4 ms·

I doubt this because AI has the implicit biases of the training data and the data scientists who trained it. My firm likes the idea of ChatGPT, but our product

I doubt this because AI has the implicit biases of the training data and the data scientists who trained it. My firm likes the idea of ChatGPT, but our product risk analysts basically concluded we need to build our own for internal usage because we can't trust OpenAI to not push various hidden biases. For our marketing research and copy writings tests using ChatGPT, we ran into an insane amount of issues with market segmentation copy and research analysis hitting various limiters.

pmarreck 4y ago
> I doubt this because AI has the implicit biases of the training data and the data scientists who trained it. The nice thing about fallacies is that they flag all biases equally, so the bias claim is irrelevant when it comes to it calling out errors of reasoning (all of which seemed valid to me, regardless of what I actually believe). > we ran into an insane amount of issues with market segmentation copy and research analysis hitting various limiters could be a prompt construction issue, which I've found may require some experimentation > our product risk analysts basically concluded we need to build our own for internal usage Well, at the current pace of things, you'll probably be able to run all of OpenAI inside your Mac laptop in, ohhhh... 1 month or so
- stuckinhell 4y ago
  It wasn't prompt construction, I can't share more because it'd be too revealing about our targets for new growth. At the current pace, I think most companies will be scrambling to build their own chatgpt and spending millions to do it. We already have approval for initial experimental budgets at 2 million just for hardware. The hype wave is real. I think data providers are going to do well in this gold rush.
  - zamnos 4y ago
    Nvidia is the one to invest in, though looking at it's performance since March 15th, it might be too late and it's already been priced in.

pmarreck 4y ago

> I doubt this because AI has the implicit biases of the training data and the data scientists who trained it. The nice thing about fallacies is that they flag all biases equally, so the bias claim is irrelevant when it comes to it calling out errors of reasoning (all of which seemed valid to me, regardless of what I actually believe). > we ran into an insane amount of issues with market segmentation copy and research analysis hitting various limiters could be a prompt construction issue, which I've found may require some experimentation > our product risk analysts basically concluded we need to build our own for internal usage Well, at the current pace of things, you'll probably be able to run all of OpenAI inside your Mac laptop in, ohhhh... 1 month or so

stuckinhell 4y ago
It wasn't prompt construction, I can't share more because it'd be too revealing about our targets for new growth. At the current pace, I think most companies will be scrambling to build their own chatgpt and spending millions to do it. We already have approval for initial experimental budgets at 2 million just for hardware. The hype wave is real. I think data providers are going to do well in this gold rush.
- zamnos 4y ago
  Nvidia is the one to invest in, though looking at it's performance since March 15th, it might be too late and it's already been priced in.