Howdy 👋🏾, major tech cities faced a sudden shortage of popcorn as everyone watched intently as Sam Altman, CEO, Cofounder, and face of OpenAI, was unceremoniously fired over a Google Meets call by what is quickly being called a coup by a faction of the board.
You heard correctly: the guy who is the poster boy for a company that made the word AI regular vocabulary in worldwide households was fired with little more than a blog post that implied he lied.
After over 500 of the companies’ 770 employees threatened to quit if Sam Altman was not reinstated, the board spent a weekend negotiating over McDonald’s and Boba Tea, ending with an announcement that Sam Altman would become CEO of a new AI division at Microsoft and the former CEO of Twitch would become the third CEO of OpenAI in just three days.
Many see this as a massive win for Microsoft, who may acquire all of OpenAI for essentially $0, but if you’re a user of OpenAI’s products, this is a huge wake-up call. All those months spent integrating OpenAI’s tools into your products feel risky with so much instability. I’ve spent the last week working with the new Assistant APIs and fine-tuning frameworks, but if the board wins and slows down the speed of commercialization, the future of OpenAI’s products and roadmap is very much in question. So I thought, what better time than now to talk about all the other great AI models out there, especially if you realize some level of AI redundancy might be necessary.
The AI stack consists of a few layers:
👉 The silicon and GPUs provide all the processing power for these large language models (LLM) that are needed to train on tons and tons of data. The big winner here is Nvidia, which continues to see its sales double as it becomes the biggest supplier of AI chips.
👉 The cloud is the next part of the stack, and for many, the cost of the hardware needed to train an LLM makes it cost-prohibitive. The big cloud providers AWS, Google Cloud Platform, and Microsoft Azure allow us to tap into these powerful models and only pay for the necessary compute time.
👉 The foundation models are the actual AI models, and we use the word foundation because these models are trained on general or narrow data. Still, we can train these models by injecting them with our proprietary data to make these models better for our use cases.
👉 The applications that use these foundational models be it custom products you build or an off-the-shelf SaaS solution that uses one of these foundational models behind the scenes.
OpenAI’s ChatGPT is one of these foundation models, but several companies have popped up with competitive models, and you might be surprised to learn that some of these models may be better and cheaper for you, depending on your use case. That’s why it’s important not to pick a model blindly and to go through a selection process to identify which model or models best fit your use case. These are a few you can check out and explore:
📌 Anthropic is an AI company founded by 11 former employees from OpenAI focusing on creating a safer AI model with fewer hallucinations. I’m a big fan of Claude 2, its newest AI model, and in my experience, it tends to give better answers than ChatGPT but is not trained on as much data. You have to apply for API or enterprise access, but try the chat interface; it’s free.
📌 AI21Labs Jurassic-2 is a multi-model solution that can respond with a combination of text, images, and audio from one prompt. Their AI tools also offer some great solutions for moderation and grammar correction.
📌 Cohere’s Command model is trained to be a business-friendly model introduced as the perfect model for internal business operations and tasks. This excels at writing press releases, summarizing financial reports, or writing work emails.
📌 Meta’s LLama2 is a powerful AI model that’s great at being conversational due to the loads of social data Facebook and Instagram bring to the table. Meta also has a very impressive AI model for generating code that I’m continuously surprised and delighted by.
This list is by no means complete. Companies with new and exciting AI models are popping up every day. I regularly check Hugging Face to explore new models trained for all tasks. It’s also worth reminding you all that this is what Mindgrub and I do, and we’re always happy to consult and advise on the best AI model for whatever the job might be. Now, here are some non-OpenAI thoughts on tech & things:
⚡️The next leap in telehealth is the CarePod, a diagnostics pod that could allow your doctor to run a complete set of tests from anywhere in the world. The technology makes doctors more scalable and potentially solves shortages in rural areas.
⚡️ Apple made a massive U-turn by announcing it will support RCS next year, the next generation of texting, making it suck less when you text an Android user. The thing is, texts will still appear green, not support end-to-end encryption, and miss out on many other features that make iMessages great, but at least this will allow iPhone and Android users to send each other photos or videos.
⚡️ We could use Google Search not because it’s a monopoly but because it’s the best product out there. Techdirt explores if Google is the best, but yours truly has been found on Duck Duck Go.
⚡️ EV Chargers continue to be a problem. Joanna Stern has an excellent article in the Wall Street Journal. In it, Stern chronicles 48 hours in a Rivian R1T, stopping at over 120 non-Tesla EV chargers, and 40% had issues. Stern categorizes the problems into three categories: chargers out of order, chargers that failed to accept payments, and chargers that experienced “handshake” issues when communicating between the charger and EV.
⚡️ I travel a lot for work, and I found myself emphatically agreeing with this great article on hotel showers and how they somehow suck.
I’m headed down South to spend time with my family for Thanksgiving, and I can’t wait to see how the Tesla fairs on a road trip. Wish me luck! I’m thankful for all of you, and I hope you have an amazing and wonderful Thanksgiving!
p.s. What’s crazy about all this OpenAI drama is that the company has had 3 CEOs in less than a week, and the dust still hasn’t settled. As I write this newsletter for tomorrow, The Verge reports that Sam Altman, now CEO of Microsoft’s AI division, is STILL trying to get his old job back. This state of chaos may drag through the holidays into next week! At least I should have tons of emergency podcasts to listen to on my drive!