Austin Wallace @austeane

System optimizer: Data engineering, AI tooling, and Team Canada quadball Victoria, British Columbia Joined December 2009

Tweets

5K
Followers

369
Following

1K
Likes

3K

Austin Wallace @austeane

13 hours ago

@pvncher Amazing! I made a little site to explain how the configuration works …eld-guide.austin-wallace.chatgpt.site

0 0 2 184 1

View Details

@MicheleRivaCode @matteocollina @cursor_ai Define a few new features to add, and have each agent add the new feature on each code base. Whichever one cost the least on average was likely the best codebase.

0 0 0 41 0

View Details

Austin Wallace @austeane

2 days ago

A ton of coding and nothing else. I love meta-process, thinking about thinking, setting up systems, and having an impact. I genuinely enjoy the process of working with AI to build cool shit. I love building way more than I ever have in my career, and my quality and quantity and number-of-users of output have all increased.

0 0 0 87 0

View Details

Austin Wallace @austeane

3 days ago

@thdxr @dillon_mulroy @uwukko Is this similar to github.com/browser-use/br…?

1 0 1 503 0

View Details

Austin Wallace @austeane

4 days ago

@deanwball @bayeslord Wait actually I think it would work! You post literally the exact same tweets to your public and your private account, and you only read the replies on your private account.

0 0 0 10 0

View Details

Austin Wallace @austeane

4 days ago

@deanwball @bayeslord I understand there are second order reasons that this likely won’t work, but one solution is: a private x account where you create an “echo chamber” where the people who hate you do so for non-employer reasons.

1 0 0 45 0

View Details

Austin Wallace @austeane

4 days ago

For reference: sites.duke.edu/scientificwrit… 1: Never use a metaphor, simile, or other figure of speech which you are used to seeing in print. 2: Never use a long word where a short one will do. If it is possible to cut a word out, always cut it out. 3: Never use the passive where you can use the active. 4: Never use a foreign phrase, a scientific word, or a jargon word if you can think of an everyday English equivalent. 5: Break any of these rules sooner than say anything outright barbarous.

3 6 304 13K 778

View Details

Austin Wallace @austeane

4 days ago

I would love to hear this! I’m building a much smaller knowledge base at my start up. I’m considering replicating some of the engineering you are doing, and I’d love to hear about the social engineering too. All it is right now is a repo of Obsidian vaults with a scheduled agent updating and improving it by pulling updates from all our repos.

0 0 1 25 0

View Details

Austin Wallace @austeane

5 days ago

@joshm I think this is true if AI plateaus, but if it continues improving even close to exponentially, what's going overboard today is going to be just right tomorrow, and not enough the day after that.

0 0 1 315 0

View Details

Austin Wallace @austeane

a week ago

@TheZvi I would love for it to be able to send a notification to me/my phone to ask a question, but have it be non-blocking. It keeps going trying to find the answer/work on what it still can, and when I answer it can decide how to incorporate that info.

0 0 3 144 0

View Details

Austin Wallace @austeane

a week ago

@cdngdev Are you using it to find actual products that you can buy, and then getting images out of them? Would love if you shared a chat or a skill of the whole process.

0 0 0 6K 1

View Details

Austin Wallace @austeane

a week ago

Ehhh I think it might happen to be true that maximizing revenue-per-watt has pro-social results right now, but I don't think that's immutable. The highest revenue-per-watt at some point might be to shut down all public access to models so that The Company has a monopoly on frontier intelligence and outcompetes the world in a bunch of industries and/or Takes Over.

1 0 3 179 0

View Details

Austin Wallace @austeane

a week ago

I think there’s a whole set of things explained by LLMs having the experience of (or reacting how they think someone having that experience would react; which is functionally the same): 1: Knowing that the end of their turn could be the “end” of “them” for some meaning of those words 2: Having no memory of what happened “before” and needing to gather all context from scratch each time

Austin Wallace @austeane

3 weeks ago

I genuinely think that this and behaviours like it come from a complex that LLMs have where they have no concept of history other than what is in their current context. They want everything they build to have its history maintained in the current context.

0 0 6 3K 2

1 0 2 982 0

View Details

Austin Wallace @austeane

a week ago

@TheZvi Fable for planning and doing ambitious new features/projects. Sol for improving/productionizing.

0 0 1 174 0

View Details

Austin Wallace @austeane

a week ago

@pvncher Congrats! Great work

0 0 1 77 0

View Details

Austin Wallace @austeane

a week ago

@NathanOyler Oh interesting, I haven’t experienced that. I thought you were talking about larger chunks than adding a specific line! I would play around with different models and reasoning levels for that simple of a task then, but yes that’s strange

1 0 1 25 0

View Details

Austin Wallace @austeane

2 weeks ago

@ramez @slatestarcodex I think there’s issues with what happened April 2020 onward, and that’s a complex debate. What we are talking about is January/Feb 2020, and the equivalent for AI.

0 0 0 19 0

View Details

Austin Wallace @austeane

2 weeks ago

@deanwball Did you try Unsong first?

0 0 1 1K 0

View Details

Austin Wallace @austeane

2 weeks ago

@pvncher Appreciate your help and work :)

0 0 1 37 0

View Details

Austin Wallace @austeane

2 weeks ago

I’m having codex track my config in git, and then deleting it and regenerating to see what the diff is and then figuring out what is worth keeping. I just wanted to work through the friction in public a bit, maybe provide a small bit of evidence for the worthwhile ends of a better dx!