Month: April 2024

AI thoughts in only 2017

Besides the latency, this AI robot is really impressive. When will it be taken for granted? Innovations coming with furious speed.

Machine Learning? Oh yeah, seen that…

Deep Learning? Yep…

Back when intelligent computer vision was developing.

Was also listening to this old TED talk below. People thought of AI as something out there in the future some day. Now this looks antiquated. We’ve taken for granted the power of AI already.

On the precipice

I have for some time now realized that the biblical flood was a local event. The idea of a world wide flood is scientifically impossible. But I was listening to Jack Hibbs about the prophecies in Ezekiel and Daniel  wherein he asserts that Magog is now Russia. Does he not realize that he is saying that Magog is now a nation by another name?

If these things are true, why don’t believers realize that the flood narrative is also ancient? The Mesopotamians experienced a local flood. The epic of Gilgamesh is an alternative account of a historical event.

Building an AI

Training Set

  1. Figure out what you want from the data.
  2. Determine the type of machine learning model you need, standard ML algorithms or artificial neural network.

a. If a standard ML algorithm, then it could use K nearest neighbor or Naive Bayes.
b. If a neural network, feed a training set and determine how much to change the weights.
c. Use backpropagation to adjust the weights to lower the cost function, i.e., the system will go back through the network and adjust the dials to increase its accuracy.

Test Set

  1. Data will be added from the test set. The test set will not be labeled.
  2. If the data in the training set is not sufficient for the test set to do its classifications, then that’s called “overfitting” the data.AI systems are only as good as the data they are fed. They can learn by trying different things. They can do things humans cannot–and vice versa. AIs can process large amounts of data quickly and see patterns we cannot, but they still need supervision and direction.

AI == one result?

Search engines give you a list of results. Chatbots answer in a stream of one result. Google will be able to serve you with exactly what you want, but with only one result, the bot’s answer. Many people don’t specify the options for results:

  1. Ask for a chatbot to take on a role, e.g. “you are a rocket scientist, explain…”
  2. Give context information to go with the request.
  3. Give x number of examples. The capabilities of chatbots are there for the using, but I think users need to learn how to prompt better in the pursuit of the diversity of answers. A reasoning engine, like ChatGPT, Claude, Bing Copilot, understands your question rather than just giving you rankings. Taking advantage of this difference–prompt engineering–must be a taught skill. Bingchat unites a search engine with a reasoning engine: you get the conversation with clickable sources. It’s a good sign of what is coming.
  4. Ask for competing answers, i.e. answer in the debate style with points for and against.
  5. Ask for an answer using an analogy.
  6. Ask for the bot to be creative, ideate, brainstorm, give ideas.
  7. Ask for a summary of text you input into the reasoning engine when desired.

A consideration is to be wary of the factuality of the reasoning engine’s response since it can be wrong. It can certainly sound convincing, but don’t trust everything you see or hear.

Where will your prompt take you?

Bias considerations

Considerations to help avoid bias in AI classification:

  1. Gather diverse data.
  2. Include a wide range of judges.
  3. Monitor the output of the algorithm.
  4. Attend to edge cases.

    Cover as much as you can and make sure you are addressing it as a way to overcome the inherent tendency toward bias when labeling and enhance the generalizability and utility of the work you do.

“Elon Musk” update

I’ve had to put down the Musk biography because of the new job. But I did have some opinions of what I have read so far.

There’s obviously some venom toward him. Some people think he’s brought chaos to the world; others, a new freedom for discourse. Both are true. I  certainly wouldn’t want to work for him. Yes, it’s true that he is an inspiration, especially for those on the spectrum–and I thought that for the first 500 pages of Isaacson’s book. But when the stories become repetitive, you get the idea of what it’s like in that brain of his.

He has “surges,” where he pushes his employees to outperform. Many times, his demands are successful–other times, not. He sleeps on the factory floor or in a conference room and has a preternatural ability to sleeplessly work through technological problems until they are solved.

My biking buddy Lennie said a friend who worked for Musk told him of his holding one-on-ones with employees where he doesn’t look up while you are sitting there and asks difficult questions. He’s certainly a genius in engineering and asks those Google-level interview questions.

But there may be a trail of beat-up workers: the ones that have managed to survive his firings. (He asked who wanted to remain at Twitter/X when he took over, but then fired some of the same people who decided to stay.)

So, interpersonally, he sounds like an intimidating man. But he certainly has made a dent.

Work in general

I forgot about the need for free time when you work full-time. I miss my mornings reading. But I am doing some really cool stuff for work.

In the news, some alarming items about the middle east. It’s easy to get into apocalyptic thinking.

Still thankful for fam and friends, people who prayed for me.

Been taking Jax to a new park when possible, but he gets overheated.