Hi, I’m Azeem Azhar. I advise governments, some of the world’s largest firms, and investors how to make sense of our exponential future. Every Sunday, I share my view on developments that I think you should know about. In today’s edition: AI risks and how to deal with radical uncertainty,
On LLM capabilities - I have found that using the OpenAI APIs is a massive accelerator for deploying NLP and information retrieval capabilities to enterprise apps.
For example, building a natural language interface on top of corporate data can be done easily using the GPT-X API. You just give it an outline of your schema and it can generate SQL for you with some creative prompting. You are not sharing any proprietary data but allow for much more sophisticated natural language or conversational interfaces on your data without building a domain specific model.
Similar, tasks such as disambiguating and classification are much easier. Knowing that React, ReactJS, react-js is the same thing or that Angular, Vue, React are all examples of fronted technologies is much easier.
There is a lot of discussion around AI safety, AGI, and hallucinations but I am more excited about how easy it makes mundane tasks than 99% of enterprise software has to do.
Amazing as always. Just a little bit of context as the AI wave keeps on building... I presented an introduction to AI Workshop last week to 100 SME business owners in the UK (broken down into smaller groups and part of an overall brand strategy day for a client) and only four had used chat-GPT and none had enjoyed the company of New Bing...
Your take on Cowan's remarks is valuable: "The idea, in its essence, is not to ask whether to go ahead but rather to assume that as something is going ahead, the focus should be on steering it in the most desirable direction."
But who does the steering?
I'm glad Reagan did not decide the US could steer nuclear weapons development without consulting the USSR. I think steering the development, deployment, and direction of AI needs broad, even multipolar, participation.
Quite right. That's the issue with all of the ways we talk about governing AI. When it's about steering: who does the steering? We need alignment to human values, but whose human values?
Taking into account participation, Cowan's remark is useful as well: do we focus deliberative / public space debate on whether or not further AI development should be paused, or do we use that time and energy to figure out ways to govern it?
It's more complicated than that, of course. But Cowan is one of these people who help you think, even if one disagrees
On LLM capabilities - I have found that using the OpenAI APIs is a massive accelerator for deploying NLP and information retrieval capabilities to enterprise apps.
For example, building a natural language interface on top of corporate data can be done easily using the GPT-X API. You just give it an outline of your schema and it can generate SQL for you with some creative prompting. You are not sharing any proprietary data but allow for much more sophisticated natural language or conversational interfaces on your data without building a domain specific model.
Similar, tasks such as disambiguating and classification are much easier. Knowing that React, ReactJS, react-js is the same thing or that Angular, Vue, React are all examples of fronted technologies is much easier.
There is a lot of discussion around AI safety, AGI, and hallucinations but I am more excited about how easy it makes mundane tasks than 99% of enterprise software has to do.
Thanks for linking the Garth Greenwell essay about Sabbath’s Theater - I read the book long ago and the essay is excellent!
Interesting tweet in response to the Tyler Cowen piece: https://twitter.com/s_r_constantin/status/1640783381150441499?s=20
Amazing as always. Just a little bit of context as the AI wave keeps on building... I presented an introduction to AI Workshop last week to 100 SME business owners in the UK (broken down into smaller groups and part of an overall brand strategy day for a client) and only four had used chat-GPT and none had enjoyed the company of New Bing...
they better hurry up! i bet their kids have used it.
Your take on Cowan's remarks is valuable: "The idea, in its essence, is not to ask whether to go ahead but rather to assume that as something is going ahead, the focus should be on steering it in the most desirable direction."
But who does the steering?
I'm glad Reagan did not decide the US could steer nuclear weapons development without consulting the USSR. I think steering the development, deployment, and direction of AI needs broad, even multipolar, participation.
Quite right. That's the issue with all of the ways we talk about governing AI. When it's about steering: who does the steering? We need alignment to human values, but whose human values?
Taking into account participation, Cowan's remark is useful as well: do we focus deliberative / public space debate on whether or not further AI development should be paused, or do we use that time and energy to figure out ways to govern it?
It's more complicated than that, of course. But Cowan is one of these people who help you think, even if one disagrees