The Hubris of Scale: A Lesson From Junkmail Detection

1 year ago

Scale won't solve every problem, and won't even solve the main problem of today's AI!

Beyond Prompting: The Other Ways to Influence LLM output

1 year ago

Beyond prompting, learn every other thing that you can do to an LLM, like the sampler, runtime modification, optimization, and…

Why Social Media Search is So Bad, and How to Fix It

1 year ago

Social media search is terrible because it's a harder problem than web search. Yet I implemented my own search engine…

Far Over the Misty Mountains Cold

1 year ago

I improvised music to go along with reading The Hobbit to my children. Grab a copy!

The Magic Promise of Inference-Time Compute

1 year ago

How much is it worth to solve humanity's biggest problems? If only there was someone to pay! Agents may enable…

AI’s Hardest Problem: Teaching Us to Use It Better

1 year ago

Not only is AI smarter than us in most ways, but it is so much smarter than us that we…

The Tyranny of Chat Interfaces to LLMs

1 year ago

So much of our interaction with LLMs is via a singular chat box. It doesn't have to be that way.…

AI Videocasts and Digging Your Way Out of a Hole

1 year ago

AI gets into a rut as you talk longer, as I found out when I worked with ChatGPT to debug…

What are AI Agents? A Precise Answer

1 year ago

What are AI Agents? Although it seems every bit of software claims to be an agent, there are real agentic…

The Archive pt 3: Don’t Hack Away on Vibes Alone

1 year ago

Let's not hack away on The Archive on vibes alone; let's evaluate our code. Using Promptfoo!