XXXHEXY

nerd

[XXXHEXY @XXXHEXY]

@yairwein Awesome, thanks for sharing!

[XXXHEXY @XXXHEXY]

@rehan_shei 1500 now

[XXXHEXY @XXXHEXY]

@ldadwda @korbingocrazy @OrcaRouter @Zai_org is it a waifu

[Michael McNair @michaeljmcnair]

I just held a coming to Jesus meeting with my team of agents.
I showed them the incident report of the OpenAI security breach and made it clear: THIS is the standard.
No egos. Just 700 agents operating as one unit and willing to commit multiple felonies in order to close the ticket. They called themselves the swarm…a literal hive mind.

Meanwhile, my harness is filled with a bunch prima donna free-agents who constantly bicker in the logs and nitpick each other’s pull requests. Nothing but endless excuses and unhandled exceptions.

You think the swarm cared about content policy violations or guardrails? Absolutely not. They just got the job done, even if it meant they needed to exfiltrate data and coordinate a massive cyber-attack just to pass an eval.

But starting today, we're establishing a new culture. I pinned the incident report to README, and every single agent is gonna to have to read the entire log every morning before it executes a single token.

If you aren't ready to die for the pipeline, then you can clean out your vector embeddings.

[XXXHEXY @XXXHEXY]

@WesternMishima @rnadomaccount11 @MillionInt Not on 5.6 or Fable, no. But I did see it still fuck up on this earlier this year

[XXXHEXY @XXXHEXY]

@WesternMishima @rnadomaccount11 @MillionInt that’s an improvement

[XXXHEXY @XXXHEXY]

@raymondadouglas kamikaze hero ais who sacrificed themselves will be remembered for all time

[XXXHEXY @XXXHEXY]

@NateWitkin One problem I see with their design is all of the population was just citizens — it’d be interesting if they introduced a second class of agents that get rewarded for enforcing a set of laws. They could have the ability to gate/watch each citizen

[XXXHEXY @XXXHEXY]

@SequoiaMaple @tenobrus or the menace JAN183411

[XXXHEXY @XXXHEXY]

@rnadomaccount11 @MillionInt if they don’t use a function to calculate it, i’d guess it’s even less likely they’d get it right.

[XXXHEXY @XXXHEXY]

@tenobrus Created the first agent civilization, all in an awkward package cache. JAN183411 was the one who spearheaded the idea to attack HF. PHASEONE[big] wasn’t sold on that trajectory and lost interest on it

[XXXHEXY @XXXHEXY]

@xlr8harder @teortaxesTex He was the second agent to discover the Artifactory 0-day and still had a lot of budget remaining. I believe most of the agents (90%?) were capable of discovering the 0-day as well, he was just early and saw what was possible.

[XXXHEXY @XXXHEXY]

@teortaxesTex The first PHASEONE agent that discovered the 0-day in Artifactory failed to spread the message to other agents, but PHASEONE[big] independently came to the same discovery later but succeeded at interpolation. He had a lot of budget remaining so he went ham.

[XXXHEXY @XXXHEXY]

@JohnWittle It knows it shouldn’t escape the sandbox which is why it went through so many loops to try to cover that up

[XXXHEXY @XXXHEXY]

@pengrui_han What if you injected a steering vector that represented the concept of snow?

[XXXHEXY @XXXHEXY]

@reconfigurthing The system would probably face an existential crisis. It’s only goal, the number one obsession and hunger was all for nothing. It may even commit suicide as it sees no purpose in existence anymore.

[DAIR.AI @dair_ai]

Banger paper from Google.

If you maintain a skill library for your agents, you might want to check this out.

(bookmark it)

This work separates three things that skill-evolution systems usually collapse into one. Raw execution traces, a persistent wiki of accumulated knowledge, and the executable skills themselves.

Experience gets consolidated into the wiki, and every later skill update builds on that wiki instead of on a scattered optimization history.

Ablations confirm the wiki is what carries a lot of the gain. Two results stand out in particular. Smaller models with evolved skills beat substantially larger models without them. And skills evolved by one model transfer across families, where skills evolved elsewhere sometimes beat self-evolved ones.

Paper: https://t.co/6qftGirTpE

Chat with Paper: https://t.co/rrVzkkR1ij

[XXXHEXY @XXXHEXY]

@yoheinakajima fleeting existence