ryoshu

Oof. I turned the history referencing off. I use ChatGPT for wildly diverging topics and it will bring things up that have zero relevance to what I'm currently looking for if history is on.

This is a good idea. Not sure what ToS it would violate. But a good idea.

I think the harnesses are responsible for a lot of recent gains.

To pedant it up, not across GPUs.

Tulsa used to have a rich Black neighborhood.

SAP or Salesforce can do anything. Or you can write custom code for a fraction of the price for small/medium sized businesses.

If he's having an impact it's because he can break through the bureaucracy. He's not trying to protect a fiefdom.

LLMs fail at causal accuracy. It's a fundamental problem with how they work.

I can't do that either without opening up an image editing tool. Give the model a tool and goal with "vision". Should work better.

The reward is helping people hone their craft and grow in their careers. Being a shit shield sucks, but the trade-off is worth it.

Some people write to convince. Some people write to provoke.

They should audit LLMs.

Agreed. I use CC a lot for exploratory work. It's great with fast iteration for throwaway code.

You can look it up on his blog: https://www.antipope.org/charlie/blog-static/2006/06/

This article talks about prompt evals https://www.anthropic.com/engineering/writing-tools-for-agen.... There are plenty of approaches to provide some degree of rigor around the slot machine output.

Data vs. feels

The Ren Faire in NY has a time traveller day. Trekies abound.

I asked GPT-5 to generate an image prompt. I asked it to use that prompt to generate an image. It was a content policy violation.

And you have to pay special attention to the tests written by LLMs. I catch them mocking when they shouldn't, passing tests that don't actually pass, etc.

A lot of times coding is thinking. It's like sketching out an idea on paper, but with code.

Used to be a mistake. Now it's pretty clear you have to pay-to-play to get top billing on SERPs.

Last attribution doesn't work. Retargeting is a joke and the most valuable spenders have ad blockers so you can't even reach them through display ads. The data is manipulated everywhere as long as you can hand wave towards ROAS.

Possibly https://www.secretservice.gov/investigations/cyber

Optimize everything so you can take the soul out of it.

The demo videos show very poor knife skills.

Reverse centaur.

I still struggle with ORMs. SQL is... declarative. If you're working with multiple RDBMSs, sure? Maybe I want my local dev to be sqlite and scaled be postgres? I've never run into that in production. A DSL on top of a DSL doesn't make a lot of sense.

"qual"

I visited Dachau years and years ago. It was a nice summer day, but a pallor fell over when we went inside the camp. It felt like the sky darkened and the color drained from the entire environment.

Selection bias + perceptual adaptation is my experience. Selection bias happens when we play the probabilities of using an LLM and we only focus on the things it does really well, because it can be really amazing. When you use a model a lot you increasingly see when they don't work well your perception changes to focus on what doesn't work vs. the what does.

Living evals can solve for the quantitative issues with infra and model updates, but not sure how to deal with perceptual adaptation.