This week had lots of little stories. We’ll be talking about big Nvidia AI announcements, ChatGPT gets a bunch of upgrades for everyone, Leonardo.ai makes enormous gains on Midjourney, Apple is building AI into every product, more evil AI launches, the Elon vs Zuck fight is on, and so much more.
ChatGPT Updates
First, ChatGPT is now getting up-to-date information on its models. Where before, you would be told that ChatGPT has a knowledge cutoff date of Sept 2021, now you’ll have access to current information. ChatGPT previously had web browsing, which pretty much accomplished the same thing, but it was taken down and has been offline for a while. It might have been too dangerous to give users the ability to direct ChatGPT to browse websites, and this could be the replacement for that functionality. Now, ChatGPT will gather the information first and allow the OpenAI team to filter and polish that data before presenting it to users.
Some facts about the GPTBot:
GPTBot filters out paywall-restricted and policy-violating sources but allows site owners to choose whether to grant access
Website owners can restrict GPTBot by modifying their robots.txt file
GPTBot is looking to solve issues around copyright infringement
ChatGPT Custom Instructions
ChatGPT custom instructions were rolled out a few weeks ago to paying users and are now available to all users. If you don’t remember, custom instructions allow users to give ChatGPT initial directions, so you don’t have to start from scratch each time you use it.
One of the most popular ways to use this feature is by telling ChatGPT that you already know it’s a large language model because it would say to you that every chance it got previously.
ChatGPT is also rolling out some quality-of-life improvements:
Example prompts: you can see some examples of popular prompts for ideas.
Suggested replies: ChatGPT automatically synthesizes follow-up questions.
GPT-4 by default: When starting a new chat as a Plus user, ChatGPT will remember your previously selected model
Uploading multiple files is now supported in the Code Interpreter beta for all Plus users
Stay logged in: You’ll no longer be logged out every two weeks
Keyboard shortcuts: Work faster with shortcuts. Try ⌘ (Ctrl) + / to see the complete list
AI Password Cracking
Next, it seems like AI is being used for evil each week. Now, AI can record your keystrokes by just listening to them. Tapping in a computer password while chatting over Zoom could open the door to a cyber-attack, research suggests, after a study revealed artificial intelligence (AI) could determine which keys are being pressed by listening to the sound of typing with more than 90% accuracy.
“I can only see the accuracy of such models and such attacks increasing,” said Dr Ehsan Toreini. The researchers added there are several ways to reduce your risk of these types of attacks, including opting for biometric passwords or activating 2FA systems. Surprisingly, they also say using the shift key to create a mixture of upper and lower cases, or numbers and symbols, in your password is a good idea. “It’s very hard to work out when someone lets go of a shift key,” said Harrison.
Nvidia Launches New Chip
Nvidia attended a big event called Siggraph this week, where they launched several AI products, including a new chip that will significantly drop the price of training LLMs. Nvidia’s GH200 has the same GPU as the H100, Nvidia’s current highest-end AI chip, but pairs it with 141 gigabytes of cutting-edge memory and a 72-core ARM central processor. This new chip should significantly lower the cost of training models.
Nvidia x HuggingFace
Nvidia is also partnering with HuggingFace, the AI model repository, to offer top-of-the-line GPUs to millions to help train and fine-tune models easily. Named Nvidia GDX cloud, this AI supercomputing cluster features eight H100s or A100s, 80GB Tensor Core GPUs for a total of 640GB of GPU memory per node.
One of the most significant issues facing AI developers today is the severe lack of GPUs available, with much of the inventory being sold out through the end of the year. This will be a huge help to ease that bottleneck.
Disney AI
Amid one of the most significant writer’s strikes in history, where their biggest concern was their jobs being taken by AI, Disney created a task force to explore how to use AI across their entire business. Launched earlier this year, and before the writer’s strike, according to Disney, the group is looking to develop AI applications in-house and form partnerships with startups. Disney currently has 11 job openings seeking AI and ML experts.
Disney had multiple enormous flops this year, including the new Indiana Jones installment, and now they want AI to help them bring costs down on mega-budget movies to help de-risk themselves. Disney had 100 artists spend three years de-aging Harrison Ford for just a few minutes of film in Indiana Jones, showing AI's potential to bring down costs.
But honestly, bringing down the costs wouldn’t have put more butts in movie theaters. People are showing they want new IP, like Barbenhiemer.
Apple AI Plans
Next, Apple is putting AI into virtually every product, as its CEO Tim Cook responded to growing questions over its AI strategy. “We view AI and ML as fundamental core technologies. And they are virtually embedded in every product that we build,” Cook told CNBC in an interview.
This is following Apple’s lack of large, specific AI launches, which virtually every one of their competitors has had. In true Apple fashion of waiting until a market matures before entering, they’re waiting until they can make AI “the Apple way” before launching it.
Leonardo.ai Photoreal
Next, Leonardo.ai had two big launches this week. First, they made huge gains on Midjourney by launching their PhotoReal Update, and it’s stunning. Photoreal is precisely what it sounds like. It is a gen AI pipeline that allows for creating impressive photo-realistic creations. Enable it, set your depth of field preference, and then you can choose from presets such as cinematic, creative, and vibrant.
Photoreal is only available for premium users right now.
Leonardo.ai iOS App
Leonardo.ai also launched an iOS app that allows you to create incredible AI generative art from your phone easily. Most features from the web app version are available in the iOS app.
I’ve been a big fan of Leonardo.ai for a while, with one of my first videos ever being about them when they launched. They’ve always trailed Midjourney in quality, but now, with their photoreal update, iOS launch, and my preference of not using Discord as the interface, they are a great alternative to Midjouney.
Realtime Voice Cloning
Elevenlabs, the company behind incredible voice-cloning technology, has now released real-time voice generation. Where previously you had to wait until you could convert text into voice, you can now do it in real-time, which opens up incredible possibilities, such as giving AIs voices as quickly as they can generate text. This brings us closer to real verbal conversations with AI models.
WizardLM 70B 1.0
WizardLM released their newest model, WizardLM 70B 1.0, which substantially improves coding, mathematical reasoning, and open-domain conversation capacities. I’m waiting for an open-source model that can rival GPT4 in coding abilities, and this could be it.
Google IDX
Google has launched IDX, a developer product that looks to compete directly with Replit. IDX enables development entirely in the cloud, which Replit popularized. They are also building AI code generation using the Palm2 model, which has been shown to work very well as a coding assistant. IDX will support a variety of languages at launch, but other than that, there aren’t many details yet. If you want to learn more or sign up for their waitlist, visit https://idx.dev/.
Musk vs. Zuck
This isn’t necessarily about AI but rather about two billionaires who happen to run AI-focused companies, Elon Musk and Mark Zuckerberg. A few months ago, these two tech titans challenged each other publicly to an MMA fight. While most people thought this was fake, there might be a glimmer of truth.
Musk and Zuck have gone back and forth, taking jabs at each other online and talking about potential details of their fight. Then, it was revealed that Zuck has been training in jiu-jitsu for a while and recently won a BJJ tournament.
And while Musk was publicly saying he wanted to fight, it seems Zuck called his bluff both privately and publicly saying that he offered dates for the fight, but Musk has not yet confirmed, and Zuck was not “holding his breath.” Musk started talking about health issues and other things, backtracking his blustering in a truly Musk way. I genuinely hope this fight happens because no matter who wins, the world will win by watching one of them getting their butt kicked.
