​STAR WARS AT PEACETIME (13 IMAGES)​

​

APRIL 17, 2024

🤖 Innovation Dispatch: Your Daily Dose of AI

Read Time: 6 Minutes

Welcome back, Innovators!

​AI hallucinations are 'easily solvable'; Big Tech Can't Agree on What 'Responsible' Means​​​​​​​​​​​​​; AI girlfriends will be a $1B industry; Star Wars at Peacetime (13 images); plus, more headlines and tools!

🎯 Today's Innovations

• ​AI hallucinations are 'easily solvable'
• QUESTION OF THE DAY
• AI Makers Can't Agree on What 'Responsible' Means​​​​​​​​​​​​​
• AI girlfriends will be a $1B industry, tech expert says​; more headlines for April 17, 2024
• AI Word of the Day: Data Augmentation
• 5 New AI Tools for Work and Everyday Life

FEATURE

🤖

​AI hallucinations are 'easily solvable'

At Nvidia's GTC developer conference, CEO Jensen Huang addressed two key AI topics - the challenge of AI hallucinations and predictions on achieving artificial general intelligence (AGI).

On AI Hallucinations:

• Huang expressed frustration with AI systems making up plausible but factually inaccurate answers, a phenomenon known as AI hallucinations.
• He suggested hallucinations are "solvable easily" by requiring AI systems to research answers from credible sources before generating responses.
• For critical domains like health, he proposed cross-checking multiple verified sources before providing an answer.
• Huang advocated for AI to have the ability to state "I don't know" when it cannot find a factual consensus.

On Timeline for AGI:

• Huang said predictions depend on how AGI is specifically defined and measured.
• If AGI means performing 8% better than humans on tests like bar exams or medical exams, he believes it could be achieved within 5 years.
• However, he declined giving a broader timeline unless the criteria for AGI capabilities are clearly specified.

​CHECK OUT MORE HERE FROM TECHCRUNCH​

​Q: Have you tried any AI other than ChatGPT?​​

TIDBIT

AI Makers Can't Agree on What 'Responsible' Means​​​​​​​​​​​​​

AI developers lack agreement on how to test whether their models behave responsibly, according to Stanford's latest AI Index released Monday. This makes it difficult for businesses and users to compare AI providers and choose ones aligned with their values.

While tests like TruthfulQA assess honest responses and RealToxicityPrompts check for hate speech, developers use different benchmarks inconsistently. For example, OpenAI, Meta and Anthropic use TruthfulQA, but Google and Mistral do not. Meta benchmarks against three tests, while Mistral uses none assessed by Stanford.

The lack of standardization may stem from cherry-picking flattering benchmarks or obscuring model limitations. However, motivations are unclear, according to Stanford's Nestor Maslej. A new Responsible AI Institute supported by major firms aims to address this gap with new benchmarking tools.

Wider debates continue around prioritizing immediate AI risks like biased outputs versus future existential threats from advanced systems. The report also found the U.S. leads in significant AI models and private investment, while China tops AI patents.

​CHECK OUT MORE FROM THE STANFORD STUDY HERE​

📰 Headlines

1. ​Microsoft invests $1.5 billion in AI firm G42 which faces U.S. scrutiny for China ties​​
2. ​Tech exec predicts ‘AI girlfriends’ will create $1B business: ‘Comfort at the end of the day’​​
3. ​Corporate Investment in AI Saw a Significant Downturn Last Year​
4. ​The State of AI in 13 Charts​
5. ​China's Baidu: Ernie Bot has amassed 200 million users​
6. ​"Top secret" is no longer the key to good intel in an AI world: report​​
7. ​DeepMind CEO Says Google Will Spend More Than $100 Billion on AI​
8. ​AMD Unveils New AI Chips For Business PCs​​
9. ​I finally found a practical use for AI and I may never garden the same way again​
10. ​Maestro: Amazon's AI playlist generator aims to dethrone Spotify​
11. ​Meta oversight board reviews the handling of AI-created 'celebrity porn'​
12. ​Child sexual abuse content grows online with AI-made images, report says​​
13. ​GovDash aims to help businesses use AI to land government contracts​

AI Word of the Day: Data Augmentation

Data Augmentation is a technique used in machine learning to make more data for training computers. It's like taking a photo and then making new versions of it by turning it, flipping it, or zooming in.

This helps computers learn better from pictures or words by showing them many different examples.

Why It Matters

Data augmentation is really important for making computers smart. It helps them understand pictures or words in different ways.

This is super useful when we don't have a lot of data to start with. By making the data we have go further, computers can get better at their jobs, like recognizing what a picture shows or deciding if an email is spam.

Real-World Example

Imagine we have a bunch of photos, and we want a computer to find all the ones that have dogs in them. Data augmentation helps the computer pick out things like the shape of ears, the color of fur, or the size of a tail that are important for recognizing a dog.

This way, the computer can quickly say, "Yes, this photo has a dog" or "No, there's no dog here," just like if you were looking for a friend in a crowd by remembering their bright yellow hat.

📚 AI TOOLS

• ​Deckee: text-input AI tool that creates websites, digital tokens, and other content from a description.
• ​Limitless: we covered the wearable AI assistant yesterday. Aside from the device, the service transcribes meetings, summarizes conversations, and creates to-do lists. Users can access their data forever through the Limitless Confidential Cloud, which uses encryption to protect privacy.
• ​Growth Makers: AI tool that uses AI agents to find marketing strategies to grow your business quickly. The AI agents are trained in marketing skills and growth hacking strategies. They can analyze data to find unexpected channels for growth, create content, and research keywords. They can also predict customer behavior and stay up-to-date on trends.
• ​August: 24/7 health expert powered by AI that has scored 94.8% in the US Medical Licensing Examination.
• ​Calesitai: an AI assistant for content marketing that creates social media posts, ad copy, cold emails, and more. The AI assistant can write in your brand voice. It can also edit existing articles.

​Read All Innovation Dispatch Back Issues Here​

Your input fuels the content. Let me know how I did today!

​Email me your thoughts, or just tell me whether you thought it was:

Great ❤️
Average 😐
Sucked 😞

Also, let me know what you would like to see more of. Have a great rest of your weekend!