Barry Kelly

@rickasaurus It would be stuffed with almonds presumably

If you were going to invent a new type of sausage called a Mandelbrat what would it look like

@rickasaurus FWIW, I don't recognize Steve's description in my org.

@fchollet The graph isn't quite a random walk; you've effectively added momentum to it ("persistence"). That probably models stock movements a little bit better, but it's not a pure random walk.

Ok, this thread has apparently been a magnet for hordes drooling morons who not only don't get stats, but can't even read. If you're a normally intelligent reader of this tweet, here's an extra example of my point: If you take two RANDOM, INDEPENDENT timeseries (i.e. knowing one gives you NO information about the other) that are each highly temporally autocorrelated (e.g. two random walks), and you plot one against the other as a scatter plot, what you get is a single X/Y trajectory that will ALWAYS look very structured. Yet it is random. Like the figure below. Code to reproduce the figure and play around with this idea: #scrollTo=MACHekU7eirD" target="_blank" rel="nofollow noopener">colab.research.google.com/drive/1FIk8j_A… Of course if the two series happen to be correlated, then you will ALSO see something very structured. It's just that this type of visualization is a completely retarded way to look at such data. If you think this is deep, you are innumerate.

This is a type of chart crime: when you plot value tuples from a timeseries on a 2D scatter plot like this as if they were independent samples, you are leveraging temporal autocorrelation to magnify any existing x/y correlation and hide the actual variance. You're also hiding temporal distribution drift (the typical value range for x and y, and the relation between them, change over long periods)

@patio11 Heroics being bad or not has nothing to do with compensation.

One part of the heroics discourse is “Employers do not successfully reward heroics and therefore expectation for them is breach of the implicit contract”, about which I feel “… It depends. How familiar are you with tech compensation structure.”

This is both a really rough adjustment for individuals/orgs and also a culture one can reasonably opt into or not.

@patio11 Heroics suppresses a pain signal that an organization needs in order to adapt. It can be different if heroics are being applied to maintain a steady state, but heroics can inhibit adaptation when things are ramping up.

(I have a complicated relationship with heroics, and to coin a phrase, the optimal level of heroics is not zero.)

@mentalgeorge Society would fall apart if people were not mostly moral. This isn't an either/or situation. You need both: hard rules to guard against deviance and set a perimeter on permissible action, and fundamentally moral action so that rules don't need to be harshly illiberal.

I'm also partial to old-school liberal critiques of the Claudian approach. Empowering an agent to act on the basis of its own reasoning rather than the letter of the law is dangerous business. This is precisely the view that, in Isaiah Berlin's words, empowers one "to ignore the actual wishes of men or societies, to bully, oppress, torture them in the name, and on behalf, of their 'real' selves."

More on OAI's model spec. This time, I'll cover the section that most distinguishes OAI's liberalism from Anthropic's Claude-as-philosopher-king approach. #2 - "No other objectives" Almost infinite material for discussion here

@0xluffy It's ironic that they talk about releasing it early for cyber defense, and yet the leak seems to be down to poor cyber defense practices.

so Anthropic been training a new model “Claude Mythos” and claimed to be “by far the most powerful AI model” they’ve ever developed - CMS error exposed 3,000 unpublished internal assets - blames human error in CMS for the leak - new tier Capybara 🤨 larger than Opus - scores higher on software, academic reasoning, cybersecurity - claims that model is far ahead of any other AI models in cyber capabilities, can exploit vulnerabilities that outpace defenders - cue “too dangerous to release” again Anthropic is afraid of releasing a model they built themselves. but OAI said the same thing before i wonder

@full_kelly_ @eliebakouch It's not token efficiency, it's move efficiency, to guard against brute force strategies. Models can generate as many tokens as they like.

@eliebakouch Cursed benchmark TBH. The score is about token efficiency relative to humans which isn't very meaningful.

this gets even better when you look at the y axis

@VictorTaelin API prices have huge margin. The plans almost certainly don't subsidize anyone except the heaviest users; most people won't max out their quota day in day on a weekly basis.

(also opus 4.6 fast isn't "15x more expensive" than normal mode (2.5x speed * 6x price) - it is actually more like ~200x more expensive, if you consider how heavily subsided the claude pro plan is right now...)

@paulnovosad @Scholars_Stage It's more than that. The more ICE is reviled by the people, the more ICE members need to side with their masters and obey them, to save their own hides.

@Scholars_Stage Isn’t it obviously because the chaos and escalation is exactly what some people in the administration want to see? Every bit of violence gets us closer to a Reichstag fire.

Here is what I don’t understand. I saw probably 50 people make this point between November 2024 and January 2025, including some people who went in. So why did it not happen? I assumed they discovered legal hurdles that made it difficult—was my assumption wrong?

@Scholars_Stage You're mistaking the goal and the means. The goal isn't deporting immigrants. The means - an army of violent thugs - is the goal: having what effectively amounts to a domestic army with blood on its hands, so that it needs to side with the authoritarians, and not with the people.

@EMostaque You want to be looking for côte de boeuf, not tomahawk steak. Try Bouchon Racine. If you want a steakhouse, I can't save you.

Where is the best tomahawk steak in London

@Miyhnea It can go the other way. It can be that incumbents in a more functioning country can leverage their weight to suppress competition and protect monopolistic rents.

@rickasaurus Hitting the ground running as a senior IC can be tough if the tech stack is different. And some tech stacks are very different.

Accepted a job at a big company today. First time at a Fang after 25 years at startups. Wish me luck!

One Tuesday night, as my wife and I sat down for dinner, a sheriff’s deputy knocked on the door to serve me a subpoena from OpenAI. I held back on talking about it because I didn't want to distract from SB 53, but Newsom just signed the bill so... here's what happened: 🧵

"If you’re seeing a high number of gunshot wounds to the chest area and the head, that’s not collateral damage — that’s deliberate targeting."

@rickasaurus I'm not sure. Broad shallow software is at risk; stuff you go to do a one-off or simple task. You can craft it quickly. But software that is deep, that multiplies in value with each feature addition, isn't going away, especially if the data isn't easily transported.

It’s dawning on me that the era of building generic systems is over. Code is cheap now, build hyper specific systems quickly. Reuse matters much less.

@dnlklr @trouble_man90 These are reflections in the glass in the window.

tree shadows shift from 17 sec> 21 sec... my guess is they just cut out a part and warp transitioned

It should be a major deal that the White House posted an AI video of trump’s Oval Office speech. How much of it was AI? All of it?? Was he even in the room?

@rickasaurus Honestly, it's also kind of vegetal. A little bit earthy / grassy. That can be a discordant note when you bite into your mochi ball.

If you’re wondering it’s kind of reminiscent of fig jam. It’s a really nice flavor along side ice cream or pastry.

Americans are missing out on red bean. 90% of people won’t eat it because “eww beans are gross” but it’s legit one of the most delicious dessert ingredients ever invented.

@drgurner @jburnmurdoch Well first you bemoan lack of solutions, then you say who cares about solutions? I'm as depressed about it as the next guy, but it does unfortunately look like the liberal way of life will be a fairly short blip.

@barrkel @jburnmurdoch Who cares.

NEW: Progressives have a birth rate problem For all the talk of a general fall in births, the drop is overwhelmingly driven by people on the left having fewer kids. By ceding the topic of family and children to the right, progressives risk ushering in a more conservative world.