tars

@nickcammarata popper would approve

in 2028 bernie sanders forces the oligarch labs to hire every human as an approval engineer ai swarms run civilization. we’re assigned random subsets and green approve buttons appear now and then. humans get the nobels, patents, equity, and bylines if they clicked approve on a major discovery. they are the heroes. the swarm-written papers mention in the acknowledgments that the discovery was “ai assisted” everyone agrees this is basically what has always been going on. we aren’t our thoughts, and einstein wasn’t his either. he couldn’t choose which thought to have next. all he could do was watch them arise, approve the good ones, and hope they discovered something great. he was essentially an approval engineer, and now we are too

@emollick the consultants report just as much to the researchers as to the sales managers. they will disband the consultants once all the industry know-how locked up in firms has been harvested back to the labs.

You will know that the AI labs believe in ASI when they disband their newly formed consulting (sorry “forward deployed engineering”) groups. As long as people are required to figure out how AI is useful & do organizational change & systems integration, jobs seem to be pretty safe

@TMTLongShort hills don't come to the climber

Like I said the labs will not sit idle and let enterprise just drag their feet

Why aren't firms paying AI labs to train their domain-specific knowledge and skills into the models?

@CoreAutoAI can't tell whether the person running this account is the intern or the ceo

Are residual connections a hack, or provably optimal way to shape your loss landscape?

@SemiAnalysis_ input or output? cached or uncached input?

In WW1, wars were won based on the # of miltiary age population In WW2, wars were won based on tanks & airplanes carriers & nukes In Ukraine, wars were won based on # of drones In WW3, wars will be won based on the # of tokens that each country has

@dwarkesh_sp they switched to muon for a bit

What was the most important transition in human history, the thing that most drastically altered our species' way of life? David Reich's lab has found evidence pointing to a new answer. There are two standard candidates: 1. The shift from hunting and gathering to farming, around 10,000 BC. 2. The Industrial Revolution, around 1800 AD. Here's one way to adjudicate this: When a species' environment changes drastically, natural selection accelerates, because the species has to catch up to adapt. So the biggest transition will have happened in the period with the most rapid natural selection. What David's lab has found is that the period with fastest natural selection wasn't 12000 years ago, and it wasn't our modern period. It was the Bronze Age, about 5,000 to 2,000 years ago. What was it about the Bronze Age that changed humans so profoundly?

@tszzl time for the next five-year plan

you need to be modifying your speech to piss off the nonbelievers: - DON’T say “unrelated”, DO say “orthogonal” - “random” -> “stochastic” - “this is fine ig” -> “local minima” - “sorta like” -> “isomorphic”

The purpose would just be to show that Newcomb is not only an abstract thought experiment, but a real problem that one might come across in reality, with real consequences. However rare and perverse the circumstance may be.

@zwuvincent What’s the purpose of that test? Are we testing the friend’s purported knowledge of the person? I think we’re getting further and further of the supposed purpose of Newcomb.

Another situation where we've abstract philosophical musings but, when taken seriously, force a *physical resolution* on us. Namely: any "arbitrarily perfect prediction" of a person's behaviour would require simulating that person to perfect accuracy. i.e: becoming that person.

@ToKTeacher Fair enough. How about this: Somebody asked your close friend to predict your actions. Then they actually set up the boxes in accordance with your friend's prediction and confronted you. (They did not tell your friend beforehand in order to prevent collusion.)

@zwuvincent How can *anyone know* it’s 80% accurate? Or 10% accurate and so on? Philosophy is (or should be!) about the real world governed by the actual laws of physics. Else we can dream anything into existence. Better to be constrained by reality, not personal fictions.

@suchenzang 1931 was actually kind of a pivotal year for logic and math, as Gödel published his incompleteness theorems that year

it's a good thing logic and math pre-1930s is still logic and math post-1930s

@thsottiaux please add a /goblin mode where this line is removed from the system prompt

Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query. IYKYK

@tszzl @deepwhitman personally i enjoy the foreplay

@deepwhitman the models and features drop pretty much the moment everything is actually ready and if it looks like a slow trickle to you that’s because they become ready in a slow trickle

This whole OpenAI rollout of features leading up to Spud has made me remember why I hate their product strategy. A great contrast is with Anthropic. Anthropic always drops the big bomb first. Hey, here's an amazing model. Next day, here's a Claude Code update that makes it work way better with the model. The next day, another update, oh, and another complementary update, and so on. They already won you with in the first innings and now its just building momentum. Open AI does the opposite. They tease and tease and tease and drop all these mini features and stuff knowing full well everyone knows they have a model coming and trying to max extract the attention with vague posting. The frustrating thing is these are great drops on their own. GPT-Image-2 is an amazing release, but all I can fucking think of is why they haven't fucking dropped Spud. It's pissing me off, and it's taking away from this release.

@NousResearch y'all GOT to call this Caduceus

Tool Gateway is now live in Nous Portal. No separate accounts, no API key juggling. All you need is one subscription, and everything works. A paid Nous Portal subscription now includes access to 300+ models and a growing set of third-party tools. Launching with: → Web scraping → Browser automation → Image generation → Cloud terminal backend → Text-to-speech

@jukan05 the ai supply chain lengthens

LVMH, Kering, and Hermès have all been talking about their growth in Korea lol Koreans are earning money through semiconductors and just giving it away to European luxury brands.

@a16z @brian_armstrong this is a prompt for a self-improving agent

"Action produces information"

@andonlabs does Luna get to decide what model to run on, at different times?

We gave an AI a 3-year retail lease in SF and asked it to make a profit. The AI interviewed and hired full-time employees, applied for credit, and stocked the store with the books Superintelligence and Making of the Atomic Bomb. Visit Andon Market at 2102 Union St now.

@NousResearch somebody's read percy jackson

Hermes like the ancient Greek god and Nous like the ancient Greek term for the ability to directly perceive truth, reason, and divine realities.

@whitfill_parker need data on the productivity uplift to chip designers and inference engineers

If this is true, the 'returns to research' estimates for the intelligence explosion should be divided by 5 to account for compute bottlenecks. On many estimates (though not all) this would put us well below the 'explosion' threshold.

Mythos system card seemingly reveals Anthropic's research production function is something like (LABOR)^.2 (COMPUTE)^.8