Microsoft launches Copilot for Office 365

OpenAI addresses concerns regarding GPT-4

Author

The Alignment
March 20, 2023

Welcome to The Alignment. Apologies for missing the Friday issue, the team was out cold with the flu 🤒. But we're back at full strength and ready to start the new week 💪

 As always if you enjoy reading our posts be sure to spread the word ! 

Here’s what we have lined up for you today - 

  1. Microsoft launches Copilot for Office 365  

  2. OpenAI addresses concerns regarding GPT-4's openness and safety

  3. Humans + AI = Productivity Machines

Microsoft has finally released Copilot for Office 365. An AI assistant purpose built to boost the productivity of professional teams. 

Through a couple of keystrokes, Copilot is capable of : 

  1. Taking meeting notes 

  2. Preparing notes for upcoming meetings 

  3. Drafting documents on Word

  4. Building Powerpoint presentations 

  5. Going in to Excel and analysing your data

What’s impressive is that it has context of your work through the data it retrieves and so the suggestions and insights it provides are personalised to the user.

Microsoft will also introduce the Business Chat feature which brings together documents, presentations, emails, notes, and contacts into a single chat interface in Microsoft Teams capable of generating summaries, planning overviews, and more. 

The launch of Copilot is massive because now users have the capabilities of ChatGPT with the added benefit of copilot having all of the business context of the user. Secondly, there are over 400 million users who use Office 365 daily. There is no better stress test than that. It also shows how strong of a distribution stronghold Microsoft has when it comes to the productivity software suite.

OpenAI’s GPT-4 launch was great. The live demo was beyond impressive, Twitter has been going crazy with folks showcasing their GPT-4 adventures and there is no doubt excitement all around. 

There is however an air of mystery about what OpenAI did not talk about in the launch announcement. Particularly on things like the data that was used to train the model, the hardware used to train the model, costs related to energy consumption of the model and the underlying workings of the model. Folks in the AI community have criticised this decision because according to them a) It goes contrary to the founding ethos of the company, b) It makes replicability hard and c) It makes it harder to build safeguards against potential AI threats. 

Emad Mostaque, founder of StabilityAI and creator of Stable Diffusion has been going directly at OpenAI expressing his displeasure at how the company chose to release GPT-4. 

OpenAI founder Ilya Sutskever directly addressed these collective frustrations when asked and listed two primary reasons for why they chose to release GPT-4 the way that they did. 

Competition - OpenAI is not the only player developing very large language models. GPT-4 is super complex and with competition emerging from a competitive standpoint you have to look at this decision as a “maturation of the field”. 

Safety - Models like GPT-4 are extremely potent. As the models get more powerful, the ability to do harm also increases and at that point it does make sense to disclose the model to the general public (as bad actors could be present anywhere). 

Here’s what Ilya had to say about the change in approach on OpenAI’s end - 

“We were wrong. Flat out, we were wrong (talking about their previous approach). If you believe, as we do, that at some point, AI — AGI — is going to be extremely, unbelievably potent, then it just does not make sense to open-source. It is a bad idea... I fully expect that in a few years it’s going to be completely obvious to everyone that open-sourcing AI is just not wise.”

As a business OpenAI has the freedom to choose the path they want in bringing their models to the world. The whole world is watching, and they’re not shying away from the attention. 

Around the industry - 

  1. Baidu’s shares drop after ChatGPT competitor seems unimpressive

  2. 1.7 billion parameter text to video generation diffusion model

  3. Microsoft Bing AI is now available without a waitlist 

  4. Responsible AI at Google’s Impact Lab

  5. Microsoft lays off ethics team ?

Product Watch - Khanmigo by Khan Academy- 

The world's favourite tutor just got a massive AI upgrade. Sal Khan demonstrates Khanmigo himself !

Humans + AI = Productivity Machines

One man, one ChatGPT join forces to build a mighty impressive iPhone app.