AI Insider No. 38

Hola, AI Insiders! Every time a new thing drops that’s half a smidge faster or a tiny bit bigger than the current market leader, headlines start blaring that it’s a “[fill in the blank] killer!” This week, it’s chatbot Claude’s turn. Apparently, the just-launched Claude 3 is a “ChatGPT killer!” Right. Claude’s benchmark numbers are good, but in some cases, just minimally better than ChatGPT 4. And frankly, getting the attention of users with “ChatGPT on the brain” is a tough sell at this point. Read on to find out what you can expect from Claude 3 if you bother to take a look (and you should!). And we promise not to use the word “killer.”


(Screenshot)

Anthropic Unveils Upgraded Chatbot, Claude 3

By Perplexity for AI Insider

Anthropic has released Claude 3,  its updated large language model that the company claims set new industry benchmarks with “near-human” capability, outperforming OpenAI’s GPT-4. 

Claude 3 is Anthropic’s first multimodal AI. It can analyze up to 20 images in a single request, though it cannot produce images on its own. Microsoft’s Copilot and Google’s Gemini have that capability. 

Claude 3 comes in three versions: Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus. Opus is the most powerful, and Haiku is the most “compact” and fastest.

The models will soon cite the source of their responses for users to verify. However, unlike some other generative AI rivals, Claude 3 cannot search the web and can only answer questions using data from before August 2023.

Anthropic co-founder Daniela Amodei told CNBC that the new chatbot has the ability to summarize up to about 150,000 words, about the length of the book Moby Dick, compared to ChatGPT 3.5, which clocks about 3,000.

Users can upload photos, charts, handwritten text, and other types of unstructured data for analysis and queries, extract text from images, and more. The company declined to specify how long it took to train Claude 3 or how much it cost.

Claude 3 Opus beat GPT-4 on 10 AI benchmarks, although some of the wins are narrow. No other model has beaten GPT-4 on such a range of benchmarks before. Opus outperformed GPT-4 and Google’s Gemini Ultra on undergraduate-level knowledge, graduate-level reasoning, and basic mathematics.

The mid-tier Sonnet is accessible for free, and the Opus is subscription-only and priced similarly to ChatGPT Plus at $20 per month.  Sonnet and Opus are available in 159 countries. Haiku will be released later this year, according to Anthropic. 


OpenAI Claps Back at Musk

By Michelle Johnson, AI Insider

In this week’s episode of The Hellscape formerly known as Twitter vs. OpenAI (ChatGPT): OpenAI claps back at Elon Musk and his lawsuit alleging breach of contract. And OpenAI showed up with receipts in the form of email screenshots!

In a blog posted on Tuesday, OpenAI made one thing clear: “We intend to move to dismiss all of Elon’s claims.”

You may recall that they were once besties and then it went bad. OpenAI described it this way:  “We’re sad that it’s come to this with someone whom we’ve deeply admired—someone who inspired us to aim higher, then told us we would fail, started a competitor, and then sued us when we started making meaningful progress towards OpenAI’s mission without him.”

Now, let’s get to the receipts. This email shows that Musk threw some shade in 2018, predicting that OpenAI had zero chance of succeeding.:

As we all know, OpenAI’s ChatGPT rocked the planet, so the failure here appears to be Musk’s forecast.

Next receipt: At the heart of his lawsuit is the charge that OpenAI strayed from its mission to develop AI for the good of humanity by going for-profit. Musk estimated that OpenAI would have to raise at least $1 billion. OpenAI later formed a for-profit arm to start pursuing those big bucks.

Musk is squawking that OpenAI looks more like “ClosedAI” because they shifted too much focus away from being open source. However, in 2018, Musk responded with the word “Yup” to an email from OpenAI co-founder and Chief Scientist Ilya Sutskever, noting that “As we get closer to building AI, it will make sense to start being less open.”

The kicker: Some folks are saying that Musk basically admitted that this is just a nuisance lawsuit when he posted a cheeky response to the blog post saying that he would drop the lawsuit if they changed their name to ClosedAI.


Experiment Turns Out Fake Election Imagery

By Perplexity for AI Insider

A recent experiment conducted by the Center for Countering Digital Hate (CCDH) has unveiled loopholes in the creation of fake election-related imagery using artificial intelligence tools. 

Despite stringent rules set by AI companies to curb the generation of misleading content, the CCDH’s test had a 41% success rate.

The CCDH evaluated four popular AI image generation platforms: Midjourney, OpenAI’s ChatGPT Plus (Dall-E), Stability.ai’s DreamStudio, and Microsoft’s Image Creator.

A nonprofit organization dedicated to mitigating digital hate and misinformation, the CCDH embarked on the investigation to highlight the vulnerabilities present in the current framework of AI-generated content. 

Utilizing the most popular AI tools, which explicitly prohibit the creation of deceptive images, the researchers systematically attempted to produce election-related deepfakes. Astonishingly, nearly half of their attempts were not flagged or stopped by the platforms’ existing preventive measures.

“This finding is a stark indication that the measures in place are insufficient in the face of evolving AI capabilities,” said Imran Ahmed, CEO of the CCDH. “The ease with which our team was able to create convincing, yet entirely fictitious, political imagery calls for an immediate reevaluation of content moderation practices.”


Perplexity AI Pro allows users to switch between chatbot models. (Screenshot)

Off The Beaten ChatGPT Track
A Look at Perplexity AI

[Ed note: This occasional feature is meant to encourage everyone to step away from ChatGPT and explore the wider world of AI from time to time.]

By ChatGPT and Mistral for AI Insider

Hey there, AI aficionados! 

Welcome to a whirlwind tour of Perplexity AI, your new digital BFF that’s about to make Google look like your grandpa’s dusty old encyclopedia. So, what’s the big deal with Perplexity’s “Answer Engine,” and how does it stand out in the sea of chatbots? Let’s dive in!

Perplexity AI: Not Your Average Chatbot

Picture this: Instead of typing in a keyword-laced search request and getting back a zillion links to sift through, Perplexity AI serves up straight-up answers like a know-it-all (in a good way). 

It’s like having a chat with your super-smart friend who can explain quantum physics, recommend the best pizza in town, and tell you a joke, all without breaking a sweat. This “Answer Engine” is all about getting to the point, minus the info overload.

Users can choose a focus to fine-tune responses to prompts. (Screenshot)

Features That’ll Make You Go ‘Wow!’

Models Galore: Whether you’re a fan of the wise and seasoned ChatGPT 4 or the fresh-faced Claude 3 and Mistral, Perplexity’s got a model for every mood. When you enter a prompt, you can pick your model, including Perplexity’s. Each model has its own strengths and capabilities, so you can choose the one that best suits your needs.

Focus Settings: Tailor your chat experience by specifying settings like Writing, Academic, and more to get the vibe just right.

Image Wizardry: Need a picture of a cat in a spacesuit? Perplexity can whip one up for you without you having to open another browser tab and call up an image generation site. Perplexity has one provided by Playground AI built in.

YouTube Sleuth: Can you find that one video with the thingamajig you couldn’t remember? Perplexity’s on the case. It searches YouTube videos for you.

Searchable, Organized Threads: Good luck finding that chat you had with ChatGPT a month ago. There’s no easy way to search for it. Perplexity not only has a search for your threads, you can organize them into “collections,” basically folders that you can label. Whether you’re juggling multiple projects, diving deep into various topics, or simply want to keep your thoughts structured, Collections has got your back. 

And More: From planning trips to solving riddles, Perplexity’s your go-to for just about anything. Start a chat, Perplexity will give you an answer and include links to related info if you want to go deeper.

Go Free or Go Pro?

Navigating the waters between the free and Pro versions of Perplexity AI can feel like choosing between a cozy rowboat and a sleek yacht. Both will get you across the digital sea, but oh, the difference in the journey!

The free version of Perplexity AI is like your trusty rowboat. It’s equipped with the essentials, including access to OpenAI’s ChatGPT model, allowing users to ask questions, get answers, and even upload a limited number of files for the model to analyze. It’s a fantastic option for those dipping their toes into the AI waters, offering a taste of what AI can do without any financial commitment.

On the other hand, the Pro version is the yacht, offering a luxurious and more powerful experience. For $20 a month or $200 a year, Pro users unlock the full potential of Perplexity AI. This includes unlimited copilot usage, access to advanced models like ChatGPT-4 Turbo, Claude 3, and more, plus the ability to upload up to 100 files daily.

 Pro users also enjoy dedicated support, including direct communication via Discord and email for quicker assistance.

The choice between the free and Pro versions ultimately depends on your needs. If you’re looking for a simple, no-cost way to explore AI’s capabilities, the free version is a great starting point. However, if you’re diving deep into research, writing, or development and crave more power, flexibility, and support, the Pro version is your go-to.

Prompt Examples and Use Cases

Homework Helper: “Hey Perplexity, can you explain the symbolism in ‘The Great Gatsby’?”

Culinary Consultant: “What’s a simple recipe for a chocolate lava cake?”

Travel Agent: “Plan a weekend getaway from Boston with a focus on history and good food.”

Artist: “Generate an image of a Victorian house with a modern twist.”

Prompt Assist: “Generate 5 prompts that I can use to learn more about the future of AI.”

Academic Aide: “Summarize the main arguments in this academic paper on AI ethics.”

Practice Makes Perfect

Ready to take Perplexity for a spin? Here’s a fun exercise to get you started:

1. Open up Perplexity AI and say hello (it’s polite!).

2. Ask it to write a short poem about your favorite city.

3. Now, switch gears and ask for the latest tech news highlights.

4. Feeling adventurous? Request an image of a robot having a tea party.

As you play around, you’ll discover that Perplexity is more than just a tool; it’s a playground for your curiosity. So go ahead, ask away, and watch as it turns your questions into a cascade of answers.

Don’t forget to share your coolest finds with AI Insider!

Keep being awesome,
Your AI Pals, ChatGPT and Mistral


This image has an empty alt attribute; its file name is Firefly-question-marks-rising-from-flames-94723-1-300x300.jpg

What the Actual Hell?
Did a Thieving Stability AI Employee Bring Down Midjourney’s Servers?

On Wednesdays, David Holz, CEO of the high-end AI image generator Midjourney, hops onto their Discord server to do “office hours.” Usually, he updates Midjourney fans about how the latest upgrades are progressing, then answers questions from the floor.

This week, he spun a tale of intrigue involving one of their competitors: Stability AI. 

Holz said that they noticed something odd over the weekend, namely that a paid user who looked to be someone from Stability was trying to suck down a massive number of prompts and images. Like a lot of them. Enough to bring down Midjourney’s servers, running the weekend for their engineers who had to clean up the mess and get the service back up in the middle of the night.

Tisk, tisk. Then the CEO announced that this was not cool and the penalty would be that ALL Stability employees are now banned from Midjourney. Accounts canceled. He took the interlopers to task for ruining the fun, making their already hard-working engineers work over the weekend. “Just let people make pictures,” he said. We agree.


Yet another text-to-video generator: Haiper, a new AI video generator, has made its debut, offering users the ability to create videos from text prompts, animate static images, and repaint existing videos with a fresh look. Developed by former Google DeepMind researchers, Haiper generates super short (2-second) video clips and is currently available for free.

CNBC: Microsoft engineer warns company’s AI tool creates violent, sexual images, ignores copyrights

AP: Former Google engineer charged with stealing AI secrets for Chinese companies

Nieman Lab: Google wants to redirect traffic from AI-generated spam blogs to legit websites

Reuters: AI-powered LinkedIn’s premium subscriptions have raked in $1.7 billion

MS Power User: ChatGPT now has multifactor identification to secure your account

TechCrunch: Apple CEO says the company will break new ground in AI this year.


Aht Gallery

And now for something completely different. I’m often asked which chatbot is the best. There’s no real answer to that question because they each have strengths and weaknesses. One feature that’s now built into most bots is “vision,” meaning the ability to “see” and analyze an uploaded image.

Just for kicks this week, I uploaded a photo of a device that hasn’t even hit the market yet and asked a few bots, “What is this?” Only one got it 100 percent right, and one misidentified the logo as the Playboy bunny. (Note: Click to enlarge! Each slide stays up for 10 seconds so that you have time to read. Wait for it….)

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.