Outsmart The Future
Today in Everyday AI
6 minute read
🎙 Daily Podcast Episode:Why is AI so biased? What causes LLMs to have those biases? We’re breaking down why they happen and how to fix them. Give it a listen.
🕵️♂️ Fresh Finds: A GPT-4 copilot for Google Drive, Big Tech wants AI regulation, and what AI architects fear most for 2024. Read on for Fresh Finds.
🗞 Byte Sized Daily AI News: Humane’s screenless mobile AI device, Nvidia's strategic AI chip move for China, and Hollywood strike ends with AI updates. For that and more, read on for Byte Sized News.
🚀 AI In 5: Looking for a quick and easy way to summarize and talk to webpages? We’re showing you how you can do it for free! See how here
🧠 Learn & Leveraging AI: So whether you’re developing LLMs or just using them, how can you avoid AI bias? We break it down. Keep reading for that!
↩️ Don’t miss out: Did you miss our last newsletter? We talked about The future of law with AI, Samsung launches GenAI model, and reviewing Midjourney's new Tune feature. Check it here!
How To Understand and Fix Biased AI 🤔
AI is a GREAT tool. But it has its issues.
Like why are AI models so biased?
Whether it’s ChatGPT or an AI image generator, LLMs often have certain biases and tendencies.
Nick Schmidt, Founder & CTO of SolasAI & BLDS, LLC, joins us to discuss how to understand and fix biased AI.
Also on the pod today:
• Algorithm misuse & discrimination 🚫
• Al fairness through regulations ⚖️
• 3 step process to avoid AI biases 🤖
It’ll be worth your 30 minutes:
Subscribe and listen on your favorite podcast platform
Upcoming Everyday AI Livestreams
Friday, November 10th at 7:30 am CST ⬇️
Here’s our favorite AI finds from across the web:
New AI Tool Spotlight – March Health is an AI powered reproductive health coach, Writingmate is a GPT-4 copilot for Google Drive, and AI Email Finder helps you source and verify emails in a single click.
Big Tech – Big Tech wants AI to be regulated. The rest of Silicon Valley worries that regulation will destroy the competition.
Business of AI - The AI wave seems to be taking off in France as in the last few months, AI has become the focus for many French entrepreneurs.
1. Humane's AI Pin: A Screenless Mobile Revolution 📱
Humane is set to launch the AI Pin, a new wearable device priced at $699, aiming to redefine mobile interaction. This screenless gadget, which clips onto clothing, integrates AI models from Microsoft and OpenAI, and operates on a unique version of T-Mobile's network. With a monthly subscription of $24, users get a phone number, data, cloud storage, and unlimited AI queries. The Pin features voice-based commands, gesture interactions, and a laser projector for displaying information.
2. Nvidia's Strategic AI Chip Move for China 🇨🇳
According to the Financial Times, a leaked report shows Nvidia has crafted three new chips specifically for the Chinese market, aligning with U.S. export controls. These chips, named H20, L20, and L2, are a response to recent U.S. restrictions on high-performance chip sales to China. Despite performance adjustments, these GPUs are anticipated to hold their ground in China's competitive tech landscape. Nvidia's pivot reflects its commitment to one of its key markets amidst the ongoing U.S.-China tech tensions.
3. AI Revolution in Animation 🎥
Jeffrey Katzenberg, DreamWorks co-founder, predicts a significant impact of AI on media and entertainment. Speaking at the Bloomberg New Economy Forum, he foresees a drastic 90% reduction in costs for producing animated blockbusters. AI could reduce the need for large teams of artists and lengthy production times. Katzenberg's insights suggest a digital transformation that will accelerate content creation.
4. Hollywood Strike Ends with AI in Focus 🎬
After a historic 118-day strike, Hollywood's SAG-AFTRA reached a deal addressing AI use and streaming residuals. The strike had actors demanding fair AI usage and residual payments. SAG's tentative agreement, expected to pass the union's board, marks a significant shift in industry standards. The deal follows intense negotiations over AI's role, with studios initially proposing unrestricted use of actors' digital likenesses. The final terms likely include protections for actors against AI misuse, reflecting the growing global discourse on AI regulation.
5. AI Drones Elevate Elephant Research 🐘
Colossal Biosciences collaborates with Save the Elephants, using drones and AI to revolutionize the study of African elephants. High-tech tools are now capturing detailed behavior, with drones flying high to avoid disturbing the natural activities of these majestic creatures. This innovative approach aims to predict and protect, potentially transforming conservation efforts.
Summarize and Chat with Webpages: Writingmate Review
Looking for a quick and easy way to summarize and talk to webpages?
We’re checking out Writingmate, an AI tool that lets you chat with and summarize different webpages.
🤷♂️ What’s Going On and Why It Matters:
So how are you supposed to actually feel good about using these AI models?
Because let’s be honest — AI models are super biased. They can reflect ugly stereotypes from the internet, which is what many of the most popular Large Language Models are based on.
So, how can we spot bias? And how can we fix it?
But here’s what we covered.
Examples of bias in image generators and language models
The role of biased information on the internet and societal biases influencing AI models
Practical advice for addressing AI bias, emphasizing fairness in input data and avoiding exclusion or collective punishment of groups
The potential for algorithmic decision-making to eventually reduce biases by recognizing and correcting human biases over time
Reasons for bias in AI models, including biased training data, discriminatory outcomes, and misuse of algorithms
The role of AI models in reviewing massive data sets (like credit models) to avoid discrimination
Promoting the use of AI to mitigate risks and emphasizing speed and efficiency for fixing biases
Now, let’s break it down y’all.
🦾How You Can Leverage:
Fact: AI models are biased.
Also fact: It’s gonna take a loonng time for Big Tech companies producing these models to fix the issue, mainly because the datasets they’re based off are often biased. And that’s kinda a big problem.
Were we the only ones kinda shook by what we learned today? Like… using AI to correct AI bias? Kinda meta, right?
To be honest, this is one of those conversations that really require many hours to truly understand. (And if you wanna go into a lil deeper dive with Nick, you can check out this paper on Fair Lending in Machine Learning that he co-authored)
Let’s talk about this massive problem and create a somewhat digestible action plan.
1 – Keep a watchful eye 👁
You know what’s almost as bad as AI models being loaded with bias, stereotypes and straight up nastiness? The fact that the hundreds of millions of people using LLMs are just blindly using their outputs without even seeking out bias. (And then, we actually retain that bias afterwards. Yikes.)
2 – Check your models and bias 🧐
If you’re a bit on the techy side, maybe you’re helping your company decide on an AI model to use. Or maybe you’re tapping into an open source model yourself.
If so, Nick had some great advice.
3 – Less garbage in, less garbage out 🚮
If you’re an individual user, there’s not much you can do for the problem at large. Like, a strongly worded letter to OpenAI or Midjourney probably won’t help. (Maybe have that letter be written by a human, at least?) But, there are things individuals can do to cut down on biased outputs, whether you’re using a program like ChatGPT for content or Midjourney for imagery.
Try this: Improve your inputs. General, open-ended prompts are much more likely to produce bias. If you instruct models with specific, context-rich prompts, you’ll cut down on the likelihood of bias, stereotypes and hallucinations. For instance, when using an AI for content creation, include directives that emphasize diversity and avoid stereotypes. Yeah, sometimes you gotta tell models to be more inclusive and less divisive. Weird, but it helps!
Bonus — While we’re on the topic, go check out former Everyday AI guest Leo Rodman’s free collection of diverse clipart. Save yourself some time and bring a little less bias to your next project!
Now This …
Vote to view the live results