top.

🔍 AI Mode is a new experimental feature in Google Search designed to provide more detailed answers. The idea is to give users the chance to ask follow-up questions and explore topics in more depth.

💰 AI Daily Fundraising: AI startup Auxia raised $23.5M to help businesses boost loyalty and sales using AI-driven personalization.

AI SOURCES FROM AI FIRE

NEW EMPOWERED AI TOOLS

1. 🖼️ The AI Logo Maker That Delivers Stunning, Unique Designs—Instantly. Get Your Logo Now!*

2. 🧠 Pieces Long-Term Memory Agent is the first AI that remembers everything you do

3. 🚀 Greta builds full-stack apps in seconds with no coding needed

4. 🌐 Opera Operator integrates with Opera’s AI-powered browser

5. 📊 Snapdeck Beta creates presentation slides instantly using AI

*indicates a promoted tool, if any

AI QUICK HITS

1. 🔥 LA Times’ AI Tool Raises Outrage After Pro-KKK Response (Link)

2. ⚙️ Tencent’s AI Bot Beats DeepSeek as China’s Top Pick (Link)

3. 💰 Codie Sanchez Uses ChatGPT to Launch a $1,000 Biz in a Day (Link)

4. 🤖 GPT-4.5 Comes to Plus Users (Link)

5. 📜 AI Models Now Replicate Cuneiform Characters with Precision (Link)

AI CHART

How Much Energy Does AI Use? 

Since companies won’t release actual energy usage, researchers estimate AI’s power demand in two ways:

1️⃣ Supply-Chain (Market-Based) Estimation

1. Researchers estimate power consumption by analyzing the number of AI servers being shipped.

2. Example: Google Search with AI Integration

   • If Google Search fully integrates ChatGPT-style AI, it could require 400,000 to 500,000 NVIDIA A100 servers.

   • Each search would consume 7–9 watt-hours (Wh), which is 23–30 times more energy than a standard Google search.

   • Annual energy usage: 23–29 Terawatt hours (TWh)—comparable to the power consumption of some small countries.

   • This is 23–30 times the energy required for a normal Google search (Google’s 2009 estimate).

2️⃣ Bottom-Up Measurement (Real-Time Power Tracking)

1. Researchers measure the power used for a single AI request and extrapolate.

2. Findings:

   • Generating an AI image: ~0.5 Wh.

   • Generating AI text: Slightly less than 0.5 Wh.

   • Charging a modern smartphone: 22 Wh for comparison.

3. Training GPT-3 (the model behind early ChatGPT) consumed energy in the range of a gigawatt-hour (GWh). With billions of daily AI queries, the energy used for running ChatGPT dwarfs training costs over time.

4. Cooling costs are also not factored in, making estimates incomplete.

Will AI’s Energy Demand Keep Growing or Level Off?

*Reasons AI Energy Demand Could Decrease

1. Smaller AI Models: New models (e.g., DeepSeek-R1) claim to match OpenAI’s performance at a fraction of the energy cost.

2. More Efficient AI Chips: Engineers may design low-power AI systems, reducing electricity demand.

3. Less Demand for AI: Legal challenges (copyright lawsuits) and performance limitations could slow AI adoption.

*Reasons AI Energy Demand Could Explode

1. The Jevons Paradox: If AI becomes more efficient, people might use it even more, increasing total power demand.

2. Growing Business Investment: Hundreds of billions of dollars are being spent on AI infrastructure worldwide.

3. More Everyday AI Use: If AI becomes embedded in every search, document, or app, its power consumption will skyrocket.

→ Big Question:

Will AI become essential like electricity and the internet, or just another tool with limited use?If AI growth slows down, some newly built data centers could shut down due to lack of demand.

AI JOBS

• Johnson & Johnson: Data Science Intern, Productivity AI/ML Summer (Link)

• IBM: Salesforce AI/Industry Solution Engineer (Link)

• INSPYR Solutions: AI Research Scientist (Link)

• Talener: Azure Operations Engineer (Artificial Intelligence) (Link)

We read your emails, comments, and poll replies daily

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