What good is something that only looks good from afar? Most new tech is exactly that, for posturing to gain shareholder value - very few innovation in actually useful things that make peoples lives better. It's all self-serving, to consume more and get people hooked on digital experiences.
I’m not arguing that social media is a net benefit to society, but you’re acting like it’s all downsides. It’s been an extreme economic enabler for a lot of people. There are many people earning money independently who would have really struggled to put themselves in a position to do that 20 years ago. For independent businesses or artists it’s an invaluable promotional platform.
You know I really don't know about that. Generally speaking people had more dispensable income after WW2 in the west than right now. People had a job lined up if they wanted to work, work was abundant. Back then any kind of job would pay well and most households had one earner. Being able to earn money through the internet today doesn't really replicate what these people had back then.
If you wanted to start a band you had multiple ways to promote them: I would argue some of the best known bands came from the pre-internet era (Beatles, Rolling Stones etc). In fact social media made it so you need a prohibitively high marketing budget to cut through the noise.
Define "good from afar". We romanticise the notion of the wild west, while ignoring the poverty death and disease. Would you rather live in the wild west or today, where the issue is AI which at this point is more a #SVproblem than even a #firstworldproblem.
That's a very interesting point. I'd think the "happyness/contentness" levels are not necessarily tied to life expectancy or poverty.
I would argue despite some global tendencies being way up (like people out of poverty) we are at a local minima right now for happiness for the middle class. It is entirely possible middle-class people had it better in the late 50s/60s despite multiple statistics being much better today (e.g. crime).
In a way it is also entirely possible, that a medieval peasant was happier than an overstimulated, modern human being, despite having a "worse" life.
Tbf I think my view is it's about improvement over your own life. We expect our children will do better than us and that our lot will improve etc.
That sets expectations. We are at a high base, but things aren't getting better thus we can't look back at our own life and see that we have been successful.
I suppose the question is, where are we in the life cycle? In the industrial revolution life was pretty crumby for a lot of the working class. Do we have to wait for society and laws to catch up so we get to that 1950s/60s heyday of life for the working class. Or is this a plateau and the disruption will makes things worse before they get better? Or is the west going to stagnate while china or whoever takes over?
Yes, very true, and I can already thank you for the thought-provoking discussion.
I think we are the first generation who don't necessarily expect that children will do better than us, in fact, in many ways we know that they won't: climate change etc.
The current gen alpha is also the first generation since measurements began, who are doing worse at school, measurably, than the previous generation. All research points to the inclusion of digital learning devices in school as the cause of that.
That's my point. It's not about the technology itself, because that's never cut and dry. It's about having a process to evaluate and criticize a technology before fully embracing it. Socrates was in a position with loud voice. Critics of technology exist today too, but they aren't loud enough. Instead, there's a far stronger force coming from private companies and their investors to push whatever they want onto us.
But I want to admit that I was indeed shoehorning my rants about AI (not the technology itself, but how it is being adopted) into this topic. Let me stop steering the discussion further. I do think the fact that we could recover Herculaneum papyrus like this is amazing!
Signing this sounds like a good way to get fired. Executive in corporations gets to make the decisions. Employment is at will, if you don’t like it you get to leave otherwise you’re not fulfilling your contract
NAL but it may be protected activity to improve working conditionals. I would guess Meta leadership doesn’t actually care very much if someone signed it. And typically the people making firing decisions are not necessarily the ones that want the AI training data anyway.
also NAL and i think you’re right, the petition even says as much. but at-will employment means they can fire you for basically any other reason, and they probably have at least 10 to choose from at any given time.
No, look a Composoer 2, it stands out starkly on its own in the pareto frontier on low cast and fast models.
Composer 2.5 was a huge leap with minimal compute from xAI.
They can compete with OpenAI and anthropic with xAI scale compute. They have a top notch model team and incredible training data and huge enterprise costumer contracts.
my employer (one of those huge contracts) dropped cursor in favor of claude and i don’t think this is true at all
while we had it i used cursor for probably eight months as my main ide (i did really like the interface for embedding code in prompts!) but had no problems switching to claude code. i asked around, and i truly don’t know a single coworker who misses cursor even a little bit.
I was fully in on Cursor for a good chunk of last year, using Composer + Gemini Pro (via Copilot / GH integration). I really enjoyed Cursor's tab completion capabilities, but when Sonnet and Opus started getting particularly good for me (think for me it was around 4.5), I swapped over to Zed + claude code in the integrated terminal. I've found that after a bit, I haven't ended up missing the tab completion. I've been perfectly fine with just LSP + claude always open. I don't miss Cursor. All my colleagues are on claude code with half of us also using Zed.
For someone who is new to agentic code or is generally somewhat junior, Cursor is very easy to get started with and is generally fairly frustration-free.
I use a cheap $20 subscription mostly for occasional use of Opus and Composer.
SpaceX made a smart move here. Someone else should have really seen the opportunity and bought them.
How the hell is an IDE a "pareto frontier"? Even if, say composer 2.5 is a huge leap forward, that doesn't mean IntelliJ or Vim or Emacs or Codex got worse.
Nope. Vim and Neovim users can use the cusor-agent cli for agentic stuff and there prefered editing tool for editing. All the major providers have a cli specific version these days. Probably because folks actually didn't want to actually use the cursor Gui and once Claude code came along those folks jumped ship and went full cli again
This doesn't match the empirical observations of a lot of people I'd trust more than you, and putting it below Haiku immediately makes it extra sus (Sonnet would have been the credibility preserving comparison).
Hmmm. Not in my experience. I don’t think it can be compared to Haiku, maybe sonnet levels? It’s obviously not Opus and never was intended for that use case. I use it quite a bit and it works well and is extremely fast for the tasks it was built for
It’s a bit misleading to say nothing special, as they are doing more than just increasing parameter count. Progress has been steady in all the sub components of training from data filtering and weighting to sparse attention, optimizers to up and down the stack various efficiency in training computing.
They’re using more compute, a bigger model and tons of training quality improvements to get more out of an equivalent model.
This has been my thought for a long time. I think all that matters from attention is that there is crosswise comparison going on.
You need some amount of parallel compute and some amount of global comparison.
And the rest is basically a ways to parameters and scale.
(This is in theory, in practice you can get a lot of small % stability and efficiency improvements that really compound in algorithmic details of model architecture)
Confidently yes. OpenAI for sure has been training larger models internally and distilling.
Pre-training scaling laws all support larger models being more cost effeceint to train then smaller models. And distillation is comparably cheap. So you can get the most juice by training the biggest model you can and distilling it.
There is endless returns to frontier intelligence, just because most people can't make use of it doesn't mean someone can't make a ton of money off of it.
Most software engineers will just need cheap tokens.
But things like physics and drug discovery have no foreseeable upper bound.
Or governance of large organizations... There are a huge number of factors to consider, counterfactuals, studies, lots of non-obvious second and third order effects, etc. We're barely able to get basic governance without creating huge problems (low density zoning rubber stamped across the nation creating a housing crisis, for example), so the bar isn't high.
We pay CEOs an enormous amount because a small improvement in performance of an org because of them can make a massive difference in organizational value.
There is endless returns to frontier intelligence, just because most people can't make use of it doesn't mean someone can't make a ton of money off of it.
Most software engineers will just need cheap tokens.
But things like physics and drug discovery have no forseeable upper bound.
Within software engineering, security, reliability, and scale also seem boundless.
Software that never breaks (including because it never runs into scaling problems) and never leaks your data is preferable to software that breaks and leaks your data sometimes, but it has been too costly to be practical.
Current models are still very far from the reasoning muscle required to build things that never break, scale to billions of users with no issues, and cannot be exploited.
> Software that never breaks (including because it never runs into scaling problems) and never leaks your data is preferable to software that breaks and leaks your data sometimes, but it has been too costly to be practical.
It's almost impossible to prove non-trivial software is invulnerable.
It's very easy to prove that it sort of works.
For one, you have hardware vulnerabilities - period. If you're running on any operating system, you have OS vulnerabilities. If you're not running on bare metal, you may have who knows what kind of vulnerabilities. If you're running literally any other piece of software on the same machine, depending on the hardware and OS, you could have vulnerabilities...
Nothing ever happens, in 20 years we will still be painfully dying from the same shit as now. Maybe there is like 5 new drugs for some exact specific type of cancer out of like what, thousands?
They would be impressed with our technology even if it has downsides. Wisdom is knowing humans and technology and imperfect tools.
reply