stormfather

Can you give me examples of services you think DOGE will cut that the economy depends on? Genuinely asking. I assume we have different mental models of what will go away.

When I think of spending cuts, I think of: USAID (23B), that rural fiber project (42.5B) and the electric charging stations (7B), stuff like that. I have a hard time imagining that these are net economic positives, but I am sure you have different services in mind that would be worse to cut. My impression is that the government is actually full of waste. I certainly don't feel like the citizens receive 8T of value in return for the taxes we pay.

I am more cynical. Congress is controlled by lobbyists which are controlled by oligarchs. Voters are controlled by the media which is controlled by oligarchs. I don't see us raising taxes significantly. Wishful thinking.

I also think we should try to cut spending BEFORE raising taxes. What seems better? Less waste and less tax or more waste and more tax? What is better for the economy (for the organization of labor and capital) in the long run? What will attract foreign investment? What will allow accrual of consumer savings?

DOGE is positive because we need a wrecking ball to come in and cut spending because the system is rotten and Congress and the media don't serve the people anymore. They plumb won't do it. I honestly couldn't think of a better personality profile than an idealistic 25 year old engineer who is young enough to hate the government and resent being born 100k in debt as much as I do. Who hates housing prices and health care and the MIC. Someone relentlessly logical whose raison d'entre is optimizing complex systems. At least they'll do something about all of this. Waiting on Congress to do something against the will of their masters is a losing game.

Ok so let's raise taxes on the wealthy. I think that's great and am in total agreement.

How long do you think we should wait for Congress to significantly raise taxes on the rich? How confident are you that will ever happen? Have you seen a graph of the national debt? How long do you think we have to wait? Why oppose cutting spending now? That's not in conflict with raising taxes for the wealthy, both should be done.

Do you see a popular mandate for raising taxes materializing? Do you see Congress enacting that? I don't. Like you said, you're not the first to think of it, but it never happens.

As imperfect as DOGE is, its the only measure I can see actually happening that has a shot at cutting the deficit. I would love to think of another one.

Say what you want, but if we don't drastically cut spending, our way of life will collapse. It would be nice if Congress itself would reign in spending and downsize the government, but everyone knows that has a 0% chance of happening.

This is better hyperinflation or a violent revolution, at least, which is what it avoids for us. Anyone who doesn't see that hasn't done the thought experiment of extrapolating out our spending for a few more decades. It can't continue, period. Period. Our only choice is how to change it and our democracy (congresspuppets controlled by lobbies) fundamentally cannot fix this. Does anyone have a better idea that will actually work?

Awesome! I write LLM powered scrapers and stuff all the time and one of the biggest pain points is HTML is full of so much crap that isn't meaningful and overwhelms the context. And being a data science guy idk how to solve this.

> As well as an algorithm for efficiently converting/compressing large html pages into a semantic format.

For the love of humanity please open source this. This seems tremendously useful by itself.

Who cares? It's a name. Let's talk about how he wants to move all the Palestinians to Egypt and Jordan.

Google is clearly catching up. Have you tried the recent Gemini models? Have you tried deep research? Google is like a ship that is hard to turn around but also hard to stop once in motion.

I wasn't clear. Model weights aren't changing at inference time. I meant at inference time the model will output a sequence of thoughts and actions to perform tasks given to it by the user. For instance, to answer a question it will search the web, navigate through some sites, scroll, summarize, etc. You can model this as a game played by emitting a sequence of actions in a browser. RL is the technique you want to train this component. To scale this up you need to have a massive amount of examples of sequences of actions taken in the browser, the outcome it led to, and a label for if that outcome was desirable or not. I am saying that by recording users googling stuff and emailing each other for decades Google has this massive dataset to train their RL powered browser using agent. Deepseek proving that simple RL ca be cheaply applied to a frontier LLM and have reasoning organically emerge makes this approach more obviously viable.

Google is silently catching up fast with Gemini. They're also pursuing next gen architectures like Titan. But most importantly, the frontier of AI capabilities is shifting towards using RL at inference (thinking) time to perform tasks. Who has more data than Google there? They have a gargantuan database of queries paired with subsequent web nav, actions, follow up queries etc. Nobody can recreate this, Bing failed to get enough marketshare. Also, when you think of RL talent, which company comes to mind? I think Google has everyone checkmated already.

Layman here. Isn't the big speed penalty in digital electronics the clock speed of the transistors changing state? Not the actual signal going down the wire from gate to gate?

How does Kagi Ultimate compare to Perplexity?

I made an objective test for prompting hacks last year.

I asked gpt-4-1106-preview to draw a bounding box around some text in an image and prodded in various ways to see what moved the box closer. Offering a tip did in fact help lol so that went into the company system prompt.

IIRC so did most things, including telling it that it was on a forum, and OP had posted an incorrect response, which gpt was itching to correct with its answer.

I see why it gets a lot of flak too. People are becoming increasingly neurotic and fragile, and coupled with that we are increasingly unable to fluidly deal with differences this creates as they arise. We're talking about playing a boardgame with friends, do we really need to have a trigger-warning session? And when does that end? Every time I hang out with my friends should we update our daily list of no-no words first?

Sorry, what? Not wanting someone to poison themselves is an expression of love

Not necessarily. Say you can make 20% a year with your system. That's very respectable. But unless you have a large account already (1M?), you will make more by providing this knowledge to others.

What brought you there?

Excuse? You must not follow chess too closely. He is the undisputed GOAT. He is clearly bored - he plays atrocious opening moves these days just to get an interesting game. He's so good he transcended the need to keep proving it. Excuse. Lmao. Gukesh is the WC only because he is not good enough to present an interesting challenge.

Or testing our own response. If we want to test what China's response might be, both public and military, but its too provocative to try, we might try it on ourselves.

I agree! It's a shame you're being downvoted just for expressing your opinion. Come on HN, downvoting is not for expressing disagreement...

And yeah. People saying this was exciting chess are lemmings. It was absolutely not. It was yet another boring draw-fest. The format incentivizes prep and penalizes creativity and risk-taking. If my child got very good at that it would be hard to be proud. What a waste of human spirit! Why would I want to watch two extremely smart young men waste months of their lives on this for the sake of boring us? It's perverse. Let's move on to Chess960 already. THAT would be exciting.

Absolutely. It would be AMAZING if no prep was possible. Memorizing engine lines 20 deep is nothing to be proud of.

He's also not really the world champion. The world champion just got bored of winning so hard.

And they have the absolutely massive advantage of being able to associate content with queries that led to it, and know which piece of content was selected by the user. That surely can be used in some way to give them a leg up with both choosing good training data, and making for o1 type agentic models.

I think they will though, I think the enormous corpus of video data and the supercluster that powers self driving development are the machine vision analog of internet scale text data that gave rise to LLMs. We'll see the same moment for vision models that text prediction models had once the data is there, where an enormous foundation model becomes much much better, especially at zero-shot tasks.

I would guess the plan is to have the foundational machine vision tech that becomes the core of robotics sensors. Not just Optimus but every robot arm in a factory, robot mule, etc. I don't think everything will have LIDAR if its proven to be unnecessary.

Also he wants to reuse the foundational machine vision tech in Optimus bot, which probably won't have lidar.

Yeah exactly. Shooting hundreds of bullets with your bullets. And each bullet you need to shoot breaks into smaller bullets upon re-entry that start randomly maneuvering. Hit to kill is just not practical against a peer firing hundreds of ICBMs.

That's why I suspect the US never stopped developing nuclear tipped ABMs with the Sprint I and II programs. They just went dark. It's the only logical way to defend against the only existential threat we face, and DC is gonna be first struck so you can bet the political class is incentivized to develop this.

Why keep it secret? Well first it interferes with MAD which is very destabilizing. Also its good to hold your most important cards close to your chest. And there's the fact that the public would be uncomfortable with the idea that our plan to survive involves so much radiation being thrown around. A lot of incentive to not talk about it...

I bet there are nuclear tipped interceptors too, though that wouldn't be public. Its a lot easier to destroy the incoming missile if your margin of error is a few miles versus a few meters. Not to mention it would blot out all the multiple re-entry vehicles. The public wouldn't want to know that's the plan but it would sure beat nukes detonating in cities.

I even remember reading about how the Soviets planned to chain detonate nukes over Moscow in the event of an exchange to destroy incoming ICMBs.

I choose LeakyReLU vs ReLU depending on if it's an odd day of the week, LeakyReLU being the slightly favored odd-days because it's aesthetically nicer that gradients propagate through negative inputs, though I can't discern a difference. I choose sigmoid if I want to waste compute to remind myself that it converges slowly due to vanishing gradients at extreme activation levels. So its empiricism retroactively justified by some mathematical common sense that let's me feel good about the choices. Kind of like aerodynamics.