Hacker Newsnew | past | comments | ask | show | jobs | submit | sharemywin's commentslogin

People assume training on past data means no novelty, but novelty comes from recombination. No one has written your exact function, with your exact inputs, constraints, and edge cases, yet an LLM can generate a working version from a prompt. That’s new output. The real limitation isn’t novelty, it’s grounding.

Synthesis is different than genesis.

Just because there are lots of tasks which can be accomplished without the need for anything novel doesn't mean LLMs can match a human. It just means humans spend a lot of time doing some really boring stuff.


That’s only part of it. The true intelligence is knowing what to recombine, and this is where AI falls short.

Wonder how much WFH could help. Seems like during covid demand went way down.

It helped on reducing traffic, spending more time with family, favouring local shops... Why we went back still is a mystery for me. Even if it was "working from a coworking space" or anything that was not the downtown open-space.

I remember reading articles about local politicians insisting companies and governments do return to office for no other reason than to have the workers spend money downtown for lunch. Not even implying it, directly stating it.

...and increasing valuation of office real state.

Because downtown city centers were missing out on office worker revenue and started giving incentives to companies who brought people back into the office. I 100% believe the reason we went back into the office at all was because of this despite all the talk of 'in-person collaboration.'

Exactly, put AI in the office instead.

WFH and the almost 100% shutdown off airline travel at the beginning of the pandemic resulted in nearly 0 change in CO2 emission and levels in the atmosphere.

https://www.climate.gov/news-features/understanding-climate/...


Of course we shouldn't expect a couple of years of shutdown to significantly reverse 200 years of man-made atmospheric CO2 accumulation, but surely it would help stop the problem to get worse if the widespread WFH effort were sustained after the pandemic.

It's a bit off topic but that didn't sound right to me.

According to the following it was a reduction but yes near zero in the context of total emissions. A few hundred million tonnes reduction ain't nothin none the less.

"plummeted from more than 1 000 Mt CO2 in 2019 to less than 600 Mt CO2 in 2020, in the context of the pandemic.

In 2023, aviation accounted for 2.5% of global energy-related CO2 emissions, "

https://www.iea.org/energy-system/transport/aviation


Heavens no! How could we ever afford to be more productive overall? Just think of the effect on corporate real estate prices.

That's true. I wonder if the government might force companies to allow WFH if there are petrol shortages.

They did it during covid so I wouldn't rule it out.


I don't have data but I was always under the impression that consumer use of fossil fuels (ie gas) was a drop in the bucket compared to enterprise use of fossil fuels (shipping trucks/boats/planes, private jets, etc).

The whole "reduce your carbon footprint PSA" was just a ruse.


> The whole "reduce your carbon footprint PSA" was just a ruse.

It is especially true in the context of this war when the US is attacking oil infrastructure which is causing catastrophic environmental damage: https://www.nytimes.com/2026/03/10/world/middleeast/iran-oil...

Similarly related to the paper straw meme that has been circulating the last while.


Ugh don't get me started on paper straws! At least EVs are a cool way to reduce reliance on fossil fuels, however ineffective it may be at scale. But nobody likes paper straws.


Yikes. I never really thought about the environmental impacts of war mainly because no one seems to talk about it.

Now that I know about it, it just infuriates me that anyone could even criticize any individual person's usage of fossil fuels.


But the companies have a large carbon foodprint to deliver a product or service for the consumer.

When I buy a large SUV / Truck, and never drive it, is that not counted negative towards my carbon foodprint?


> But the companies have a large carbon foodprint to deliver a product or service for the consumer.

I agree that that's _why_ they have a large carbon footprint, no company is just burning fossil fuels for fun. But it doesn't change a) the fact that they do have a large carbon footprint, and b) entire cities could ban gas cars and everyone could take public transit and it still wouldn't make a dent in the global carbon footprint.

As I think you're alluding to over-consumerism as a cause of companies having a large carbon footprint, that's part of it. But unless everyone just stops consuming, it's not gonna change anything. If it were legislated that big companies needed to reduce their carbon footprint by X% by Y date I think that would be the most effective, short term at least.

> When I buy a large SUV / Truck, and never drive it, is that not counted negative towards my carbon foodprint?

I don't know why it'd be negative. Zero or neutral, at best, but not negative. Negative would entail you're somehow removing CO2 from the environment.


Oh my bad, I used negative wrong.

I mean if I buy a car (and never drive it) and have a phone, ipad and computer that I never use. Is it not ME who has a huge carbon footprint?


But think of the commercial real estate market?!?! What about the chopped salad slop bowl market?! Dry cleaning?! This would shatter the fabric of our precious society. We need butts in seats. We need to foster open communication and cross functional pollination.

so by that logic is transaction data allowed without a warrant?


I think the big win with AI is being able to work around jargon. Don't know what that word means ask AI. what the history on it no problem. don't understand a concepts explain this at a high school reading level.


what if cool new tech is just slowing down and AI is masking it.


Not a "what if". Can you name 3 new cool technologies that have come out in the last 5 years?


1. Copilot for Microsoft PowerPoint

2. Copilot for Windows Notepad

3. Copilot for Windows 11 Start Menu


Nah man, I’m still waiting for Copilot for vim.


Yeah. Where are all the great new Mac native apps putting electron to shame, avalanche of new JS frameworks, and affordable SaaS to automate more of life? AI can write decent code, why am I not benefiting from that a consumer?


Well, if you're a consumer of code, then technically you benefit. Otherwise, you probably won't notice it as much.


It's almost like a lot of our technologies were pretty mature already and an AI trained on 'what has been' has little to offer with respect to 'what could be'.


oof that's profound. Really nice closing thought for 2025.


LLMs, Apple Silicon, self-driving cars just off the top of my head without really thinking about it.


GPT-2 was 6 years ago, the first Apple silicon (though not branded as such at the time) was 15 years ago, and the first public riders in autonomous vehicles happened around 10 years ago. Also, 2/3 of those are "AI".


> the first Apple silicon (though not branded as such at the time) was 15 years ago

Nobody, not even Apple was using the term "Apple Silicon" in 2010.

The first M series Macs shipped November 2020.


1 year is being pedantic. Apple Silicon is clearly referring to the M series chips which have disrupted and transformed the desktop/laptop market. Self driving also refers to the recent boom and ubiquity of self driving vehicles.


M series is an interation of A series, "disrupting markets" since 2010. LLMs are an iteration of SmarterChild. "Self driving vehicles" is an iteration of the self-parking and lane assist vehicles of the last decade.

I'm bored.


Damn, what an LLM roast. Smarterchild couldn't even recall past 3 messages


I would be bored too if I was disingenuous. Everything is an iteration of ENIAC right? Things haven't changed at all since then right?


All of those things are more than 5 years old.


I could not get in a Waymo and travel across San Francisco five years ago, are you serious?


Neura Link, Quantum computers are making interesting milestones with Microsoft releasing a processor chip for Quantum computing. Green steel is another interesting one, though not as 'sexy' as the previous two.


didn't believe the quantum stuff, so I googled it. I'm shocked how far its come. Even China has some kind of photonic quantum chips now.


Wait so quantum is going to actually deliver something useful within the next 10-20 years??


Incredibly cheaper batteries and solar panels. Much better induction stoves.


I was thinking computer-related, but those are good, and better battery technology helps with computing.


Uhhh, LLMs? The shit computers can do now is absurd compared to 2020. If you showed engineers from 2020 Claude, Cursor, and Stable Diffusion and didn't tell them how they worked their minds would be fucking exploding.


So LLMs exist therefore nothing else is worth the time? That’s sort of the gist of HN these past few years


Moreover: people’ve been crowing about LLM-enabled productivity for longer than it took a tiny team to conceive and build goddamn Doom. In a cave! With a box of scraps!

Isn’t the sales pitch that they greatly expand accessibility and reduce cost of a variety of valuable work? Ok, so where’s the output? Where’s the fucking beef? Shit’s looking all-bun at the moment, unless you’re into running scams, astroturfing, spammy blogs, or want to make ELIZA your waifu.


No I was just skeptical of the GPs assertion that tech hasn't produced anything "cool" in the last 5 years when it has been a nonstop barrage of insane shit that people are achieving with LLMs.

Like the ability for computers to generate images/videos/songs so reliably that we are debating if it is going to ruin human artists... whether you think that is terrible or good it would be dumb to say "nothing is happening in tech".


This was posted earlier today:

https://www.danshapiro.com/blog/2025/12/i-made-the-xkcd-impo...

The xkcd comic is from 11 years ago (September 2014).


Surely you have realized by now that a large portion of the HN userbase is here for get rich quick schemes.


ahh brings me back to the blockchain days, and the many excuses people tried to use them instead of a SQL database for whatever reason


It’s really incredible how quickly people take things for granted.


LLMs are one, granted. GP asked for three, though.


GGPs question doesn't make sense though. What does it mean for a technology to "come out".

Also what does three prove? Is three supposed to be a benchmark of some kind?

I would wager every year there are dozens, probably hundreds, of novel technologies being successfully commercialized. The rate is exponentially increasing.

New procedural generation methods for designing parking garages.

New manufacturing approaches for fuselage assembly of aircraft.

New cold-rolled steel shaping and folding methods.

New solid state battery assembly methods.

New drug discovery and testing methods.

New mineral refinement processes.

New logistics routing software.

New heat pump designs.

New robotics actuators.

See what I mean?


Great list, and most of those don't involve big tech. I think what your list illustrates is that progress is being made, but it requires deep domain expertise.


Technology advances like a fractal stain, ever increasing the diversity of jobs to be done to decrease entropy locally while increasing it globally.

I would wager we are very far from peak complexity, and as long as complexity keeps increasing there will always be opportunities to do meaningful innovative work.


1. We may be at the peak complexity that our population will support. As the population stops growing, and then starts declining, we may not have the number of people to maintain this level of specialization.

2. We may be at the peak complexity that our sources of energy will support. (Though the transition to renewables may help with that.)

3. We may be at the peak complexity that humans can stand without too many of them becoming dehumanized by their work. I could see evidence for this one already appearing in society, though I'm not certain that this is the cause.


1. Human potential may be orders of magnitude greater than what people are capable of today. Population projections may be wrong.

2. Kardachev? You think we are at peak energy production? Fusion? Do you see energy usage slowing down, or speeding up, or staying constant?

3. Is the evidence you're seeing appear in society just evidence you're seeing appear in media? If media is an industry that competes for attention, and the best way to get and keep attention is not telling truth but novel threats + viewpoint validation, could it be that the evidence isn't actually evidence but misinformation? What exactly makes people feel dehumanized? Do you think people felt more or less dehumanized during the great depression and WW2? Do you think the world is more or less complex now than then?

From the points you're making you seem young (maybe early-mid 20s) and I wonder if you feel this way because you're early in your career and haven't experienced what makes work meaningful. In my early career I worked jobs like front-line retail and maintenance. Those jobs were not complex, and they felt dehumanizing. I was not appreciated. The more complex my work has become, the more creative I get to be, the more I'm appreciated for doing it, and the more human I feel. I can't speak for "society" but this has been a strong trend for me. Maybe it's because I work directly for customers and I know the work I do has an impact. Maybe people who are caught up in huge complex companies tossed around doing valueless meaningless work feel dehumanized. That makes sense to me, but I don't think the problem is complexity, I think the problem is getting paid to be performative instead of creating real value for other people. Integrity misalignment. Being paid to act in ways that aren't in alignment with personal values is dehumanizing (literally dissolves our humanity).


Not even close. I'm 63. You would be nearer the mark if you guessed that I was old, tired, and maybe burned out.

I've had meaningful work, and I've enjoyed it. But I'm seeing more and more complexity that doesn't actually add anything, or at least doesn't add enough value to be worth the extra effort to deal with it all. I've seen products get more and more bells and whistles added that fewer and fewer people cared about, even as they made the code more and more complex. I've seen good products with good teams get killed because management didn't think the numbers looked right. (I've seen management mess things up several other ways, too.)

You say "Maybe it's because I work directly for customers and I know the work I do has an impact". And that's real! But see, the more complex things get, the more the work gets fragmented into different specialties, and the (relative) fewer of us work directly with customers, and so the fewer of us get to enjoy that.


Ah my bad, that was a silly deduction on my part.

Yes I see your point better now, however I still think this is temporary. It's probably something like accidental/manufactured complexity is friction, and I'm this example the friction is dehumanizing jobs. You're right this is a limiting factor. My theory is that something will get shaken up and refactored and a bunch of the accidental complexity that doesn't effectively increase global entropy will fall off, and then real complexity will continue to rise.

I'm kind of thinking out loud here and conflating system design with economics, sociology, antitrust, organizational design, etc. Not sure if this makes sense but maybe in this context real complexity increases global entropy and manufactured complexity doesn't.

Manufactured complexity oscillates and real complexity increases over longer time horizons.

So what you see as approaching a limit (in the context of our lifetimes) is the manufactured complexity, and I agree.

My point is that real complexity is far from its limit.

I'm a lot less confident, but suspect, that if real complexity rises and manufactured complexity decreases we will see jobs on average become better aligned with human qualities. (Drop in dehumanizing jobs)

Not sure how long this will take. Maybe a generation?


I see your point better also. I'd like to think you're right, especially about the accidental complexity getting removed. That would do much to make me feel more positive about the way work is.

And in fact, if you have multiple firms in competition, the one that can decrease accidental complexity the most has a competitive advantage. Maybe such firms will survive and firms with more accidental complexity will die out.


That sounds right to me. It also makes me wonder whether artificially low cost of capital (artificially low interest rates) would result in more manufactured complexity.


that doesn't even consider the buying the competitor across the street and paying lobbyist to have congress ignore you for the benefit of the consumer because with the combined stores you can gain market efficiencies. of course ignore the price gouging that actually happens.


if you've ever shopped at Giant Eagle instead of Kroger you'll understand the flaw in the logic.

higher prices usually equals better service, less busy shopping. get in and get out.

so if your time is worth more than your money you aren't sensitive to price at the widget scale. most widgets are bundled with some kind of service.

I bought some printing and it was super cheap but no service not an email not a phone number nothing. not ordering from their again.


So the products are not the same


I think if you built some kind of game state server it would make a great front end for it. it could even generate the "rooms" as some kind of graph with items, and foes, and descriptions and directions between the rooms. items might need actions to transform or use items.


it's annoying to me that there's not a doc store with vectors. seems like the vector dbs just store the vectors I think.


Elasticsearch and MongoDB Atlas and PostgreSQL and SQLite all have vector indexes these days.


> MongoDB Atlas

It took a while but eventually opensource dies.


My search service Lens returns exact spans from search, while having the best performance both in terms of latency and precision/recall within a budget. I'm just working on release cleanup and final benchmark validation so hopefully I can get it in your hands soon.


Pinecone allows 40k of metadata with each vector which is often enough.


Elasticsearch and Vespa both fit the bill for this, if your scale grows beyond the purpose-built vector stores.


chroma stores both


As does Azure's AI search.


I just use sqlite


I always love the the solution thrown out is we're all going to be plumbers...


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: