Turtle

@djmicrobeads.bsky.social dj is back!

I’d expect nothing less, this is really great, Ha! Are you planning to write or present more about the process and work in the future?

Also nice to see your writing again, missed enjoying your working out loud after the departure from the other site.

bsky.app/profile/gord...

Indeed

This week I added emotional valence and logarithmic compression to an LLM memory system. Going for a ‘Darmok at Tenagra’ approach for the overall system. Subjective time.

Framed product development principles, organizational research, etc as the solution to horizontal problems across org systems.

Yes, covid created existential threat for a company I was working with at the time.

Previous services model meant ‘r&d’ was outsourced.

Created alignment around a
compounding consolidation model, allowing us to model second order effects across org systems for demonstrating value chain quickly.

I’ve found it significantly increase the rate at which I can identify / map / create ‘systems of systems’ work and manage abstractions, explore the translation or abstraction of domain-specific patterns.

A smarter rubber duck with pattern matching capabilities?

@michaelnielsen.bsky.social You had a question on Twitter about whether/how ai tools have improved work. Wanted to share here as I’m inclined to use this platform in a looser / scratchpad capacity

Indeed, and there’s so much strategy / framework writing available (of variable quality ofc) that the synthesis of experts is likely to generate a better approach than the majority of individuals.

Also, a moat that scales proportionally to context windows. And they’re going to keep increasing in size.

‘How can the system self-improve?’

‘How can the system improve how it self-improves?’

Will be sharing more throughout the year, both here and Twitter.

Also curious how @gordon.bsky.social has been thinking about the space

It feels like the current approach to applying ai, even from the leading labs, is to shove an old model onto a new set of capabilities.

Working on something that diverges drastically.

Nice to see you continue on here after leaving Twitter, Ha! Don’t recall how much we interacted there, but appreciate your writing and thinking regardless. I’ll have to spend more time here again to catch up on it.

This is a system for ingesting and pattern matching across vast quantities of personal data and metadata btw

Performance is substantially better as a result of clearer separation of sections and their functionality, data models, user-specific behaviors.

Surprised how much room for compression there is when managing the information structure of my prompting system myself, rather than offloading entirely to sonnet 3.5.

How would you feel about, in addition to this model, the ability to explicitly ‘follow’ a thread?

The idea here being that you might want updates from a thread without following the original poster, or want to ensure that the post and updates from the OP are guaranteed to be part of your timeline

and it turned out Lord Michael Ashcroft got it after all 🔥💪

When I go to my profile, and look at my lists, and tap a user on a list, the transition is abrupt and janky compared to other instances of navigating to a user profile

When I am looking at someone’s profile who hasn’t posted any content, the views associated with the different tabs are still scrollable

Similarly, with the lists tab

when I am on my user profile, and I have scrolled, so the header is fixed to the top of the view, and I swipe to the feeds tab, disappears, rather than the regular interface that shows up if I scroll from the initial state of the profile

janky* sorry, dictation

Tapping a users profile picture from the main view of the search tab has a Jenkee animation similar to a vape their name in a regular post. Preference for consistency across all the ways I can open a user profile.

when any of the main views are scrolled, and I tap the tab associated with that view, I am scrolled to the top of the feed. Except for the search tab. Can that be made consistent?

I also enjoyed the twitter pattern where long pressing the search tab focused the search field and opens the keyboard

if I tap on the link associated with my name in a post, the transition is different than if I tap my profile picture. The animation is Jenkee, but I would also expect that both interactions should be consistent, rather than having different effects.

when exceeding the character limit of a post, the ‘reply’ button changes to an inactive ‘post button’, causing that toolbar to jump (to accommodate the different length of the button string).

preference for keeping strings consistent across states.