altcognito

If you ask a model, they will generally tell you where to get data. Modern frontier models have the large advantage of having tens if not hundreds of millions of users providing use cases to train against to improve their responses.

Nothing fundamentally changed when we "discovered" that we weren't the center of the universe, or when we "discovered" that we aren't the only tool users.

There is no prize and no trumpets that will play if humanity agreed we understand what consciousness is or whether or not something qualifies as conscious.

I haven't tried what you guys are talking about myself, but I think when agents begin talking to other agents, the chances for unpredictable goal mixing and confusion resulting in really unexpected and bad behavior is a lot higher.

That being said, the danger is measured in computer damage, which can be a lot personally and to a company, but less existential, so your mileage may vary as to how "scary" it is.

So weird, cause it seems a lot of services are suddenly adding them. Huh, wonder what changed?

Irony is that they will get the benefits from upstream projects (like the Linux kernel) that does accept AI inputs.

It can be done, the US did it for decades, but the budget can't be reckless. You can't ignore the top line forever. I accidentally put spending first in this post, and that was a mistake. The US has cut taxes, cut taxes and while it is true that we've done little to curtail wasteful spending, we've not actually addressed the wasteful part, we've just moralized about "who deserves what"

It is funny that so many comments vascilate between "It is so expensive these companies can't make money and will go bankrupt in seconds" and "Inference is so cheap that these companies can't make money and will go bankrupt in seconds".

I never take them seriously, I just assume they are coming from countries that don't understand how capitalism works or are operating out of bad faith. The underlying reality of the market is always changing and needs are always changing. Some AI companies will fail, that is a given. Remember alta-vista? Yahoo? Did search go away? How about Microsoft phones? Nokia? Motorola?

OpenAI and Anthropic are not in the inference business. That is a commodity. They need to sell products and solutions.

On the front page the counter is up to nearly 4 billion since 1998. The real problem is that we continue to do things to accelerate this, and as we all know, humans are terrible with exponentials.

Wonderful! Careful with any sorts of holes in "gates" within the home to keep animals or children out of rooms as they can get their hands caught in them. That's why the ones at the store have very tiny holes. I just used any sort of solid thin laminate type surface (the ones with a whiteboard like finish are easy to clean!)

Looks good, and great hinges and latches.

We had better have even more time to help each other as a result of LLMs, or what are we even doing here?

I realize that everything seems to be about increasing the pace of life, but this is another "chance" for humanity to figure things out.

This is an area I think I've really enjoyed AI coding: DSLs are pretty fun to make and you can get pretty creative. Bravo on this

I really appreciate the post.

I completely agree that we haven’t even gotten past the first post yet in terms of the magnitude of change that is likely to occur. People fail to comprehend that we’re still assimilating the Internet as a means of communication. Cell phones. It will be decades before what we have today is fully utilized, let alone what is likely to be developed. Anyway, thanks for the article.

And would actually be protected, unlike data centers in space

You can "solve" human biology? For what optimal outcome? We can all be Elons running around trying to procreate as much as possible? What is the optimal human model? How is human biology NOT going to become anything except the same distorted race to the bottom that everything else has been? We've not solved the mental model of survival-of-the-fittest.

The best analogy for a positive outcome I can come up with is living within our means as a species. Leverage AI, leverage all the tools we have for production, but we have to find equalibrium with ourselves and the universe we live in.

I'll vote for the terminators and meteors if we insist on living beyond our means.

You can't have your cake and eat it too.

Fundamentally the problem with living forever goes beyond billionaires. People get stuck in their ways of thinking, the mindset of living forever is completely different. Why should I even listen to someone who only lives a mere 40 years? What is a suitable punishment for someone that lives forever? How does it change murder?

Philosophically, living forever may be corrupt by nature.

The only way to tear down tiers of society is for some of those tiers to literally die off.

Probably should have left off the kids, but lets start with just distrusting anybody who wants to live forever. At the level of influence billionaires have, it is downright dangerous.

For everyone else confused: Think of all the people throughout history we would prefer would not have lived forever. Then multiple that by A LOT. Then consider how greedy and sociopathic most of the billionaire class is already.

Now, we could spend time getting distracted by childless. I don't think it matters.

I keep repeating this too.

The history of the self driving car is apt to repeat as well. I remember thinking "2015, 2018 maybe at the latest." But as 2015 (and 2018) came and went, expectations were tempered.

We will see shortened development cycles, and maybe even orders of magnitude shorter development cycles.

But hard problems remain hard (there are only so many chips, materials research and biology are difficult problems). Manufacturing remains a choke point.

I'm more concerned that the absolute worst persons on the planet are in charge of many of these technologies. If those that are in charge are more concerned about allocating resources and power in their favor, the more dismal the future of all humanity will be (including for those in charge, but they can't see that due to self-interest bias)

I mean sure but reality is pesky. Most ads buys are targeted. Once a company has accepted putting ads in, difficult to know when they've started allowing information gathering. Technically it is already the case that companies sell targeting information before they accept ads even.

Agreed, especially since the more frontier models are able to accomplish in a vacuum, the more people will trust them. That being said, tool use is still really important for pulling in the right information.

It needs influencers!

I would describe it as "fidelity" and "verbosity" (or just amount of token generation to complete the task, sometimes that works out to scratch space, or literally how large the "solution" is).

If you have something that needs to be done right, might be a bit complicated, up the model size.

You can see this in the pelicans. Big model pelicans are pretty accurate by default. Up the reasoning and only more so, but with more detail. For Astra, it is 105 lines for low, 250 lines for max reasoning.

Small model pelicans will lack the fidelity of a large model. Bits will be out of place etc. For luna, it's 90 lines for low, 150 lines for xhigh.

Additionally the amount of time taken is increased for the larger models. Luna takes 11 seconds on low, and 1:33 for xhigh. Astra is 33 seconds on low, 4 minutes on max.

And naturally, there is the cost. There's some overlap in functionality between luna xhigh and Astra low in the sense that luna really can do quite a suitable job for some tasks. But there are just some tasks that just don't make sense for Luna, even at high reasoning.

The other thing to remember is that sometimes high fidelity isn't ideal. It can lead to overdesigning. My recommendation is to commit early, commit often, and review everything you do, which we've all been doing since before LLMs right?

Keeping people looped into your product is pretty important, but yeah, there's not clean way currently to separate "structured" outputs from the token stream and to start using a different billing structure there. And I also appreciate that they aren't going to be keen on gving free or near free output either, so gotta figure that.

Can't have ads and tell me you're privacy focused.

I don't see it fundamentally any different than knowing when to use a tool. Is this tool like RAG an important enough corner case to train for it? I dunno.

LLMs already shell out and write code to solve certain problems. This is just a special case of that.

Yeah, it's not just over-engineering, it is just that generic systems that look like everything else usually don't solve a business problem, but I think I would be less dismissive of its world model in general.

LLMs are not a "random walk", they take in information and they explore the space according to the way they've been instructed.

You should definitely try it if you haven't already. Most of the models are deeply familiar with domain specific areas in ways that most of us aren't. In fact, I would say the problem is generally the opposite. If model size and effort is is high, it will over architect what an app needs. I find myself reigning in a giant email notification system with subscriptions and "channels" and other such nonsense when sometimes you just want a simple one off.

Something we can all agree with is we need more foundries and green power.

Thanks for your time.

I tried the demo, and I ask it modify the game (hitting the w button immediately proceeds to the flag) and it doesn't do it but does something else. Is that the desired outcome? I think the desired outcome would be "What you're asking for doesn't make sense given the rule."

They compared against crocodilian teeth and those teeth were at a different temperature. Crocodilian teeth are known to be at a similar temperature to their environment because they are known to be cold blooded.

It is correct to say that we don't know they were warm blooded, just that their body temps were likely the given observed temperature.

I was unaware that Amazon even sold single AZ S3. 20% discount. Doesn't seem worth it. By the time I commit to purchasing S3 space, it has to be important data.

I get that S3 is convenient and reasonably performant, but it is not cheap at all.