sashank_1509

I should have realized sooner though, because I had no idea what that table was trying to say, and that should have been a signal already.

I will also add, AI code is ugly, yes including Opus 5.5. And I haven’t seen this improve much, gpt 5.6 to Astra was a big step down in code quality, and Opus 5.5 is a step up from fable, but I feel 5.6 is at the level of Opus 5.5. To this you might see, I only care about functionality of the code, fine, you might only care if the code was making you money, I don’t, I care for how good, nice and well architected it looks and feels, and expert humans are still miles ahead on that.

For example, recently Opus 5.5, in some multithreaded app I was working on at work, used complex timing analysis with hardcoded time values 1.2, 0.95 etc to ensure that some activities only happen in a sequence (A->B->C). I had to tell it to just serialize these activities by putting them in a buffer. Both versions passed all tests (I’m sure it wasted a lot of time tuning the first set of numbers to pass the tests, but no human would do that nonsense). Increasingly, you feel it’s thinking diverging from humans more and more as the labs lean into RL.

> We need the best and brightest if very country to stay in US.

These are not coherent positions at all.

Frankly I think we need the best looking people, not the most talented/ hard working/ intelligent people anymore, because that might be the only thing AI won’t replace for a while (decades).

I’m honestly surprised pangram got this right, with my human eyes it feels indistinguishable but apparently not !

I think for small snippets and well defined codebases, we probably can’t,

But if we say a fully vibecoded project vs handmade, I suspect there are AI tells outside of basic stuff like bar names etc that a model can learn, and it will be fun to know.

We could probably also do these for maths.

I guess what I’m getting at, is even in these days extremely verifiable fields, I guess AI has diverged enough that I think we can train a detector to separate AI from how a human thinks?

There are 2 types of tasks people do throughout their life.

1. Creative tasks that provide self fulfillment

2. Uncreative monotonous menial tasks that degrade life.

AI takes 1 and ruins it to 2. The jury is out if it will fully take your job, I don’t know if that’ll happen but no one asked for automation of these creative tasks.

When you point these AI bros talk nonsense like cure cancer, Dyson sphere, China bla bla bla.

At the very least, models need to be post trained to stop outputting art, this will never happen though, meanwhile I will enjoy my half a mil a year, while all these AI bros lose their mind over graphs, loops or whatever nonsense they’re concocting up.

Nothing about this is inevitable, AI is as inevitable as nuclear, it seems to require even more capex than nuclear.

Hates as in they would rather shoot themselves in the head than read AI slop, or listen to AI music or appreciate AI art etc.

If we are going to say, humans should learn these skills to produce this and not just keep typing prompts into a chatbot, I agree, but I also think it applies to maths, coding, engineering and everything else. I don’t see why we need to draw an arbitrary line around supposed art.

> my uncle can create software

Is considered great

> my uncle can create music or art

Is not considered great.

I guess humans have agreed, that art is a skill you must learn if you want to produce it, but engineering or software is not a skill worth learning and we are happy if computers can do that for us.

So it’s great and amazing if anyone can create their gizmos and software to help them but not if they create music or digital images, which apparently it is wrong to prompt, and not a sign of a great future.

1. Make mathematics a closed community, only people who join this community can read and work on its stuff. I imagine a VPN that contains all maths knowledge that only members can view, similar to how big tech handle their codebases. It needs to be hard to copy and paste that stuff into the open internet and doing so should involve serious punishments like jail time, similar to posting say proprietary Google code on the internet.

2. Anyone caught using AI is excommunicated from the society, his access to the VPN permanently revoked, and his work is never interacted with again.

3. There will be a bit of a disruptive period as maths moves from open to closed model. During that time AI will keep solving more problems, but as long as mathematicians stop contributing more open problems, at some point AI will stop (either because it has solved all open problems, or it can’t make progress anymore because scaling is no longer economical).

I think Association for Human Mathematics should do this, I think they will be surprised by how much funding the public is willing to allocate to this (public hates AI), because fundamentally humans like thinking about and doing maths, and do not mind dedicating a bit of the surplus produced from society to let others do this, similar to arts. Just because I can prompt a better melody from Suno, doesn’t mean humans will stop wanting to make music.

Of course other mathematical associations can and should form, which fully embrace AI. But if the VPNs secure, these associations will not be competing as they won’t even have shared consensus of what problems to solve. They both will still need to compete to get public funding, which is fine, I think AI math will have a harder time justifying their funding since anyone can prompt a robot.

An agentic OS, that you can speak to a majority of the time to get stuff done, also seems like an interesting direction

Even outside of controls based games, LLMs fail on sokoban like puzzles weirdly enough, see MazeBench, they’re also not that good at chess, probably around 1600 elo.

More and more it feels like as humans we are re-entering the pre 1700s era where most humans never learnt any technical knowledge, circa they never needed it.

The LLM boosters will claim, this is good and fine, LLMs will create an age of miracles and we will climb the kardashev scale or whatever other nonsense. I don’t think that will happen, I’m like 99% sure we won’t even cure cancer. We’re just making large swathes of intellectual work that we invented since the 1700s to now redundant for some reason and hoping that’s a better future, we are liberated from the “toils of thinking”. Well humanity has lived for 1000s of years without thinking of any of this stuff, so we’ll be fine but I suspect the world was richer when we did think about all this.

How much additional productivity do they think admin staff can crank out, and does that even matter. Does a 100X more productive research administrator mean the University will not be “left behind”, in this new world. I think we are officially reaching mass hysteria phases of the AI revolution.

No one’s gonna care for formalizing a 10,000 page math proof with 46M lines of Lean code, unless it is very cheap to do, which seems unlikely.

This gives a sense of embellishment by a clueless layman who think very smart people can memorize every single thing they ever do. Computers memorize well, very smart humans, it’s not clear if they memorize any better than a man of the street. I was struck by a fields medalist recent comment that he sometimes struggle remembering proofs he wrote himself in the past.

In the past building something fast, complicated etc required quite a lot of skill and some time. Now you need far lesser skill and lesser time. This just means what you are building is commodified, and has far lesser value that it had before. Take typescript, if it becomes clear that LLMs can speed it up with no regressions and a week or so of prompting, then most likely every company will have their own typescript compiler and no one will use your new one that you built with your 200$ plan.

A lot of tasks are much easier now and also monetarily they’re not worth the 200$ or so tokens of the ChatGPT plan you’re using, because anybody (or at least a number of people) can do it now.

Rather than being more ambitious with AI, you will be far better off finding a unique combination of skills and services that does not compete / AI can’t provide. Only the big labs can be really ambitious with AI.

Can AGI figure that out. Maybe, but I think it is very likely that AGI will also have to fail in the real world, and then develop those learnings that those companies have done. There is no way to circumvent this computation, you need to do these steps to even know what the correct solution is. I like to think of it as playing chess. Intuition helps in directing exploration, but any agent, human or machine has to explore down a minimum number of paths before converging to the optimal move. There is no algorithm that can generate an optimal move without this search, P!=NP says something similar, exponential search can’t be done in polynomial speed.

Stephen wolfram says something similar, superintelligence, whatever cannot predict the output of the trillionth step of a cellular automate without calculating all the steps before it, there is a minimum amount of computation required, the problem is computationally irreducible below that.

So maybe in the future, with humanoid robots, AGI can decide to start creating a Walmart competitor. Then it will make a large number of mistakes and it needs to learn from, this could mean continual learning or maybe just memory in markdown files if that’s enough. It needs to hit all those failures quickly till it finally has something as efficient and capable as Walmart. And this is to just become a capable competitor, during this time you will keep burning money and to surpass Walmart, you have to do far better which is hard, how much more efficient can Walmart be, idk.

All to say, OAI starting its own company is a losing proposition, huge capital, unclear benefits, high risk. If OAI just provides AI to Walmart, they can make money immediately and Walmart can become more efficient by firing its human workers, which is the problem we are stumbling into.

“ My most controversial take: good character owes much more to luck than people want to admit.”

https://chatgpt.com/share/6adaa65a-e5c0-83e8-8a98-d9d5db0b27...