Fiyi

[Fiyi @Supercech1]

Weekend never even start, Chelsea don spoil am

[Fiyi @Supercech1]

Henderson and Barco pivot? We’re actually finished this season

[Fiyi @Supercech1]

Hmmm idea ball

[cowboy mai laushi 🤠🐎✨ @naftaleethrivin]

we did it, joe

[Fiyi @Supercech1]

Really struggling to believe the hugging face incident happened. The agents that openAI described in the incident report is not the same agent I’m using to work. These agents are from from as brilliant as openAI is suggesting, the reasoning is not that good yet

[Fiyi @Supercech1]

I’m sorry he was a ghost against Arsenal. You guys can’t forgive him so easily

[Ellis🥋 @elIisburner]

This is arguably the best palmer has ever looked for Chelsea

[Fiyi @Supercech1]

@nunumojeks They don’t know MJ in ojuelegba is all I’m saying 🙂‍↕️

[oluok @oluok1]

Good effort Chels but that’s a classic know your place pat on the back loss.

[Fiyi @Supercech1]

@bcherny petition to add “Locked in” as one of the spinner verbs? 🙂‍↕️

[Fiyi @Supercech1]

Our season might actually rest on this guy showing out and staying fit. We need it

[💫 @RoyceCFC3]

Romeo Lavia vs Brighton (H)

First Goal in Chelsea Blue. 💫

@RomeoLavia

[John Ternus @johnternus]

hello

[Fiyi @Supercech1]

Happy CP10 day to all who celebrate! 🎉🥳

[Chelsea FC @ChelseaFC]

Cole Palmer signed for Chelsea #OTD in 2023! 🥶

[Fiyi @Supercech1]

Didn’t realize that trying to determine what your agent finds useful/useless is a very difficult problem, but this approach is promising. I imagine we can use this to properly evaluate how an agent uses tool calls and this can even help improve mcp as a protocol

[Akshay 🚀 @akshay_pachaar]

Google just dropped a banger paper.

If you write or curate skills for your agents, this one is for you.

(bookmark it)

Agent skills are just folders holding the instructions an agent follows for a task, and people increasingly let agents write those instructions themselves.

The loop is simple. Run the agent on some tasks, read the runs that failed, rewrite the instructions, and keep the rewrite only if the score goes up on a held-out validation set.

That last check matters, because an edit that quietly makes the agent worse would pile up over time. So the system always applies the edit, tests it, and reverts the file if the score drops.

Here is where it breaks. Working out why the agent failed is the expensive part, since it means reading full traces and comparing failed runs against successful ones.

But that diagnosis never gets written down. The only thing saved is the new instruction text it produced.

So when that instruction reverts, the finding behind it goes too. The next round reads the same failures, reaches the same conclusion, and often proposes the fix that already lost, with no memory that it was ever tried.

WikiSkill fixes this by splitting the one folder into three:

1. Raw traces, written once and never touched.

2. A wiki of accumulated knowledge, holding what keeps breaking, what has worked, and every edit tried so far with the reason it passed or failed. Nothing here is ever deleted.

3. The skills themselves, reverted whenever an edit makes things worse.

So the instructions can roll back, but the understanding behind them never does.

Their own example makes it click. An early skill gets rejected for being too vague, and instead of vanishing, that rejection is logged.

The next round reads the log, sees the vague version failed, and writes a concrete rule in its place. That one gets accepted.

Same failure, same analysis, but the second attempt starts from knowing what already did not work.

The authors confirm this is where most of the gain comes from by turning the knowledge layer off. The average score drops by about a quarter.

Two other results are worth the read on their own.

Smaller models with evolved skills beat much larger models running without any, and skills evolved by one model transfer across families, sometimes working better than the ones a model wrote for itself.

Paper → https://t.co/ungqaQVTma

I wrote about a related idea (GEPA) earlier, which improves a model by having it reflect on its own mistakes in plain language and rewrite its prompt, with no weight updates involved.

The detailed article is quoted below.

[Fiyi @Supercech1]

Jackson played well, let’s not be annoying 😂

[YAGS @yags_szn]

two chelsea scrubs going gba for gba 😂😂😂

[Fiyi @Supercech1]

I’m a bit worried about the lack of pace up front with Morgan and Palmer, especially when gusto and hato have to be pinned back into a back 5. Could easily become a problem. All in all performance today was okay. 6/10

[Fiyi @Supercech1]

Levi colwill is a liability and the rest of you guys will notice soon enough. Poor decision making and a bit erratic. Lacroix has been there for 5 minutes and he’s easily the best CB we have

[Fiyi @Supercech1]

My childhood love, we’re so fucking back 🩵

[Chelsea FC @ChelseaFC]

Let’s get it.

#CFC | #FULCHE

[Fiyi @Supercech1]

This is really where things are headed. I’m building something similar!

[PostHog @posthog]

https://t.co/yYGp6J8W66

[Fiyi @Supercech1]

Pulling this up when FixDoc blows

[a16z @a16z]

Cursor, 4 years ago

[Fiyi @Supercech1]

As he should be. If I speak my mind on this guys I’m getting banned

[B/R Football @brfootball]

Enzo Fernández returned to the pitch for Chelsea and was booed by the home fans 🫣