🧠 Safe superintelligence

🧵 In today's edition:

• 🧠 Safe superintelligence

• 🍦 Bacon ice cream

• 🤵 Salesforce unveils groundbreaking GenAI benchmark for CRM

• 🛠️ AI Tools to Check Out

• 🤑 AI Fundraising News

• 🐦 Tweet Post of the Day

🧠 Safe superintelligence

What was that quote? 

You either die an OpenAI employee or live long enough to start your own AI company…

Ok, not exactly - but I think it applies in this context. 

Ilya Sutskever, Co-founder of OpenAI and one of the world's most respected AI researchers, announced the launch of a new AI start-up called Safe Superintelligence (SSI) Inc. 

This announcement comes just a month after Sutskever's dramatic departure from OpenAI following an unsuccessful attempt to oust CEO Sam Altman.

Here’s the tea:

• As the name suggests, SSI's sole focus is to develop "safe superintelligence" – a type of machine intelligence that could surpass human cognitive abilities. 

• Sutskever has co-founded the company with former OpenAI employee Daniel Levy and AI investor and entrepreneur Daniel Gross.

In a statement released on X, the founders emphasized that SSI's "singular focus means no distraction by management overhead or product cycles, and our business model means safety, security, and progress are all insulated from short-term commercial pressures." 

Overall: Sutskever's move to launch a rival AI start-up dedicated to safe superintelligence comes after a period of turmoil at OpenAI, where clashes over leadership direction and safety concerns led to his attempted ousting of Altman. 

With SSI, Sutskever aims to develop advanced AI systems that prioritize safety and benefit humanity.

Read more: Financial Times

🍦 Bacon ice cream

In a hilarious turn of events, McDonald's has removed its AI-powered voice ordering technology from drive-through restaurants across the US after customers shared countless comical mishaps online. 

IBM developed and announced the system in 2019. It was supposed to streamline the ordering process with voice recognition software, but the AI had other plans.

What happened? 

• Social media was flooded with viral videos of the AI gone rogue, misinterpreting orders in the most bizarre ways imaginable. 

• From topping ice cream with bacon to ordering hundreds of dollars worth of chicken nuggets, the AI's mishaps were both cringe-worthy and hilarious.

• One video shows a customer's frustration as the AI repeatedly adds butter to her order instead of caramel ice cream.

• The AI even confused a request for a caramel sundae with an order for a bacon-topped ice cream.

Overall: After this trial run turned into a comedy of errors, McDonald's has decided to discontinue the AI ordering tech by the end of July.

The fast-food giant stated that it remains confident in the future of such technology but will re-evaluate "long-term, scalable solutions" for a voice ordering system by the end of the year.

While the trial's failure may have provided some comic relief, it also serves as a reminder that replacing human workers with AI isn't always as straightforward as it seems, at least not without a few hiccups (or, in this case, a few hundred nuggets) along the way.

Read more: BBC

🤵 Salesforce unveils groundbreaking GenAI benchmark for CRM

Salesforce introduced a first-of-its-kind GenAI benchmark designed explicitly for CRM applications. 

The goal? To empower businesses to make informed decisions when choosing LLMs for their CRM needs.

According to Clara Shih, CEO of Salesforce AI, customers don't just want the best model; they want one that aligns with their security standards, compliance requirements, and budget constraints. 

The benchmark addresses this by evaluating LLMs based on four key metrics: accuracy, cost, speed, and trust and safety.

Unlike existing academic benchmarks, the Salesforce GenAI benchmark for CRM is tailored to real-world business scenarios. It leverages actual CRM data and human expertise to comprehensively assess the practical utility of AI in various CRM use cases, such as sales and service scenarios.

"The interesting aspect of this study is that the answer isn't always the biggest model," says Silvio Savarese, EVP and Chief Scientist of Salesforce Research. "You can actually obtain very satisfactory performance by using models that are smaller and more cost-effective from the cost and latency perspective."

Overall: This benchmark represents a significant step towards empowering businesses to optimize their AI technology stack for CRM applications. 

Salesforce plans to expand the benchmark further by incorporating additional metrics, use cases, and data and evaluating the performance of fine-tuned models on CRM data.

Read more: CIO

🛠️ AI Tools to Check Out

• Lummi: The best free stock photos and royalty-free images. Powered by robots everywhere. Check it out!

• Strut: Capture notes, organize projects, and collaborate with your team alongside AI. Check it out!

• HiTalk: An AI language learning platform that offers stages of vocabulary, phrases, daily dialogues, writing, and downloadable exercises. Check it out!

🤑 AI Fundraising News

• Materia raises $6.3M in funding to provide a comprehensive generative AI platform for public accounting firms.

• Substrate raises $8M in Seed funding to provide an API for modular AI, enabling users to build fast multi-step AI programs using optimized models, colocated file and vector storage, a code interpreter, and logical control flow.

🐦 Tweet Post of the Day

Cool demo on how good Claude is 👇️