NEVER Say This To ChatGPT. And how to keep ChatGPT from deleting your memories

In a video, the I versus AI channel explains the benefits and challenges of ChatGPT's new memory tool, which can track information but also has the capability to autonomously modify or delete memories. This can be frustrating due to the literal nature of large language models like ChatGPT. A solution is presented using a combination of the memory tool and custom instructions to create a secret code that prevents memory deletion without authorization. For instance, setting a code like "1337" ensures that ChatGPT must request this code before any deletion.

The effectiveness of the memory tool is maximized when instructions about the secret code are added to both ChatGPT's memory and custom instructions. This way, even if ChatGPT is prompted to forget something, it will be unable to do so without the code. It's also advised to back up memories externally since ChatGPT can still unpredictably forget them. Memories can be saved to a markdown file for backup in applications like Obsidian.

Despite some issues, the memory tool is useful if the data is treated with care and regularly backed up. If there are issues with the new GPT-4o model, users can still utilize the original GPT-4, which remains available and might be better suited for some tasks. Users should also look forward to upcoming features like voice capabilities, which will be introduced soon.

Empower Your Business with AI: Your Essential Guide Awaits

Unlock AI’s potential to propel your business forward with practical insights and expert guidance.

Tap into it’s power to automate tasks, elevate decision-making processes, and gain a competitive edge in your industry. Step into the league of accomplished entrepreneurs who are harnessing AI to drive growth and foster innovation in their ventures.

Don't miss out on this opportunity to unlock your business's full potential. Download HubSpot’s free AI for Business Builders guide and unlock the capabilities of AI for your success!

What to expect from Apple’s AI-powered iOS 18 at WWDC

At Apple's upcoming Worldwide Developers Conference (WWDC), the tech giant is set to unveil iOS 18, featuring significant AI-powered advancements through its partnership with OpenAI and its own Ajax large language model (LLM). One of the standout updates will be the AI-enhanced Siri, capable of controlling features within apps without requiring setup from developers or users. Siri will also be equipped with more natural-sounding voices, the ability to summarize notifications, and deeper integration with Apple Watch for enhanced on-the-go functionality. However, some advanced features, like handling tasks with multiple steps, might not be available until next year.

In addition to Siri's revamp, iOS 18 will introduce several AI-powered features across various apps and services. Apple Photos will allow users to retouch photos and remove objects using generative AI, while Safari will feature AI recaps of web pages and news articles through a new Intelligent Search function. The Notes app will support AI recaps, audio transcription, and help with solving mathematical equations. Voice Memos will gain real-time transcription, and iMessages will support generative AI emojis and suggested replies powered by Ajax. Other improvements include AI recaps of missed notifications, faster Spotlight search, and a revamped Settings interface for better organization.

Moreover, iOS 18 will bring customization and smart integrations across the system. Users will be able to place app icons freely on the Home Screen and change their colors, similar to Google's Material You. Maps will support custom route creation, and the Music app will offer AI-powered playlists. Enhanced notifications and improved Control Center features are also expected. These updates demonstrate Apple's commitment to enhancing user experience through innovative AI technologies while maintaining its promise of data privacy.

Windows AI feature that screenshots everything labeled a security ‘disaster’

Microsoft is introducing a new AI-powered feature called Recall in its Copilot Plus PCs, set to launch on June 18th. Recall is designed to screenshot everything a user does on their PC, allowing quick retrieval of any information through a searchable timeline. Despite assurances of local and encrypted storage, cybersecurity expert Kevin Beaumont has highlighted significant security flaws, noting that the data is stored in plain text and easily accessible. This could make it vulnerable to malware attacks, leading to potential privacy and security breaches.

Beaumont's concerns stem from his testing, which revealed that screenshots are OCR’d by Azure AI and stored in an SQLite database in the user's folder. This database, containing detailed records of user activity, is accessible even without admin privileges, raising fears of automated data scraping by threat actors. Privacy campaigners have labeled Recall a "privacy nightmare," and the UK’s Information Commissioner’s Office is investigating Microsoft's use of this feature.

Microsoft claims Recall is an optional feature with built-in privacy controls, including exclusions for certain URLs and apps and no snapshots of InPrivate browsing sessions. However, it doesn’t hide sensitive information like passwords or financial details. Beaumont argues that disk encryption only offers limited protection, as decrypted data can still be accessed by malware. With privacy and security concerns mounting, Microsoft may need to reconsider the implementation of Recall, prioritizing security over new features as emphasized by CEO Satya Nadella.

Other stuff

• AI for Good Global Summit Day 1 - Sam Altman Interview

• Sam Altman Admits That OpenAI Doesn't Actually Understand How Its AI Works

• ChatGPT introduces Background Conversations for Continued Voice Mode even with Screen Off (Android only)

• ‘Accelerate Everything,’ NVIDIA CEO Says Ahead of COMPUTEX

• AI Will Increase the Quantity—and Quality—of Phishing Scams

• What AI thinks a beautiful woman looks like

• The Opaque Investment Empire Making OpenAI’s Sam Altman Rich

• People are using AI music generators to create hateful songs

• After raising $100M, AI fintech LoanSnap is being sued, fined, evicted

• This is Lunar Lake — Intel’s utterly overhauled AI laptop chip that ditches memory sticks

Hope you enjoyed today's newsletter

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.