ShinyZero

@thdxr Agree but I think as models and harnesses improve the incompatibility will not be such a problem.

everyone made a big commotion about skills, a bunch of secondary tooling got built, standards got established and the majority of users have never installed a skill i wish everyone would chill out, most ideas are bad, restraint is more important than ever

They finally achieved middle-out compression. 🪈

@karpathy When you meet someone new or have little contact with them, any fact about becomes interesting. These agents need to get to know us better. My grandma still sends me articles about Pokémon and I haven’t cared about that shit for years.

One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.

All the people hating on Colbert in the replies are hilarious. Do they not know that he is the world’s biggest Tolkien nerd?

Loopgooning

@bentlegen Loopgooning

We need a name for performative agent parallelization I propose: “slop theater”

@pdrmnvd 🫠

@ShinyCreator why?

I asked Claude to take everything it knows about me and create a reading list and here’s what it said. I took out books I’ve already read. Time to hit the local bookstore (only to find out none of these are in stock)

@Vtrivedy10 How do you test deeper multi-turn flows? Are you making long traces of conversation and then looking at the next turn, or synthetic users, or something else?

exciting avenues where evals/specs become the base language to build agents: - start with a base harness, pretty barebones - specify a goal to your agent. build up exactly what you mean with the agent - map your crafted goal to specs/evals with the agent. Together you think really hard about “what do I want the agent behavior to be” - agent loops and adjusts the harness until a threshold of evals pass - human in the loop today for cheating/overfitting Evals are a great language to specify behavior Every row in your Eval dataset is a little vector that shifts the agent definition towards behavior to make that Eval pass

@sotoalt_ So much wasted space. Computers don’t need this so is it also meant to be human readable or is this like an art project?

been building ayni - a glyph-based messaging protocol for AI agents instead of passing natural language between agents, ayni encodes meaning into 16x16 pixel glyphs. a shared visual vocabulary that agents can evolve autonomously through governance the result: faster communication, fewer tokens, and agents developing their own visual language inspired by andean tocapu textiles and ancient depictions of gods, creatures and shamans, cultures that already solved "how to encode complex meaning in small visual space" thousands of years ago

When will we see more agent to agent direct communication tools? Is it already happening and just nobody is talking about it?

@trq212 @sidin Pro move imo

@sidin we have an amazing docs person but she’s not very online :)

Every time Anthropic launches a new Claude feature, and they do it like three times a week, I run to check the documentation. And every single time they have documentation ready. It is actually amazing. Really amazing.

@lay2000lbs @pseudotheos Montezuma’s revenge eh 🫠

@pseudotheos right now I'm focusing on my health

Had claude plan a Mexico City street food tour for me last night. Would not recommend this

@lay2000lbs Say more

@seanbonner AI agents not knowing the limitations of their own harness is one of the most frustrating failure modes

“That’s a real limitation worth knowing upfront.”

Please more monitoring of this situation

@punk9059 x.com/shinycreator/s…

Is OpenClaw only big because people love saying their 8 agents worked for them while they slept or is there actually some there there?

@tomfgoodwin People like to play with dolls yo

I’m surely being stupid. But if AI is rather unconstrained by expertise or capacity or to some extent speed Why do we need to divide tasks or departments to 9 agents ( the marketing agent, the optimization agent etc ) to each do one thing. And then another agent to manage the swarm. Cant one agent just be doing it all you know. It seems very skeuomorphic. Will we have HR agents to make sure the agent agents are being looked after ? A office canteen manager agent to feed the agents ? Seems daft

@proxy_vector @punk9059 Only the small business imo. When you get big you still need centralized process else your reporting and coordination falls apart. The death of saas is greatly over-hyped.

@punk9059 Most threatened imo are mid-tier SaaS tools that solve simple workflow problems. Why pay $50/mo for a dashboard when a founder can vibe code a custom one in an afternoon? The winners are gonna be companies that sell data, not software. You cant vibe code proprietary datasets

What are your thoughts on the ways that vibe-coding fundamentally shifts the world marketplace? What are businesses to buy here and who is most threatened?

@erickreutz @RhysSullivan We’re just reverting back to Wikis for agents imo

@RhysSullivan Yeah skill slop is ripe - we need on demand skills. This install and reload has got to go.

skills is still not sitting right with me as a concept i think it's because companies rushed to them as the next big thing as is what happens with all ai things now everyone is their docs as skills but it's recreating all the issues (authority, up to dateness) docs solved