Agentic software engineering @DaltonMillsAI. Localist. Global perspective. Whimsy.
Colorado, USA
Joined February 2008
- Tweets6.5K
- Following4.2K
- Followers623
- Likes16.6K
Pinned Tweet
For the last few days, I’ve been using /poteto-mode to help with hill-climbing improvements to the agent we’re building into @DaltonMillsAI. It’s been a huge help teaching agents how to coordinate the process and measure at each step.
Another experiment this week: comparing different models for a specific phase of an agent.
Using a standard set of prompts, we captured repeated control and experimental measurements for using Haiku instead of Luna during a tool-selection pass.
Haiku selected tools similarly or better but almost twice as fast. Due to how the tool data is tokenized though, there was a 30% cost increase. We rolled it out.
Important to break down the parts and measure them on their own!
The AI security apocalypse will be an interesting forcing function for AI adoption. Businesses will not be able to combat the attacks at the velocity they will start coming and will need to use AI to keep up.
Are folks using @claudeai subscriptions with Hermes or OpenClaw again?
Canceling 8 x ChatGPT 500 subs w/@OpenAI.
If you value your time, then OpenAI just isn't it right now.
Poor service availability, and inference is super, super slow. This isn't even about model quality; the problems go far beyond that. @claudeai is 100/100 better choice right now, faster inferencing, better availability and the models are better too.
This is one of the more interesting ones I’ve heard! Allow retirees who don’t need SS to forego it or give it to children in exchange for other benefits.
Running a process like this in the past was always fun/educational, but also time consuming. Teaching agents how to do it well just relentlessly improves performance pretty easily. Pick the top 5 most used parts of your app and run it.
For the last few days, I’ve been using /poteto-mode to help with hill-climbing improvements to the agent we’re building into @DaltonMillsAI. It’s been a huge help teaching agents how to coordinate the process and measure at each step.
Is anyone doing sovereign wealth funds with AI yet? I wonder if we should be treating AI more like minerals from public lands due to the potential for labor disruption.
In this scenario, one other question would be: what is an AI company? We shouldn’t leave any room for games around this definition. I’d say any company developing models, offering inference as a product/service, or whose hardware components are critical to those industries.
Would it be hard to offer a “Mark all Notifications as Read” in the push-and-hold menu for every iOS app? It’s sometimes impossible to find how to dismiss a notification within the app.
Did Boomers feel this about the Greatest Generation? Will Zoomers feel it about Gen X? These people have had their whole lives to accumulate assets, it’s natural that they have more than you.
BUT! There are specific structural disadvantages around housing, Euclidean zoning, and government-supported home loans that we have to solve. I don’t know the answers.