NOTE: This is a republication of the latest issue of our weekly newsletter. To subscribe, click here.
Last week was big for A.I. updates. Although…so was the prior week. And, well, the week before that, too.
Ok, but… last week was BIG.
- GPT-4: OpenAI’s upgrade from 3.5 to 4 was the biggest news of the week. The upgrade promises an improved alignment by making it more truthful and generating less offensive or dangerous output. GPT-4 also shows improvement in factual correctness and the ability to change its behavior according to user requests. It also adheres better to guardrails and refuses unsavory requests. A major improvement in GPT-4 is that it is multimodal, and now has the ability to use image inputs in addition to text, which enables it to interpret complex imagery such as charts, memes, and screenshots from academic papers. This release also marks OpenAI’s first release as a true for-profit company.
- Midjourney v5: AI artists rejoiced this week, as the new version of Midjourney seems to have corrected the “hands issue.” It also is delivering some stunning renders.
- Google PaLM: First announced last April, Google’s large language model finally was released as an API this week. Similar to OpenAI’s GPT models and Meta’s LLaMA family, it has the potential to perform a variety of text generation and editing tasks. PaLM can be trained to function as a conversational chatbot, text summarization or even coding.
- Google Workspace AI: Googs also powered up Workspace this week, announcing AI functionality to Docs and Gmail.
- Anthropic’s Claude AI: Claude is governed by “Constitutional AI”, which refers to AI systems that align with constitutional values and principles, such as transparency, accountability, and fairness. The model is only available as an API, which you can use on…
- Poe: This new tool by Quora, was actually released earlier this year. However, with the new of Claude, it got a boost, as it’s one of the few ways you can use Anthropic’s new tool. Poe pulls in APIs from Anthropic, OpenAI as well as Sage and Dragonfly, making it possible to compare LLM results – and toggling between the models is really interesting (Dragonfly, for example is built for short answers), and can help boost productivity. Read our deep dive on Poe here.
- Microsoft’s Copilot: Of the major tech co’s, it’s been Microsoft leading the way so far. This week, they announced another stride forward with Copilot. Copilot is designed to work within Microsoft Teams and Microsoft 365 apps, and can perform a wide range of tasks, including scheduling meetings, responding to emails, creating documents, and more. The AI system uses natural language processing to understand users’ requests and provide personalized responses. Copilot can also learn from users’ preferences over time and adapt to their work style. The feature is expected to launch later this year.
- …and this, we guess, is an update?
This week probably won’t be so frantic – but again, that might change. Expectations can be high because we are in it now. And by it, we mean the AI Revolution. And the thing about revolutions is that they spread as slow or as fast as the technology can provide. The Industrial Revolution took decades to take hold and impact the economy. This one, we can surely estimate, won’t.
Let’s stay in touch. Sign up for The Dept’s newsletter here.