OpenAI and the balancing act of being Open

Welcome to The Alignment, where we help you understand what’s happening in the world of AI without any of the jargon or bs.

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

1. OpenAI shares how its going to address biases and safety issues

2. Wharton Prof solicits Bing as his Research Assistant

3. Time Magazine's Cover vs Our AI Generated Magazine Cover

OpenAI and the importance of being (sufficiently) open

OpenAI has finally addressed how they are thinking about developing the behaviour of ChatGPT and AI systems in general. In a post on their blog they highlighted a three pronged approach to build better behaving systems:

1. Improving default behaviour and reducing the biases of how ChatGPT responds to different inputs right out of the box.

2. Defining your own AI’s values and bringing updates to ChatGPT such that users can easily customise its behaviour.

3. Soliciting public input and feedback to engineer future iterations of AI systems. On that end, OpenAI has been in talks with Educators to understand how ChatGPT is affecting education in the classroom amongst other efforts that they will begin piloting.

The post clearly indicates OpenAI’s intentions to ensure safety but also helps understand why Google has not publicly launched their models into mass adoption given the risks with biases, trust and safety. So while OpenAI’s approach has certainly gotten their technology in the hands of hundreds of millions of people we’re still miles away from declaring which strategy is better. Have systems out in the open or be overly protective before releasing them ?

OpenAI has come a long way from its beginnings as a non-profit AI Research Lab to a multibillion dollar (valuation) organisation competing head on with Google. All that in a span of 7 years since founding and only 4 years since they formed their for profit entity. So matters to do with ethics and safety will only form a larger part of the organisation in the coming years.

A Wharton Prof and his Ethical Research Assistant

Public intellectual and Wharton Prof Ethan Mollick shared how powerful Bing is as an out of the box research assistant. He prompted it to go through his academic work and suggest research ideas for future work. Amongst other things Bing suggested reseacrh ideas based on previous papers, gap in literature, suggested methods and offered potential data sources. Deep into the conversation Bing also indulged in an ethical conversation about the ethics of Bing “writing code” for Ethan. Fascinating but like kinda weird.

What’s slightly unsettling is that folks are not clear what’s happening under the hood for Bing to carry out conversation like this. There is clearly a “personality” assigned to Bing. So the sooner Microsoft releases some literature on how Bing moulds its answers and personality the faster we will be able to understand what we are interacting with.

Product Watch Perplexity AI -

Founded by OpenAI and Meta Alumni, it’s an information retrieval engine built using Large Language Models that cites the answers it gives and now even lets users “edit answers by either adding sources for more perspectives or deleting irrelevant sources.”

Around the industry -

1. Famous solo capitalist Elad Gil shared his outlook on the AI market landscape from an investor's perspective.

2. Google has developed a deep learning model that beats state of the art physics models in rain prediction.

3. ML leader and Big Tech exec Xavier Amatriain has released a 36 page preprint about the different transformer models used in industry.

4. Anthropic which recently raised $300 million from Google to develop AGI is now hiring “prompt” engineers for a base salary of 250-330k whilst only requiring a basic programming background. Certainly they’re not feeling the recession.

Time Magazine vs Our AI Generated Magazine Cover

Personally I like ours, but then again I might be "biased".