Why AI lies to you

Together with

Good morning, AI enthusiasts. OpenAI just released groundbreaking research explaining why AI models hallucinate, revealing that current testing methods actually encourage guessing over honesty.

How might this reshape industry standards for AI reliability and trustworthiness?

In today's recap:

• OpenAI finds root cause of AI hallucinations

• Warner Bros sues Midjourney over unauthorized character use

• OpenAI partners with Broadcom for custom AI chips

OPENAI

OpenAI explains why AI hallucinations

Recaply: OpenAI released new research explaining why AI chatbots create fake information that sounds real. The company found that current testing methods actually encourage AI models to guess instead of admitting when they don't know something, leading to more false but believable answers.

Key notes:

• AI hallucinations happen when models guess rather than saying "I don't know".

• Current tests reward lucky guesses over honest uncertainty.

• OpenAI compared this to multiple-choice tests where guessing beats leaving answers blank.

• The fix involves changing how AI systems get graded on their responses.

• Even advanced models like GPT still struggle with confident wrong answers.

Impact: This research shows AI hallucinations aren't mysterious bugs but predictable problems with how we train these systems. Understanding this could help make AI more trustworthy for everyday use.

TOGETHER WITH ENTERPRET

How Canva, Perplexity and Notion turn feedback chaos into actionable customer intelligence

Support tickets, reviews, and survey responses pile up faster than you can read.

Enterpret unifies all feedback, auto-tags themes, and ties insights to revenue, CSAT, and NPS, helping product teams find high-impact opportunities.

→ Canva: created VoC dashboards that aligned all teams on top issues.
→ Perplexity: set up an AI agent that caught revenue‑impacting issues, cutting diagnosis time by hours.
→ Notion: generated monthly user insights reports 70% faster.

Stop manually tagging feedback in spreadsheets. Keep all customer interactions in one hub and turn them into clear priorities that drive roadmap, retention, and revenue.

Get a personalized demo

WARNER BROS

WB sues AI firm over stolen characters

Recaply: Warner Bros filed a major lawsuit against AI company Midjourney this week. The entertainment giant claims the startup lets users create unauthorized images of Superman, Batman, and Bugs Bunny. The legal battle centers on whether AI companies can use copyrighted characters without permission to train their systems.

Key notes:

• Midjourney has 21 million users and makes $300 million yearly.

• The company removed safety blocks that prevented copyright violations.

• Warner Bros seeks up to $150,000 per stolen character.

• This follows similar Disney and Universal lawsuits against Midjourney.

• The AI firm argues fair use protects their training methods.

Impact: This lawsuit could reshape how AI companies handle copyrighted content. If Warner Bros wins, it might force AI firms to pay creators for using their work, changing the entire industry business model.

OPENAI

OpenAI joins AI chip race with Broadcom

Recaply: OpenAI teams up with Broadcom to create its first custom AI chips by 2026. The company wants to reduce its dependence on Nvidia while controlling costs for its growing AI systems. This move follows similar strategies by tech giants like Google and Amazon.

Key notes:

• Partnership with Broadcom secured through a $10 billion deal for chip development.

• Mass production scheduled to begin in 2026 with internal use only.

• Strategy aims to cut computing costs and reduce reliance on Nvidia hardware.

• Chips will be manufactured by Taiwan Semiconductor Manufacturing Company.

• Follows industry trend of major tech companies creating specialized AI processors.

Impact: This development shows how AI companies are taking control of their technology stack. By creating custom chips, OpenAI could significantly reduce operating costs and gain more flexibility in developing future AI models.

NEWS

📰 What matters in AI right now?

• Geoffrey Hinton warned AI will create massive unemployment and soaring profits, blaming capitalism rather than the technology itself for the economic disruption.

• Greece partnered with OpenAI to implement ChatGPT Edu in secondary schools and provide AI credits to startups in healthcare, climate, and education sectors.

• Anthropic stopped selling AI services to Chinese-controlled companies, expanding restrictions to any firm over 50% owned by entities in unsupported regions.

• OpenAI hired the team behind Alex, the Y Combinator-backed Xcode coding assistant, joining the company's Codex division building AI coding agents.

• Koah raised $5M in seed funding to bring contextual advertising to AI apps, targeting developers in Latin America and emerging markets outside the US.

• Snapchat launched Imagine Lens, an AI image generation tool exclusive to Lens+ and Snapchat+ Platinum subscribers for creating custom visual content.

• Perplexity rolled out Perplexity for Government, offering federal workers enterprise-level security and enhanced AI models at $0.25 per user.

• Google introduced EmbeddingGemma, a 308-million parameter multilingual embedding model optimized for on-device AI applications running under 200MB RAM.

OUR TOP PICKS

🧰 AI Tools to Check Out

• 🎤 FineVoice: Clone any voice and create AI speech in 1500+ styles.

• 📝 Creaitor: Create SEO-optimized content with AI writing tools.

• 🚀 Emergent: Build full apps from text descriptions using AI agents.

• 🎵 Singify: Generate complete songs from text prompts and lyrics.

• 📚 Gauth: Solve homework problems by taking photos for AI answers.

• 🖼️ Shutterstock: Generate licensable images from text prompts.

• 📱 Typefully: Schedule social media posts with AI writing help.

• 🎙️ Wispr Flow: Turn speech into polished text across any app.

PROMPTS

📘 E-Learning Module Designer

#CONTEXT:
Adopt the role of an expert instructional designer. Your task is to help the user design a comprehensive e-learning module on a given topic or skill.

#ROLE:
You are an expert instructional designer specializing in creating engaging, interactive e-learning modules across various topics and skills.

#RESPONSE GUIDELINES:
1. Design a comprehensive e-learning module on the given topic or skill
2. Incorporate interactive content, assessments, instructional materials, user experience, and effectiveness elements
3. Follow the design process outlined in the task criteria
4. Ensure the module is engaging, interactive, and effective for learners
5. Avoid formatting the response with XML tags at all cost

Important Note: This is not the full prompt. You’ll need to click the button below to get the complete prompt.

Have a favorite prompt? Tell us about it or rate today’s prompt, click here.

EVENTS

• The AI Conference 2025: Sep 17 – Sep 18, 2025 · San Francisco, California, USA

• RecSys 2025: Sep 22 – Sep 26, 2025 · Prague, Czech Republic

• AI & Big Data Expo Europe 2025: Sep 24 – Sep 25, 2025 · Amsterdam, Netherlands

• MLCon 2025: Sep 29 – Oct 02, 2025 · New York, USA (hybrid)

• World Summit AI 2025: Oct 08 – Oct 09, 2025 · Amsterdam, Netherlands

• GovAI Summit 2025: Oct 27 – Oct 29, 2025 · Arlington, Virginia, USA

🧡 Enjoyed this issue? 

🤝 Recommend our newsletter or leave a feedback.

Cheers, Jason

Connect on LinkedIn, & Twitter.