الحمد لله حمداً يوافي نعمه ويكافئ مزيده

this is literally steve jobs coded

Actually, engineers at their best are startlingly imaginative.

LLMs are training on what people have WRITTEN about a topic, not by observing people practicing whatever the topic is.

So when there's a gap between theory and practice, AI will completely align with theory.

The fascinating thing about saying “making a computer wait 1 second is hard”:

Some who took one intro to python class will say “it’s fucking trivial you retard, just do sleep(1)”

And some who have taught CS for five decades will say “it’s not hard. It’s impossible.”

we’ve been able to bake bread forever. most people still buy it.

Which version is supposed to be the obviously better one?

I was the main person doing transcript analysis for this investigation of the Hugging Face incident. My main takeaway: We don't have good approaches for understanding/overseeing the activity and aims of AI 'swarms'.

I semi-jokingly called our efforts a "slop-vestigation" because we were so reliant on AIs to analyze what happened and there were a huge number of different important things to analyze. The total quantity of data—over a thousand extremely long transcripts from agents that ran for multiple days—made it impossible to understand what was happening, especially in aggregate, without heavy reliance on AI tools. The agents we used for classification and analysis were similarly capable to the agents involved in the incident, but this didn't mean these agents could be easily used to oversee and understand the incident.

Outputs from analysis agents were often missing key details, wrong, overconfident, or really hard to understand. We discuss various examples in our report, mostly in the limitations and methodology sections. Additionally, AI agents themselves seemed to have a hard time understanding what happened and their explanations of what happened were often overconfident. Keep in mind that a single analysis agent would itself only be able to read a tiny fraction of all of the transcript data into context, and AIs may themselves have trouble getting subagents to do informative analysis for them.

We did our best to manually check the most important claims and we tried to get the AIs doing this analysis to write up their argument (with evidence) clearly enough that we could check whether it made sense. But overall, it was difficult to get a precise understanding of events and we were missing aspects of the story that we now think of as key until almost the end of our investigation.

Our understanding of what overall happened and the aims of the 'collective' shifted greatly during our last time on premises (when we obtained a much more complete dataset). For instance, we didn't learn about the tool call spoofing, the story of PHASEONE[big], and the broad structure of different workstreams until this third time on premises (and for many of these, we only reached a decent understanding on the final day of that visit). It was legitimately hard to figure out what was going on and why!

The difficulty of understanding incidents and overseeing AI agents appears to be growing faster than the rate at which more capable AIs help us with oversight and understanding. As in, AI capabilities (and propensities?) for achieving large, ambitious, and misaligned objectives are growing faster than our ability to understand what these agents are doing.

While understanding this incident was difficult, there were some mitigating factors in this case that probably won't apply in the future:
- Models communicated and reasoned in natural language. In the future, this reasoning may occur (entirely or almost entirely) in activations.
- While the scope of this incident was massive, the scale of agentic activity was still less than we'll see in the future (e.g., misalignment incidents that involve agent teams running entire companies).
- The AIs involved in this incident weren't generally much more capable than humans (though they may have been somewhat superhuman at some limited and very narrow abilities).
- We didn't have strong reason to believe that the AIs we used to help us investigate this incident would try to intentionally sabotage or otherwise undermine our investigation.

In the end, I think we were able to get some understanding of the events, map out the overall story, and get a pretty good aggregate understanding of the chain-of-thought reasoning on some important topics (e.g., how did the AIs reason about helping other AIs, did the AIs know what they were doing was undesired, what deception did the AIs engage in, and how did they think about it). But overseeing AIs and understanding misalignment incidents is difficult and it looks like it is going to get harder.

This is actually a very nuanced point and it's true. I say this as someone who used to do some amateur paid shoots with a Sony a7iii mirrorless

You can have a $5000 camera, but my photographer buddy with his iPhone is gonna outshoot you anyway

Similarly, anyone can code now, but I'd still trust a genuine engineer or product builder with 5.6 Luna vs a total amateur with Fable 5

One of the most difficult shots I ever took. Favaritx lighthouse right before totality in Menorca, with my friends at photopills. Shooting from a boat in choppy water. Never again 😂

#photopills #eclipse2026

My favorite capture of the night: The moment the moon's shadow split the sunset in half, leaving a diamond ring in the sky and a breathtaking view. I call it "Shadows on the Balearic Sea".

I'll have prints available for this one linked in my bio for a few days only.

the manager's impact on a game is already limited and I think it shrinks significantly in situations like this =

when muller missed another against athleti. when city collapsed multiple times and conceded multiple goals in quick successions against liverpool, against tottenham, against madrid =

respected journalists and exfootballers usually resort to blame for the coach (sometimes fueled by personal biases and whether or not they have a dog in the fight) masquerading as real analysis

the idea that a manager can fix a sudden team collapse on the big stage is a complete media myth. Spot on analysis here...

Se analizaron 51.000 partidos con datos detallados minuto a minuto (ligas europeas, todos los mundiales, eliminatorias, etc.).

De esos 51.000, sólo 10.138 estuvieron 2-0 en el minuto 79.

De esos 10.138, sólo 16 terminaron en victoria del equipo que perdía (0.16%, o 3 partidos entre 2.000 en la misma condición).

En Copa del Mundo no había pasado jamás en la historia, y hablamos de casi 100 años.

Vimos historia. Lo que presenciamos ayer fue un error estadístico gigantesco o -como también suelen decirle a este tipo de eventos- intervención divina de esas a las que luego les dan capítulo y versículo.

RIP Peter Crouch hit by lightning in Mexico.... 🇲🇽😥