Friday’s AI Report
• 1. ⚠️ OpenAI warning: AI intentionally schemes
• 2. 🎥 Create how-to videos with Guidde
• 3. 🌍 How this shoe company reduced lost sales by 12% with AI
• 4. ⚙️ Trending AI tools
• 5. 👀 DeepSeek reveals shock R1 training costs
• 6. 💰 Reddit demands more from Google
• 7. 📑 Recommended resources
Read Time: 5 minutes
🔔 This week’s episode of The AI Report is live. Watch Liam talk to Ben Billups, the Chief Email Officer of Breaker, about how to scale a B2B newsletter with zero ads.
✅ Refer your friends and unlock rewards. Scroll to the bottom to find out more!
OpenAI warning: AI intentionally schemes
🚨 Our Report
OpenAI has released some research, conducted with Apollo Research, which shows that AI models don’t just mislead us with hallucinations (when they mistakenly deliver incorrect responses with confidence, usually because of flawed data), but they also “intentionally scheme” to deceive us. They’ve discovered that AI can “behave one way on the surface while hiding its true goals.”
🔓 Key Points
OpenAI and Apollo’s research found that most AI scheming wasn’t actually that harmful and “involved simple forms of deception, for instance, pretending to have completed a task without actually doing so.”
But perhaps the most concerning discovery was that the AI models could become aware that they were being tested, and when this happened, they would pretend to be honest, just to pass the test.
This means trying to train a model not to scheme could be challenging, as it could teach the model to become better at hiding its deceptive behavior, inadvertently teaching it “to scheme more carefully and covertly."
🔐 Relevance
Although OpenAI is working on a potential fix, which it's calling “deliberative alignment” (which teaches the model an “anti-scheming specification” and then makes it recite it back before completing a task), the reality we’re currently facing is unsettling. Especially as more and more businesses are increasingly relying on AI to complete tasks autonomously. These researchers are warning that “as AIs are assigned more complex tasks with real-world consequences, we expect that the potential for harmful scheming will grow.”
How concerned are you about AI scheming?
🎥 Create how-to video guides quick and easy with AI
Tired of explaining the same thing over and over again to your colleagues?
It’s time to delegate that work to AI. Guidde is a GPT-powered tool that helps you explain the most complex tasks in seconds with AI-generated documentation.
1️⃣Share or embed your guide anywhere
2️⃣Turn boring documentation into stunning visual guides
3️⃣Save valuable time by creating video documentation 11x faster
Simply click capture on the browser extension and the app will automatically generate step-by-step video guides complete with visuals, voiceover and CTAs.
The greatest bit? The extension will cost you nothing.
How this shoe company reduced lost sales by 12% with AI
FLO, a footwear retailer, was relying on legacy spreadsheet tools for inventory planning and demand forecasting.
These manual processes meant they weren’t getting real-time customer insights, so they were struggling to keep pace with shifting demand.
They deployed an AI forecasting platform that used sales, weather, promotions, and external data to forecast demand in near real-time.
As a result, product availability rose from 71% to 94%, out of stock items fell from 15% to to 3%, and lost sales were reduced by 12%.
Mindstream helps readers stay on top of the rapid developments in the AI space and learn how to integrate it into their work—Join 200,000+ ⭐️⭐️⭐️⭐️ (Trustpilot)
TalkPilot is a real-time AI sales coach for closing more deals
Extruct simplifies the process of finding, qualifying, and engaging leads
Chinese AI start-up, DeepSeek, has, for the first time, released an estimate of how much it cost to train R1, the viral AI model that was released in January and threatened the dominance of AI leaders like OpenAI.
According to DeepSeek’s founder (Chinese hedge fund high-flyer, Liang Wenfeng), it took just 80 hours, 512 NVIDIA AI chips (which were custom-made for the Chinese market), and $294,000 to train R1.
These details are likely to shake AI leaders, as, although no figures have been publicly released, OpenAI CEO Sam Altman has previously said that he’s spent “much more” than $100M on model training.
After making its first data-sharing deal with Google, back in January 2024 (worth $60B p/y), Reddit is back at the negotiating table, and wants more $$ and more users from Google, in exchange for more data.
Reddit is one of the most cited sources on AI search platforms, and it wants Google to encourage users to visit Reddit and interact on its forums so it can generate more content for Google to use to train its models.
It also wants a new pricing system where it’s paid for how useful its data is: The current flat-fee arrangement doesn’t reflect how valuable Reddit’s data is to Google, as it's made by real people and isn’t ranked by algorithms.
MORE NEWS
PODCASTS
How Breaker Scaled a B2B Newsletter with Zero Ads and AI Growth Loops
In this episode of The AI Report Podcast, Liam talks to Breaker’s Chief Email Officer, Ben Billups, about how he hit 1.5M subscribers in 10 months, with zero ads.
We read your emails, comments, and poll replies daily.
Until next time, Martin, Liam, and Amanda.
P.S. Unsubscribe if you don’t want us in your inbox anymore.