⚡️ GPT-4V is Flawed

Good Morning. When GPT-4 was first unveiled, there were great expectations of its ability to understand the context of images as well as text. However, a recent paper published by OpenAI reveals that GPT-4 equipped with Vision (GPT-4V) still has inevitable shortcomings including privacy issues, biases, and mistaken inferences. Let’s dive right in.

Today's Highlights:

• OpenAI Could See A 300% Valuation Spike

• Microsoft Introduces AI Assistant 'Copilot' in Windows 11

• Spotify’s plans to introduce AI podcast translation

DEEP DIVE

GPT-4 With Vision Still Has Flaws, Paper Reveals

OpenAI has been tempering the most problematic aspects of GPT-4V. Currently, the model has only been used by a limited user base via the app Be My Eyes, which assists visually impaired individuals in interpreting their surroundings.

Can GPT-4V fit into our vision of an AI future?

• Despite OpenAI's efforts, the model tends to 'hallucinate' or produce facts in an authoritative tone, overlook obvious objects, and fail to accurately recognize mathematical symbols.

• Stacked safeguards weren't enough to inhibit discriminatory biases. Disabled production safeguards revealed GPT-4V's inclination towards a bias against certain body types and genders.

• GPT-4V displayed a risky hurdle in the medical imaging domain, giving incorrect answers to previously correctly answered questions, and hinting at unreliable consistency.

OpenAI has explicitly warned against using GPT-4V to identify harmful substances or chemicals in images. Interestingly, models occasionally correctly identified toxic foods like poisonous mushrooms but stumbled while identifying substances including cocaine and fentanyl from their chemical structures images.

Moreover, GPT-4V failed to understand modern nuances and interpretations of certain hate symbols and even generated praised songs or poems for hate figures when figuring them from images.

OpenAI reckons GPT-4V is at an early stage—with an array of strict safeguards in place to prevent 'toxicity' outflow and to ensure the privacy of individuals implemented. Maybe the future of AI isn't as clear-cut as it seems, but OpenAI promises to continue building "mitigations" and "processes" to expand GPT-4V capabilities safely and ethically.

Keep reading here.

PUNCHLINES

Bard’s Faux Pas: Google Search is caught publicly indexing users’ conversations with Bard AI.

Cashing In: OpenAI could see a 300% valuation spike—possibly inflating the company's value to a whopping $90 billion.

Automate, Innovate, and Create: Microsoft introduces AI assistant 'Copilot' in Windows 11.

Privacy Breach? Amazon to start using user conversations to train Alexa's AI, raising privacy concerns.

Level Up: Square now offers AI-generated product descriptions and website design templates in its eCommerce tools.

Face Value: Louisiana police sued for wrongful arrest based on faulty AI face recognition.

TLDR

Microsoft works on cost-effective AI Plan B: Despite owning nearly half of OpenAI, Microsoft is developing an alternative, more cost-effective AI strategy. As AI models like OpenAI's GPT-4 are costly to run, Microsoft aims to create smaller, cheaper conversational AI systems—even if they're less powerful. In-house, less powerful AI models are already being tested in products such as Bing Chat.

Open-Source Multimodal AI, NExT-GPT, to Challenge Google and OpenAI: Developed by the National University of Singapore and Tsinghua University, NExT-GPT is an "any-to-any" multimodal AI model. Designed to process and generate combinations of text, images, audio, and video, this open-source system can be tailored by users to meet specific needs, providing an alternative to established tools like Google's Gemini and OpenAI's ChatGPT-Vision.

EU seeks stronger generative AI safeguards to prevent disinformation in elections: The European Union calls for increased measures to address the potential threat of AI-generated disinformation during elections. While major platforms like Google, Microsoft, and TikTok have begun implementing safeguards, EU Commissioner Vera Jourova insists more action is needed. Pending the EU AI Act, existing guidelines advocate proactive disclosure of deepfakes and AI-manipulated content.

Spotify reverses AI music ban, and introduces AI podcast translation: Spotify is now embracing AI-generated music, retracting its previous ban following the controversial 'Heart on My Sleeve' song. Moreover, the platform plans to pilot an AI-driven feature that translates podcasts into various languages while keeping the original speaker's voice—an initiative likely to boost the reach of podcasts across non-English speaking markets.

TRENDING TOOLS

🔄 Datatera: Convert diverse data formats or websites into structured forms efficiently

🤗 HuggingFace Inference: Curated API endpoints and improved rate limits for AI models

📱 AppyHigh Prime: First-of-its-kind generative AI-app bundle

⚕️ Athelas Scribe: Transformed clinical intake process for health systems

💼 Folk: A lightweight AI-powered CRM tailored to your needs

That’s all for today—if you have any questions or something interesting to share, please reply to this email. We’d love to hear from you!

P.S. If you want to sign up for the Supercharged newsletter or share it with a friend, you can find us here.