> So there will be endless work for humans to go over machine proofs and turn it into something we understand
I'm not convinced humans will be better at this than LLMs. Frontier models are quite good at handling even deliberately lazy/ambiguous eli5 prompts on complex topics IME.
>I'm not convinced humans will be better at this than LLMs.
they don't need to be. Any scientist is already surrounded by tools that are better than they are at anything. A scientist from 200 years ago would have the same existential shock if they see how a modern scientists works.
But that's to misunderstand what science is for, which is to expand human knowledge. And a tool by definition can only ever change how you do this, not that you do it because you understanding it is the point. When people say learning to program is now useless, what they really mean is it might not make them any money, or using the AI is faster as an instrument. But if you care about programming itself, about you understanding programming, nothing has changed, it never can. And the smartest AI in the world can always assist you, but only that.
> But that's to misunderstand what science is for, which is to expand human knowledge
I agree ideologically and aspirationally but I don't believe it's the economic reality of how and why most science is actually funded (and I think you agree based on your comment about not making any money).
> But if you care about programming itself
I'm...not so sure I do any more? At least, not any more than I care about using Java over Python, or C++ over Java, or C over C++, or ASM over C, or machine code over ASM, or an abacus over machine code.
> nothing has changed, it never can
We now have programs that can write any program 100x faster and better than even the best humans and solve the hardest math problems known to humankind that have stumped the greatest minds the planet could produce for decades. That, to me, is a change with profound implications for understanding the nature of programming.
No one who values democracy is complaining about Trump getting kicked off social media for inciting a mob to kill the vice president certifying the election in the Capitol building.
We also shouldn't really be removing hydropower at this point in time from a net zero/global warming perspective. It is orders of magnitude less environmentally harmful than generating electricity with fossil fuels, and one of two carbon-neutral sources of significant amounts of baseload power at scale along with fission nuclear.
a ~5HW hydropower is insignificant and can be replaced by a handful of wind turbines and solar panels. Not sure if it's worth sacrificing local biodiversity for that.
I did some quick napkin math on the net economic benefit, including carbon pollution and local economic benefit, of 3 scenarios:
1. Keep dam: -$2M to -$10M
2. Dam removal with 5 MW capacity replaced/absorbed by natural gas plants (no solar/wind/BESS): -$28.3M to -$249.8M, depending on whether you use the EPA's $190/tCO2 cost of damage or $1000/tCO2 DAC remediation.
3. Dam removal with 5 MW solar/wind/BESS replacement: -$3.5 to -$25M
In other words, removing the dam without equivalent solar/wind/BESS immediately in place is a net negative for society (and even then, keeping the dam might be neutral or still better).
You did not calculate the impact on the salmon population since dams were added to the Kennebec since 1837. Both the impact on the salmon themselves as well as all of the other animals and people who relied on that salmon fishery.
It's certainly a lot larger than the benefits of the dams in my calculation.
Ok, I found a study estimating a net willingness to pay per citizen for this type of ecological restoration at ~$72.31/per person/year. If we assume 100,000 people derive this level of benefit over a 30-year time horizon (same as I used for the previous numbers) that yields $21,693,000 which brings the removal WITH immediate solar/wind/BESS replacement scenario on par or marginally better than keeping the dam as is.
But even then, I believe that is being overly generous to the dam removal because electricity demand has been rising steeply recently and is projected to continue doing so. Until we are meeting 100% of current and future demand with new renewable energy, every existing carbon-neutral source that gets removed is creating a gap in future demand that will be partially filled by new fossil fuel generation and therefore incurring many of the same associated negative societal costs.
I would flip the question on its head and ask why the response to uncertainty is to default to an assumption of zero. Environmental benefits are notoriously hard to calculate and it's unfair to weaponize that difficulty as if it confirms the default assumption should be zero. I also don't think the calculation needs to hinge on one-for-one timed tradeoffs in the form of "this windmill exists to replace power from that dam" or assume that backfill is 100% fossil fueled, given that solar is now arguably the most rapidly deployed form of new generation.
It's not that the cost never matters but I think if the cost is really just $25MM for a transformatively beneficial environmental impact that has equivalent no alternative, in exchange for scaling back a fungible resource, is a no-brainer.
Most of those small and medium hydropower plants generate near insignificant amounts of electricity, even if you damn all the non major rivers you can (and destroy all the surrounding ecosystems) it would be hardly noticeable.
My napkin math disagrees. Small/medium hydro (<100 MW) appears to account for ~1.7% of annual US grid energy output. If we replace that with fossil fuel generation and account for the cost of dam decommissioning, account for estimated benefit of ecosystem restoration in terms of quantified dollars willing to pay by benefiting citizen, and net annual CO2 cost I come up with an annual cost to society of removal of small/medium hydro dams between $1.7B/year on the low and $41.7B/year on the high end. Happy to show detailed assumptions and calculations if desired. But tl;dr fossil fuel generation is very, very bad and replacing even comparatively small amounts of carbon-neutral generation with it results in a significant societal net negative.
> More than 7,800 research grants terminated or frozen. Some 25,000 scientists and personnel gone from agencies that oversee research. Proposed budget cuts of 35% — amounting to US$32 billion.
> These are just a few of the ways in which Donald Trump has downsized and disrupted US science since returning to the White House last January. As his administration seeks to reshape US research and development, it has substantially scaled back and restricted what science the country pursues and the workforce that runs the federal scientific enterprise.
> If all states had imposed restrictions similar to those used in the 10 most restrictive states, excess deaths would have been an estimated 10% to 21% lower than the 1.18 million that actually occurred during the 2-year analysis period; conversely, the estimates suggest counterfactual increases of 13% to 17% if all states had restrictions similar to those in the 10 least-restrictive states. The estimated strong vs weak state restriction difference was 271 000 to 447 000 deaths, with behavior changes associated with 49% to 79% of the overall disparity.
The paper you linked also doesn't take into account economic impacts, which aren't negligible, and lead to deaths of despair, drug abuse, unaffordable housing, etc.. They also admit the school closures were ineffective and unnecessary and we've seen the mental health and educational impacts on kids and now young adults that had those years taken from them.
Keep in mind also that the favorite phrase of the COVID era was "abundance of caution." That's itself a tacit admission that they knew their recommendations went too far and they weren't putting enough weight on the secondary effects.
> Our meta-analysis finds that lockdowns in the spring of 2020 had a relatively small effect on COVID-19 mortality and is consistent with the view that voluntary changes in behavior, such as social distancing, played an important role in mitigating the pandemic
I find it strange you bemoan the public health establishment's credibility when your own source shows that people following its guidelines played a greater role in saving hundreds of thousands of lives than legal mandates.
Social distancing in an infectious disease pandemic is obvious and people would have done that regardless. My point is that the specific advice of the public health establishment was nonsensical (don't wear masks, actually do wear masks and cloth is fine, don't go to parks or beaches but political protests are fine if we're sympathetic to the cause, wear a mask when you walk to your table but not at your table, natural immunity isn't as good as a vaccine, and on and on). It was ridiculous and a big reason public trust in institutions is at an all time low.
The fundamentals of writing correct code have become dramatically cheaper by orders of magnitude. I have been involved in hiring dozens of developers. Fable/Opus 5.5 would be a superior drop-in replacement for perhaps 75% of those hires at roughly 1/100 the cost.
Perhaps for writing boilerplate, or generic websites from scratch, or early stage prototypes.
I work in embedded systems where it's harder to offload the burden of testing and maintenance. Systems knowledge is important: experienced developers can be costly to replace.
LLMs make a lot of things easier, but costs have certainly not fallen by 2 orders of magnitude.
> The people that sat around handwringing about tech debt and trying to read every line of LLM code will really struggle to find a job. The profession fundamentally changed, and these people did not catch up.
I'm inclined to agree - the latest models seem to have crossed a threshold where trying to review by hand becomes akin to a manager insisting on reviewing PRs.
God forbid we figure out the best way for the smartest and most talented people of all backgrounds to contribute to science. Why take diamonds from the rough when you can just leave them there instead?
I'm not convinced humans will be better at this than LLMs. Frontier models are quite good at handling even deliberately lazy/ambiguous eli5 prompts on complex topics IME.
reply