⚡️ Garbage In, Garbage Out

PLUS: Netflix is splugring on AI roles (apply quickly)

August 28, 2023

Good news for your data! It's now in the 'safe' hands of generative AI and we're here to dig deeper. What else is brewing, you ask? Well, Netflix is splurging on AI roles and a UK startup claims to have designed novel cancer immunotherapy using generative AI. Sound interesting? Let’s dive in.

DEEP DIVE

Enhancing Enterprise Data Security

Generative AI offers a unique opportunity to boost productivity and hammer out innovative ideas in the business world. But with great power comes great responsibility. We’re talking about the risks in security, privacy, and governance that come bundled with it.

So, how do we minimize the risks associated with generative AI in enterprises?

1. Firm Boundaries: Rather than sending your data out to large language models (LLMs), bring the LLM to your data. Implement LLMs within the existing security and governance perimeter of your business. This way, your data stays secure while your team can make the most of these AI models.

2. Garbage in, garbage out: Customize AI models to be savvy about your business. Make use of options such as StarCoder from Hugging Face and StableLM from Stability AI to create domain-specific models that are fine-tuned to your business requirements. Fine-tuning can be effective in filtering out offensive responses that create further risk for businesses.

3. Surface Unstructured Data: Around 80% of the world’s data is unstructured—we're looking at emails, images, contracts, etc. Invest in data parsing pipelines to extract info from these sources and make them comprehensible for your AI models.

While these are all ways to maximize the potential of LLMs and minimize risk, enterprises should be cautious—only working with reputable vendors that can back their guarantees.

PUNCHLINES

Netflix and Chill: Netflix is offering up to $1 million for AI roles while writers are left out in the cold.

Smokin' Hot AI: California switches to AI for wildfire detection, but the system keeps setting off false alarms.

AI vs Cancer: UK startup Etcembly claims to have designed novel cancer immunotherapy using generative AI in record time.

Wireless wonder: A Waterloo-based company is using robotics to predict power outages before they occur.

Carrie on: Stephen King says he’s not afraid of AI, but wonders if it can ever write a story that makes people cry.

TRENDING TOOLS

📚 PromptFolder: Create, organize, and share prompts with advanced features

🦓 IndieZebra: A/B test your Product Hunt launch

📅 Vimcal: Schedule and organize events efficiently with AI

💬 Longshot: Upload documents, fact check, ensure zero hallucinations, and integrate anywhere

✍️ Yaara: Leverage AI to craft compelling, high-conversion copy

TLDR

Britain to host global AI safety summit at Bletchley Park: Britain plans to host the first-ever global summit on AI safety in November at the historic site of Bletchley Park, where Alan Turing cracked the Enigma code. The summit will bring together tech leaders, policymakers, and academics to address issues such as misinformation, warfare, and human rights.

The USAF’s A.I. Robot Wingmen for Aerial Combat: The USAF is developing a new generation of AI drones called collaborative combat aircraft, with plans to build 1,000 to 2,000 units for as little as $3 million apiece. These drones will serve various roles, including surveillance, resupply missions, attack swarms, and as “loyal wingmen” to human pilots.

Alibaba Cloud launches AI models for image and text understanding: Alibaba Cloud introduces two new open-source AI models, Qwen-VL and Qwen-VL-Chat, that are trained on Alibaba Cloud’s LLM Qwen-7B and can process both images and text. These models offer improved performance and applications compared to other vision language models and can generate captions, answer questions, and assist visually impaired users with shopping.

Bi-Touch: A new system for bimanual manipulation with touch: Researchers at the University of Bristol have developed an open-source system that enables robots to learn bimanual tasks with touch from a virtual helper. The system, called Bi-Touch, uses deep reinforcement learning and tactile sensors to control two robot arms. The robots can perform delicate and complex tasks, such as lifting a Pringle crisp or reorienting an object.

And that’s a wrap! If you have any questions or something interesting to share, please feel free to 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.

What did you think of today’s newsletter?

Feedback helps us improve!

Reply

or to participate.