Content Credentials
In this age of AI generated everything, it is quite useful to understand the origins of content that you are consuming. For media files such as images, videos, audio files etc there is an existing standard C2PA. Digitally signed metadata is attached to it, which can be verified. However, for text e.g. this blog, it is less mainstream. Browsers don’t natively support showing such metadata or verifying content provenance. In an effort to provide such origin info about the content on this …
How We AI
Have you ever wondered how folks are using AI ? At work or in their personal lives ? AI is changing faster that anyone can keep up. We constantly hear of flamboyant/pie-in-the-sky use cases. But what about the ways it is being used “right now”. Yes it is a great coding and peronal assistant…but “how” exactly are others using it ? Have you ever booked a vacation using AI ? What worked and what did not ? What is your dev setup ?
Berkshire Hathaway Shareholder Letters
In case you have not seen it, the Berkshire Hathaway Shareholder Letters are in the public domain, all the way from 1977! They are a great read about how one of the most iconoic companies in the world is run. Warren Buffett writing style is a mixture of phenomenal insight and humor. A very hard thing to do propely. Here are some gems I came across after reading them: The textile industry illustrates in textbook style how producers of relatively undifferentiated goods in capital intensive …
An MCP server for this blog
Caus, why not ? There are north of 6300 MCP servers out there in the wild ! If you don’t know what MCP is, you can read about it here It’s a way for AI applications(e.g. LLMs, AI Assistants etc) to tie into your own tools, workflows etc. AI Apps do a great job of reasoning / articulating, but it needs data for this. Data that is often siloed, on which it did not train and does not know about.
The killer use case for GenAI
In the next few years, approx $2T will be invested in AI !(For comparison, the entire US economy is around $30T). One of the reasons for this exuberance is that there are probably many many areas in which AI can be applied, in theory. We just may not have found a compelling narrative for them all. One use case which I find personally extremely useful is that of a Personal Assistant(PA). Now a PA can mean very different things to different people. When I speak of a PA, it’s someone who can …
Economic Possibilities for our future
In 1930, the English economist and philosopher John Maynard Keynes wrote an essay titled Economic Possibilities for our Grandchildren. An excerpt from it: 1 2 3 4 5 6 7 8 9 It is common to hear people say that the epoch of enormous economic progress which characterised the nineteenth century is over; that the rapid improvement in the standard of life is now going to slow down --at any rate in Great Britain; that a decline in prosperity is more likely than an improvement in the decade which lies …
AI and the coming inequality
A lot of focus thus far on AI has been on safety. And rightly so. Safety of any new technology is critical before it should be made available to mass markets. But how about societal inequality? Inequality today is quite stark. The wealthy 1.5% own 48% of the world’s wealth. The fundamental premise of AI is to boost productivity. What happens when rich nations / corporations (who also happen to own the means of producing these models) get boosted productivity by AI ? AI enabling tech will …
Vector DBs for Data Protection
I was reading up some information on vector embeddings recently and started thinking about different use cases of it. One interesting use case was for Data Detection and Response. The thought exercise I had was that if we are able to track data access patters or data lineage and represent them as embeddings, then the embeddings of any anomalous / nefarious access pattern would have very little similarity with other embeddings.
Using External Functions with OpenAI
An easy way to boost the capabilities of AI models is to provide it with external information that it may not have. e.g. maybe your company has some proprietary data that would never be included while training the models. But if you could provide the model with this info during inferencing, it would be able to provide more accurate results. One way of doing that with OpenAI is via function calls. You define a function with all its input parameters and describe what the function does. The model …
Having Fun with OpenAI
AI is all the rage these days. New capabilities, that seemed impossible just a few months ago, show up almost weekly. So I decided to take it for a spin and see what different tasks I could code up quickly. I chose OpenAI, but you should be able to replicate these using any of the other models as well. Printing the text of a html page Book Recommendations Converting csv to json using json_schema Printing the text of a html page The first quick task I tried was to ask AI to print out the text of …