Working with LLMs

I’ve followed Rui Carmo for more years than I’d care to remember. I laughed out loud at his latest post trying to get a tool which gives GPT a harness to use web search and scraping.

Notes from the AutoGPT Frontier

Today is a bank holiday, so I had some time to peruse my AI and ML backlog and ended up playing with SuperAGI, which is a cute (but hilariou…

Tao of MacRui Carmo

“I got it running inside a sandbox and asked it to summarize the top ten Hacker News posts...The thing is, NONE OF THOSE ARTICLES ACTUALLY EXISTED. The HN home page looked nothing like it….Somehow gpt-3.5-turbo decided to hallucinate them instead of using the web scraper, which just goes to show you how stupefyingly fallible these things are right now.”

That’s the key thing to remember with these models. It wants to accomplish the goal and how it does that, well that’s where hallucinations can crop up.

Large, diverse and dispersed data summaries would be fantastic (I still read RSS feeds). I’m not tackling it yet though.

(This is a good read about LLMs as well)

On Large Language Models

I’ve been pretty quiet about ChatGPT and Bing for a number of reasons, the most pertinent of which is that I have so much more going on in m…

Tao of MacRui Carmo