It needs a more coherent cogent discussion, leading to dare I say it sane _policy_ ... which the frontier model companies and hyper-scalars are not incentivized to finance, and governments too slow and weakly funded to finance.
Other topics we seem to have been unable to have deep and informed discussion on, leading to sane policy :
- inequality / taxation
- demographics / aging population
- climate change mitigations
- affect of social media esp on young people
- safety of self-driving cars
- economic impacts of LLMs
- educational "" ""
- security "" ""
- surveillance of in-home devices
- data collection and sharing
- right to repair
Failing to react rationally to a rapidly emerging global pandemic a few years ago should have been a wake up call but wasnt.
If the technorati are woefully unprepared for the age of AI, which is being rapidly thrust upon us by the big money, then the wider public dont stand a chance of being informed / prepared.
Enjoyed this. More importantly its an emerging problem we need to urgently solve.
Apart from my own desire to read the internet to find out what actual people think .. Im worried about a kind of bit-rot, where any decade now, gen-pop reader has no idea what content came from humans and what came from AI.
Im assuming the people who _train_ AI [ LLMs ] actually need to separate the two, and avoid the feedback loop of training the next LLM on the output of the previous LLM.
Presumably, this would lead to a kind of reversion to the mean, akin to making photocopies of photocopies back in the day, resulting in copy degradation.
We already have a kind of bit-rot from moving away from physical media - many documentaries on TV, and audio / video / movies / games are being lost as they are not moved to permanent archive storage.
Old out of print books should be scanned and made public and permanently available online, as they are part of our cultural heritage [ not least for the purposes of training current and future AI as the defacto archives / oracles ]
> Im assuming the people who _train_ AI [ LLMs ] actually need to separate the two, and avoid the feedback loop of training the next LLM on the output of the previous LLM.
Oh they don’t have this data degradation problem at all. They have large data curation teams that scrutinize the “data mix”, ensures the model is always performing better.
I think the current gold-rush of nearly all money into GPU Datacenter and frontier LLMs is essentially starving the economy of innovation.
Academics and founders who might work on developing practical products using NN / ML / RL techniques to solve a realworld problem in engineering/logistics/medicine are not getting investment money. VCs and most people are blind to the fact there is AI outside of LLMs, despite the fact that we have seen AlphaGo and AlphaFold as evidence of non-LLM AI progress in hard domains.
This is perhaps a sub-problem of a larger issue - hyper-inequality means that capital is not allocated to talent [ capital is localized, talent is more widely spread throughout the population ].
We are not getting money to things that will grow our future such as :
- small innovative startups
- university science research
- people who are young enough to have kids, being able to afford them
- new garage bands / authors / musicians / photographers
- public works / infrastructure / libraries
- local retail : bookshop, artisanal bakery, cafe
My thesis is that during the 70s-90s we had higher tax, lower inequality, lower median income to median house price ratio, higher levels of innovation and more original art, literature and music being made.
AI could be a golden age of human flourishing - but thats not where we are heading, what we are seeing is a territory rush by the megacorps.
The fact that RAM and GPU prices have risen so fast, is evidence of supply and demand effect where inequality steals resources from the commons [ middle of the economy ].
Can a talented garage inventor / math or arts student afford a Ryzen AI dev platform, let alone a DGX spark on which to create the next important technology innovation ?
I see results like the OP as speeding up progress. Tests like ptau make it faster and cheaper to screen for participants in clinical trials. This allows more resources to be spent on pre-screening, to identify candidates that might be healthy at time of trial start, but who are likely to develop cognitive impairment during the trial, which can be helpful for better measuring efficacy etc.
There was a good, accessible guidelines paper in the Lancet that described helpful interventions at all stages of life to decrease the odds of developing cognitive impairment later in life.[0]
In terms of late-stage treatments to halt or reverse progression, that still seems very far away. The causes are so diverse... My father died of vascular dementia not too long ago, so I'm familiar with the caregiver side in addition to the business.
While AlphaFold is great, to my knowledge, these proteins have several non-significant IDRs (intrinsically disordered regions), which have been and continue to be hard to model with AlphaFold. I work on computational modeling, so if I had to guess, I'm very unsure if people are using it for whole protein modeling, rather than looking at more specific domains where it probably does much better.
Recent studies estimate climate warming is happening faster than it was, ballpark +0.3C per decade.
As we are at or nearing +1.5C now, with global emissions at an all time high [ perhaps peaking soon ] theres a good chance we are headed toward +2.6C by ~2060.
Keep in mind the CO2 stays up there for a long time, centuries, so when we get to net-zero its max-CO2 and therefore max-heat.
does it stop? the more co2 the more heat is retained. so the relation between co2 and temperature is an integral over time. were we suddenly to stop burning things how far does it go up before infrared emissions equilibrate with the larger thermal flux? that and the lower albedo and other knockoff effects may continue to make things worse even if we were to stop.
Perhaps there are other mechanisms at play here? What about the AMOC collapse. IF too much heat disrupts some weather patterns which are heating the northenr emisphere, that would mean an increase in the ice cover again and therefore more heat reflection again.
yeah, Id actually like to see a reasonable prediction of the trajectory, factoring in recent few years data [ faster warming, current emissions, cleaner air / reduced aerosols and issues like those you mention such as equilibration and ocean heat buffering ]
The graphs I see all have ridiculous ranges, either +1.5C hopium or +6C hellscape - whereas +2.6C within 20 to 30 years seems a most likely [and unpleasant] scenario.
Let us bow and pray to the LORD of technological progress, for HE, in all HIS infinite might and glory will save us.
Sometimes shit just doesn't happen. To prime your brain for the empathy required, imagine the sensation that you feel when you jump off a bridge or building and realize that you've made a huge mistake, and you'd actually like to live a bit longer, but now there's no turning back and it is over for you.
Now imagine that same sensation, but apply it to the lives of all the less privileged people that will die due to first-order (wet bulb, drought) and second-order (war, famine) effects of climate change. But maybe we're not screwed. Maybe you actually survive the fall because a benevolent sea lion decides to keep you afloat after you impact the water and break all of your bones too severely to tread water.
Still fucked up to just jump anyway, especially if it's not your life that you're gambling with (thanks Baby Boomers, esp. oil baron Baby Boomers).
We will not. Carbon capture is a joke. It does not work anywhere near the scale it's needed. We would need many times more capture facilities than we currently have power plants. And to run them we would need many times more power plants, not to mention all the construction emissions etc.
Carbon capture is completely infeasible. It's already too late anyway, our civilization is pretty much doomed barring some miracle.
It's not about tech. It's about the fact that the atmosphere has like 0,00something% CO2, so you need to process like a thousand tons of air to theoretically extract one ton of CO2.
You basically need to suck the entire atmosphere through a factory, it's not about tech it's just a big fan. And then of course you need something to process all that air as fast as it comes in.
We've already been doing it for 50 years, and the pinnacle of achievement in the field is a facility being built in Texas that is planning to sequester 500,000 tons of CO2 annually.
The world produces over 38 billion tons of CO2 a year. Assuming they reach their target performance, we would need 38bn/500k=76,000 of these facilities just to offset our current emissions.
The number of "power generating facilities" in the entire world, for reference, is about 30,000. So we would need more than twice as many capture plants as we currently have power plants. And of course they all need power, ways to actually get rid of the carbon they're collecting, chemicals etc.
So sure, if you want to live in a dream world where we magically just invent a 1000x better way to do it it's totally feasible. And while we're at it I'm sure we'll just make a 1000x more efficient combustion engine and a 1000x more effective dynamo and 1000x more effective ways to grow food and 1000x more effective solar panels, desalination etc. The next 50 years will totally revolutionize all these technologies we've already close to perfected, surely.
In case it wasn't clear that's sarcasm. Nobody is expecting to significantly improve any of these things because we are already closing in on hard physical limits that can't just be technologied away. A 1.2x improvement in any of them would be staggering, and even that is unrealistic for most if not all of the random technologies i mentioned.
Furthermore, 50 years is way too late. In 50 years we will be at like +3-4C or beyond. If things haven't already completely gone to shit at that point I can guarantee you there will be chaos and much more to come. They will have more pressing concerns than building 70,000 carbon capture facilities.
I'm not having kids, I hope my corner of the world stays relatively stable until I'm gone and I pity the people who have to live beyond that.
I'm just explaining to you why this technology will almost certainly not solve the problem, it likely won't even make a dent in the problem.
I believe this is an important point to make because it looks to me like the primary effect (and maybe also intended effect?) of carbon capture efforts is to encourage complacency. "Don't worry about these emissions, we'll just slurp the CO2 back up later and fix everything".
It is orders of magnitude harder to remove the CO2 that it is to emit it. Relying on "future technology" to save us is a prayer not a plan.
What if there is no solution? Do you understand that we are talking about the very imminent end of the world? Literally a complete collapse of human civilization, likely starting within this century. Maybe within our lifetimes.
"Let's hope someone finds a solution" is not enough. We have the solution, it is to stop emitting billions of tons of greenhouse gases every year. We just aren't doing it. So yeah, I find it just a little frustrating when people deny the problem, deny the severity of it, or claim that magic future technology will fix it. This is why we are in this position in the first place, because people do not take it seriously.
Can you point to any evidence for the idea that climate change will be the end of human civilization? I’m fully onboard with the scientific consensus around climate change, but I haven’t seen anything credible suggesting it’ll be anything close to civilization-ending. Even RCP 8.5, which was dropped for being implausible, doesn’t predict effects anywhere close to ending a global civilization of ten billion people.
> the climate change mitigation goals which have been pledged are far too low (ambition gap), and the measures which are currently being implemented and planned are not sufficient to achieve even these inadequate goals (realisation gap). This has far-reaching consequences: There is a strong probability that the ambition gap will lead to a temperature rise of 2 to 5 degrees Centigrade compared to pre-industrial temperatures by 2100, the realisation gap to a further rise of several degrees Centigrade.[1,2,VI] There is a danger that the mean temperature will already have risen by 3 degrees Centigrade by 2050
there is no certainty that adaptation to a 4°C world is possible
That's very different from your certainty that our civilization will end. Even just reading through their list of impacts at +4C, it's horrific, but pretty clearly not civilization-ending.
IMO, this kind of hyperbole does a lot of damage to your cause. The reality is bad enough, you don't need to make things up. People eventually stop believing the boy who cries wolf.
What do you think adapt means? If we do not adapt we die. Those are the options. If we cannot adapt to the new climate on Earth we can not live on Earth. And we sure can't live anywhere else.
I'm not "crying wolf". You, like most people just don't want to hear it. That's why most (especially the "officially accepted") climate models are unrealistically optimistic even with their worst case scenarios. Because people who tell the truth are considered alarmist and people who sugar coat it are ignored.
Just to be clear, and because some people inexplicably seem to consider this worth even talking about, I'm not saying there won't be any humans anywhere. That might happen but there might also be some habitable places left at least for a while, maybe for a long time. Nobody knows exactly what's going to happen. If 80% of the global population were to perish I would consider this "prophecy" fulfilled. It's not important to me whether some people survive somewhere because the point is I don't want to live through (or, overwhelmingly more likely, die during) the process leading up to that and I don't want to put a new person into the world who will have to live through that. That's where my concern ends, I don't care what happens after the world as we know it is gone.
Humans do inhumane things to each other during hard times, and I guarantee you, regardless of how you define words like adaptation and civilization, there's extremely hard times coming.
Climate change won’t even kill 10% of humanity. I do not understand why people are so pessimistic about our ability to adapt. That’s how we created this problem in the first place! Of course we’re not going to just lay down and die. 80% is completely ridiculous and there’s not a shred of evidence for it, is there?
I understand that you don’t think you’re being alarmist, of course. But why would I think some random internet commenter is more credible with their predictions that are much worse than the IPCC and other scientific consensus?
it's not just reasonable to question, it's all but obvious and anyone who comes up with a retort like "what about millions/billions of years before civilization" is being, at best, insincere
... theres a lot of anecdotal evidence executives aren't good thinkers.
Why not replace all mid-execs with LLMs trained on every MBA textbook and linked-in blog post ? why stop at grunt coders, graphic designers and phone marketeers ?
Can we reduce the role of CEO to the following prompt : "make money, by any legal means"
I personally look at react-foo and see something thats the opposite of how I want to work as a software craftsman [ small code, lispy / function flavored style, opinionated, few dependencies ]
For similar aesthetic reasons, have only dabbled with code-gen LLMs. Im happy to vibe code css and html but not, well actual code. I'm aware of my own double standard, Im fine with median-reverting banal css and html.
I do worry about some aspects of LLMs :
- google giving AI code snippets so easily, prevents traffic to sites like stack-overflow, where discussion happens
- vibe coding means less people hit the actual hard parts of coding, which result in learning to think as a developer. Thence where the next generation of developers who understand code ?
- code will become a magical, inscrutable black box where no-one understands how it works. DNNs and LLMs have this feature already. At least with LLM discovered math, the proof in lean exists, which is independent at how it was arrived at / searched for.
- if I write a novel, and the DNN AI detects it as AI slop with 68% probability, then what recourse do I have to prove its human written ?
- circular training : subsequent rounds of LLMs trained not on human input, but on LLM slop of the previous generation [ the %ge of which increases ]
- you will own nothing : fortunately it looks like self-hosted open weight LLMS will remain a thing, but the massive spend by hyperscaler companies on Datacenters/GPUs/RAM has driven down supply and driven up prices of CPU, GPU, RAM for enthusiasts / gamers / home AI nerds / garage startups .. which might stifle innovation.
This is possibly more exciting and of more future value than solving Navier Stokes.
It also shows us a way, when it seems we are lost.
reply