Oh, you're an AI safety expert? Name every LessWrong post.

Writer's Block is a desirable problem.

Of the opinion that terms like "email job," "knowledge work," and "work you do on a screen" should be retired. The first two were basically defined by alt/far-right actors as "jobs that inherently lack value."

The way some people talk about "email jobs" makes me think: (a) they believe entire sectors of the economy are just people checking in and circling back: (b) for jobs that *are* light on workloads, that employers haven't noticed, and it would take a profound tech advancement to make them see this.

This idea that data centers are like mini-Suns is interesting, because I think it's one of those things people believe only in the abstract (like lots of things that proliferate online), but people's electricity prices are also going up significantly, so it gets intertwined with very real politics.

Terrific work with important lessons for the perils of brain-LLM alignment research 🧠 ⚡ "Stimulus dependencies—rather than next-word prediction—can explain pre-onset brain encoding in naturalistic listening designs" elifesciences.org/articles/106...

Somebody should interview Don Knuth to get his post-shock feeling on Claude and co. Curious if this has led to him using it regularly, or if the shock was a one-off.

Time is a flat circle. Google will be forced to withdraw so that Palantir takes its place. Forever and ever.

Autism? But I hardly know him!

Did the self-organizing autonomous Moltbots cure cancer yet

"Revenue share payments from OpenAI to Microsoft continue through 2030, independent of OpenAI’s technology progress, at the same percentage but subject to a total cap." openai.com/index/next-p...

Turns out I seek LLM content (I'm the problem), so I'm gonna be a little more Galaxy Brained about this

Interesting discussion within. I might have to change my view on what exactly "next token prediction" means, but I'd maintain that we should think of pre-reasoning LLMs as distinct from reasoning models, just to get our grips on what's changed since 2022.

This was fun. I scored literally below random.

This is complicated because there are a lot of misunderstandings on both sides of this question. A common misconception on the other side is that with post-training / RLHF, at inference time they are doing something other than generating a sequence of tokens one token at a time,

I also think people are largely unaware of just how extensive this particular effort is. If your mental model is an LLM's training data is the internet, I invite you to look into this use of custom, human generated data.

I dunk a lot on the pro side. One thing I would like to see curbed though is the idea that LLMs just predict the next token.

Yes, there is a war happening in the Persian Gulf region and yes, this war is choking of energy supplies. Yes yes yes, there's also a war in Ukraine and a civil war in Sudan. But I think the biggest issue today is a hypothetical machine whose consequences are speculative. Get on my level.

Yeah, I study consciousness. My favorite philosopher is Blake Lemoine.

I think this is a good thing to be thinking about - far more important than the usual tech stuff a number of us are usually concerned with.

I just find this bizarre

Mood www.youtube.com/watch?v=od_P...

I feel like I'm going crazy seeing Meghan McCain make this about the ballroom seemingly before Trump did

An enduring confusion for me is trying to understand why a particular computational model can itself serve as an explanation, and how anyone could even adjudicate between models that offer comparable performances, if the performances are all that matter.

My best suffer from being references to things only 10 people would ever care about.

I cannot believe people saying white collar work is in trouble because of AI. As if they all wear the same color shirt.

Writing is back in full swing. The previous 2-4 weeks have been mostly burnout (from other things) with only brief reprieves. Some real bullshit frankly.

An underrated possibility is that the energy crisis which we are coming to know and love does in fact curb or undermine data center expansions (at the least, bringing them online), while increasing the importance of open-source and/or locally run models.

Shocking the degree to which debates about the usefulness of AI are entirely at the meta level.

Francis Fukuyama famously predicted that the end of the Cold War in fact signaled the end of history itself. It was not until years later that Fukuyama's thesis was refuted, when physicists demonstrated the passage of time.

I didn't get DUIs until I started using AI agents: Delegating Under the Influence

I promise you, we are not ready for what's coming.

Me? I guess you could say I'm something of an intellectual. When I encounter new information, i consider it my solemn duty to inform you of your wrongness 😌

Guy who thinks LLMs have free will but humans don't

Mood

In case anyone is interested in the US Census Bureau data on AI adoption that was published today, my newsletter has a summary post. Link: macromostly.substack.com/p/an-update-...

I'm really interested in how the mind works. My favorite theory is Claude 4.6

Cognitive science? The field that studies proprietary NLP products?

Shitposts too cheap to meter

I am experiencing a rapid onset of Aging as a result of Current Events

There's a male loneliness crisis and you want me to pay my student loan bill??

You (a tech bro): thinks in terms of "capabilities" and says "a robot will do my laundry." Me (a man of the people): understands that shared coin-op laundry is the true bottleneck.

Gonna keep this brief and stick to the business angle: you don't leave money on the table during a gold rush.

Yeah, I'm a Knowledge Worker: K i N think O W therefore L E i D G am E

Took off today cause god damn

Yeah, I'm into LMs: Large Marge

I have thought up a shitpost so powerful, so menacing, that I cannot release it to the public.

Basically a nothingburger IMHO. whitehouse.gov/wp-content/u...