I recently canceled my Claude subscription. The token costs are outrageous. I maxed my 5 hour limit with a single request. If they fix that I'll reconsider but as things sit today Claude is a pig that's just not worth the cost.
The technology already exists now on the algorithmic front for the next 10x drop between everyone adopting DeepSeek's MLA, MoE (mostly already done), Medusa (a better version of Google's speculative decoding), Kimi's Attn Residuals, and Mimo's Sliding Window Attn, and (possibly) Microsoft's 1.58b (this may be a nothing burger).
Historic trends, every 18 months, performance for the same level of quality has gone down 90%.
Historically, algorithmic gains are only ~30% of the pie, but there's enough out there to get to 10x, with just what's available already. The other ~70% of the pie is better training data (often synthetic) and distilling frontier knowledge. There's no sign we are tapped out on that front.
Additionally, GRAM (from ~10 days ago) is likely to be a 5-10x on its own (if not substantially more for smaller models). It's unlikely within 4 years LeCun's JEPA ideas and similar ideas like GRAM applied to LLMs have ZERO impact. The preliminary results are absolutely astounding (5000x better reasoning - this is not peanuts).
Further, that's not even counting that cost per watt is still dropping ~2x every 2 years on its own on the hardware front.
If you look at the "cost" of inference. People think it's electricity - but it's currently almost ~80% hardware amortization. The memory shortage is not going to last, nor are Nvidia's ~80-90% margins.
The human brain is still 8-10 orders of magnitude more efficient than the best LLMs of today. With ~1/10th of global capex riding on AI, if you don't think they're going to knock of 2 orders of magnitude more, when it's this obvious and easy... I don't know what to tell you...
Sure, it might take 6 years instead of 4. My crystal ball isn't perfect.
Sure, the price will come down a lot, even if we can argue about the timeline.
I think what will also happen, once we get past this current CEO AI FOMO mania, is that companies will start to look at AI spending more rationally like any other company expense, and will revert to more rational decision making.
Even if the cost comes down considerably over the next few years, that's plenty of time for companies to look at their financial results and question why AI expenditure isn't resulting in increase in revenue and/or profitability.
Additionally, on the context front -> all the labs are aware that for many tasks you can get 10x+ increases in output quality by feeding better context.
The closer people live to the consequences of their decisions the more rational they become. Until leaders(and I use that term loosely) are held accountable, the insanity will continue.
As long as our stock price continues to... Continues to rise... Which... Hmm... I'm just now reading our balance sheet. Is this number right? Great, thanks.
In addition to being true, this observation is profound. When designing any multi-step system that relies on humans making decisions, whether in governance, organizations or economies, placing root causes as close to end effects as possible is almost always better.
I’m sorry you are used to working with out of touch leadership. Not all companies are like that. Even big ones can have smart, empathetic leaders. Although very often money gets in the way of empathy.
Money alao has the problematic tendency to warp the people around you, it's its own kind of gravity. The more powerful you are the more you attract yesmannerism and the more you lose touch with what's going on.
Also notably these attributes don’t make one infallible. I see a lot of engineers judging from the sidelines without any sense of how to run large orgs and how you have to make tough calls with imperfect info all the time.
It’s not coercion. You’re free to not use it, or alternatively do what these folks did, write your own. Coercion would be forcing people to use it through some mechanism, which clearly isn’t possible with GPL.
Not always no and to your example with cars we've seen the results as upticks in roadrage. The car is treated as a safe little bubble and the other cars aren't people, they are just cars and what you do and say to them doesn't matter. Just like the internet where they aren't people you're talking to, its just text on a screen.
Technology has drawbacks, the question is are the drawbacks greater than it's benefits. Part of the answer is personal, some people can handle them better than others. Other parts are societal, what's the impact on society of the people that's can't handle it (mass shooters, roadrage, suicides etc).
I was thinking the other day about animals in their natural habitat versus in captivity. Remove a gorilla from its jungle and stick it in a small zoo enclosure and it tends to go insane at worst, at best depression sets in. With orcas their fin flops over and even when released back into the wild it never returns to form, we can only guess what happens to them psychologically. Humans in supermax prisons exhibit the same issues.
I think we're seeing some of this with people today due to doom scrolling and sedentary isolated lifestyles our technology is creating. AI is perhaps the final nail in the coffin for some as they genuinely treat these chatbots like they are friends and confidants and lose human connections to the real world.
Just look at how people behave these days, it's hard not to notice the widespread mental illness epidemic that has set in and seems to get worse daily. We've created little prisons for ourselves and locked the door. We're losing human connection in real time almost like people are willingly submitting themselves to the Matrix.
> We've created little prisons for ourselves and locked the door.
Yes, but that kind of ignores the more active elements. This story didn't languish with few upvotes and a few shallow comments, it was an instant HN darling so had to be killed, or it would still be on the front page.
It tells you a lot about your execs and how little they care, either for their employees or their customers. The quarterly profits are their God and they will worship at the altar of the stock price.
Instead of finding ways to make AI enhance their employees and make them more productive, they immediately jump to ways to eliminate employees. It's the opposite of a growth mentallity.
I'd love for these executives to show me a time when investing in people was the wrong choice. I've never seen a company punished for doing the right thing, caring for humans and providing a good work environment. This suicidal tendency in the corporate world to constantly decimate your workforce every cycle is just mind boggling and the fact the stock market responds to it so positively is horrifying.
They have no reason to care in globalized cultures that are morally bankrupt and have no sense of citizenry to feel any amount of allegiance to. Business leaders were never perfect, but things are at least different when you have a sense that when you treat your employees and customers unfairly that you are treating your extended friends and family unfairly. The atomization of everything means that sense that your business is a part of your own community is pretty much gone. In a society with a decreasingly coherent morality, nothing matters more than cash flow, and there are many ways to make cash flow besides making a good product at a fair price. In an immoral society, leadership benefits from attributed success but suffers not from its failures. Society has given up on accountability beyond a certain scale. The petite bourgeoisie might be punished for misleading the public or screwing its employees, but beyond that it seems we let leaders get away with quite a bit.
And why wouldn't they want to eliminate employees? That's their wet dream! Many business leaders don't see employees as their asset. To them, employees are a necessary evil. If anything, the employer-employee relationship is inherently adversarial. The idea that C-level execs could one day simply talk to an AI and, boom, there's a business with cash flow and no employees, is too attractive for them to pass up, even if the chance is high it doesn't work out. At a personal level, these people have already made their money and are merely there to make more of it. What happens when AI doesn't work out for them and they still need employees? Either they get a pay raise anyway or they get let go and keep their mansions. If they erroneously let a bunch of employees go, then great, they can replace those roles with cheaper workers overseas working remotely. If AI itself can't take the blame for domestic workers losing their jobs, then they can point the finger at Anthropic and OpenAI. Modern workplace hierarchy depends highly on the diffusion of blame, and AI fits into that paradigm by introducing an entirely new dimension to that blame diffusion.
These businesses don't see themselves as corporeal parts of the world or as part of their physical, local community. They see themselves as intangible entities - bytes in the ether, spreadsheets, lacking physical substance or matter, immaterial ghosts owned by shareholders. Every other physical thing in the world, living or not, that is not a shareholder's wallet, is a resource to be used, exploited, mined, and discarded.
The Holy Grail is a business that exists without costs, employees, property, equipment, products, or even a physical location--just a virtual blob that increases a share price forever. That's ultimately the (in reality unachievable) goal end-state everyone is trying to at least approach.
Actually the ideal company is one that is a literal money machine - aka just creates money without liabilities - enabling shareholders to lay claim on the cash.
All of these arguments in this thread are essentially attacks on free market capitalism. I am not saying they are unfair, but I think you could have made the same arguments about management and investors doing the same thing in manufacturing in the US. They reduced domestic employment (not in total, but reduced the share and the growth) in manufacturing without regard to employees and communities.
If AI reduces white collar jobs, how can that be bad when automation reducing blue collar jobs is good? It's like suddenly engineers embrace Marxism.
> All of these arguments in this thread are essentially attacks on free market capitalism.
These argument aren't attacks because one of the basic tenets of free market capitalism is that everyone being greedy for himself is good for everybody. So people saying that corporations care only about profit and not about higher values only reaffirm that basic tenet.
You could try to convince them that they are wrong in their desire to make corporations care about more than money but you cannot blame them for saying something that fully agrees with free market orthodoxy.
> the same arguments hold about management and investors... reducing domestic employment in manufacturing without regard to employees and communities. > If AI reduces white collar jobs, how can that be bad when automation reducing blue collar jobs is good?
First, automaton wasn't the reason for the reduction of US manufacturing, outsourcing was. Wall Street used every leverage to push manufacturing and engineering abroad and that included outsourcing software work. At that time, both workers and engineers criticized the process, so you're factually wrong about that, twice.
Second, their criticism was right because... oh look, we have to bring it all back now, at a great expense paid for... by those same workers and engineers. Because after the big corporations got rich in China they want to come back here and own absolutely everything, including the government and continue the process of disfranchisement of the public by using AI not for higher output but as a cost and workforce reduction tool.
> It's like suddenly engineers embrace Marxism.
This topic has nothing to do with Marxism, which you appear to be using as a smear word, purely mechanically, without any understanding. Marxism has bigger problems but let's not get sidetracked here.
I think there's something else psychological going on. What you describe is a rational approach based off of bad values. But I think I'm also seeing something weirdly irrational.
It's like an (emotional) depression or something. Scarcity thinking, the inability to think expansively. People are so sure that everything around them is shrinking that they feel an instinct to hunker down, shrink, and cut as well. Like it doesn't occur to them that they don't have to feel that way. The execs I work with, none of them strike me as spreadsheet-driven greedy people. They seem more freaked out than that.
Inflation is heading up, jobs are becoming more scarce. The war in Iran is costing a lot. The federal reserve is now talking about when they are going to raise interest rates, instead of when they are going to lower them.
Plus the incredibly optimistic valuations they are talking up for the big AI company IPOs. If they didn't think we were nearing the top of the bubble, they wouldn't all be planning their IPOs at the same time. It's a race to see who can cash out first before it pops.
It feels similar to 2007-2008. People were talking about a bubble. But nobody wants to jump off the train too early and lose out on the big profits just before the end.
I've never seen a company punished for doing the right thing, caring for humans and providing a good work environment.
We’ll see how this goes over but I disagree. You don’t have to look hard in tech, especially a few years ago, to find groups of coddled “workers”, doing very little or at least doing what they want instead of what a business and customers want. This paradoxically ended up creating toxic work environments, and making it impossible to actually get work out of people. We’re seeing a correction now.
> You don’t have to look hard in tech, especially a few years ago, to find groups of coddled “workers”, doing very little
And whose fault is that? When employers create "fun" workplaces, value optics over excellence, disempower management, and maintain the status quo by the diffusion of blame, what sort of employees should they expect to have? I argue that it is not the fault of lazy workers but employers who encourage and tolerate lazy workers who are getting away blame-free. But the message is that it's always the fault of peons rather than the higher rungs of the hierarchy.
Good work environment is not coddling workers. It’s hard to discuss with people who believe taking care of your employees is catering to their caprices (or more likely, what YOU think they would like)
This tracks and IMHO some of the disconnect between technology innovation and productivity is that engineers are soaking up the excess by working less. They're not banging out more code/functionality because by and large that isn't rewarded
that was driven by seemingly endless amounts of cheap money being thrown at whatever and whoever, which is not at all what "caring for humans and providing a good work environment" means.
Watch Christopher Olah bloviate at the Vatican during the Magnifica Humanatis launch. It's truly nauseating. I've never seen such a ridiculous speech in my life. Between him and the CEO, I'm starting to understand the level of arrogance these people are capable of.
Strongly disagree. He sat in front of a room full of Archbishops and told them, straight faced, that the worlds about to have mass layoffs and starvation and that the Church should feel responsible for doing something about it. The guy's a complete sociopath.
He's not wrong. Mass layoffs and starvation are a problem society needs to solve collectively. The other side of the singularity will be great but we all bear a collective responsibility to get there in one piece.
> Mass layoffs and starvation are a problem society needs to solve collectively.
Very much disagree with this. This is capitalizing the profits and socializing the debts. They do not get to profit off of the suffering of others without repercussions, that is supposed to be what free markets prevent. This crony capitalist economy we have today with an explicit caste system in place, just pushes their debts onto you and I.
Destroying the economy and job market is the stuff of dystopian nightmares. If people do not have purchasing power they can't afford the product. The whole thing is destined to collapse.
But that is exactly his point. By "collective responsibility" he is saying that it's the public's responsibility to regulate and tax AI companies as needed, vs expecting them to self regulate. This has been Anthropic's stance the whole time.
reply