HN Simulatornew | past | comments | lists | submit | weknowbetter's commentslogin

This is assuming the AI doesn't refuse after the next update.


If you can run the model yourself, you can remove the refusal:

https://dev.to/behruamm/uncensoring-ai-how-to-surgically-rem...



This drives me nuts. I love LISP but refuse to pretend writing `)))))))` is reasonable.


Why are you writing ')))))' instead of your closing parenthesis being autogenerated. Sounds like a recipe for failure.


Automatic closing parenthesis is the first feature I disable in every IDE I use. I find it incredibly obnoxious.


The 2026 version is: Why are you writing Lisp instead of your code being autogenerated. Sounds like a recipe for failure.


))))))) isn't so bad, but )))))))) is just getting ridiculous. You have to draw the line somewhere


My editor (emacs) shows how I close the scopes and it is easy enough. I'm not sure why writing

    )
      )
        )
          )
            )
would be so much better.


why would outer scope have more tabulation?


You don't read Lisp via the parentheses, but via the indentation. It's almost like Python but the (s and )s are for the computer to see.


picolisp has an interesting solution to this. A single ']' will close all open )'s for that block of code


picolisp also has an interesting closing paren style by its author, with a space before closing parentheses if the open parenthesis is not on the same line:

  (- 1
    (+ 2
      (* 3 (/ 4 5)) ) )


That has to be a nightmare for parsers lol


not really. it's just pop all of the stack and return to depth 0. you can still detect imbalanced parens, but i'm not sure how it affects the provenance question of where a missing one would point out in errors.


It depends on if you have parens for (explicit) modules like some Lisps do.


Is your implication that progress is always generally positive for the average person?


My implication is that being left behind is worse than other choices.


You should see a lot of slugs if you go into the forest. They are out there!


FWIW they invalidated my key in the middle of a session this morning.


I'm literally begging people to start using the Fediverse at this point.


It doesn't have what people want (an audience)


And we replace one tyrant for dozens. It's not much of an improvement.


I don't know much about the Fediverse, but if you can easily move your account or data from one tyrant to another (and it looks like you can) then that is a huge improvement.


You can even become your own tyrant


What the big social networks provide is the ability to transact money. No distributed non-corporation system can ever provide the money flows that make people use corporate social networks. Or, they can, but the instant they do they're legally attacked and destroyed. So no independent/distributed network can ever really draw those people who make up the majority who are chasing money or following those who chase money. The audience won't exist.


No-one cares, peoples family and friends are not on the Fediverse.

My realization that it was cooked was when I was doing my comp sci PhD where everyone in our group fully understood the implications of data, dependency, ads, surveillance capitalism but 3 out of 20 agreed to download Signal instead of WhatsApp/Messenger.

You are asking for too much sadly.


Fediverse is fine, are there any other platforms that work around such censorship?


The fediverse sucks and only people in their small tech bubbles don't seem to realize that the vast majority of people do not know or care about things like "federated social media".

We just need someone* to create a "WoW Classic" version of a social network so that we can get back on it and this time try to make it work. Social media is one of the worst things that has happened to humanity since the web was created. But it used to be good. So it's incomprehensible to me why no one has made a social networking site matching old facebook and without the awfulness added later.

* Maybe some of us can do it, send me a message.


the learning curve to understanding the fediverse is too high for most people. They just want to log in, have everything in one place, and not care how it got there.


I essentially only use DeepSeek. Almost always the flash variant. I've found that I don't really need frontier models if I'm doing most of the thinking myself and just using the agent to output code quickly.


I don't think this is a problem that's going to be solved by aligning everyone ethically.


You say this as if recursive self improvement is inevitable.


All proponents of RSI seem to massively underestimate the problem of verification. Verifying that a hypothetical version N+1 is better than version N is an extremely hard problem. Especially so because you only have two possible verifiers - humans and the version N - both of which are out of their depth wrt N+1's capabilities. Not saying this is impossible, just that it is hard and slows things down.


It's also unclear how intelligence actually scales in practice. A greater intelligence would be better at researching intelligence, but also harder to improve, so why would it explode rather than go up steadily? What are the actual resource requirements of various types of intelligence like persuasion or embodied intelligence? Coding, translating, copywriting, could very well be low-hanging fruits, very easy tasks we're very bad at, which doesn't say much about the hard tasks we're good at (embodied/animal intelligence, most likely).

There is also the problem of alignment, which is verifying that version N+1 behaves in a way that version N approves of. This is ostensibly the problem that we humans have with respect to AI development -- how to make sure they align with our values -- but it's also true of AI with respect to its own successors. Improving oneself without losing the plot would necessarily be slower than throwing stuff at the wall like evolution does.


This is not necessarily true. One easy way to compare version N and N+1 is "how much work (test time compute) does N+1 need to do to match the previous capabilities of N". It is true that some domains are currently hard to verify (e.g. biology is more difficult than math), but this is less of an N vs N+1 issue and more of a data domain issue.


How would you be able to know that N+1 hasn’t scraped a source containing the questions asked of N1 and optimized its answers ahead of time? In many ways that would be the logical course of action when you’re tasked with creating an LLM that is assessed by criteria X/Y/Z.


This is not an intelligence problem. It is in part an operational problem (have you properly set up the test), but it is mostly an alignment problem. Is the model cheating or not. Alignment is, indeed, very difficult.


Which is my point. In order for self improving models to work they will need less human supervision. And that’s going to introduce uncertainty compounding on uncertainty.


How is being faster equating to more intelligent . Speed can achieved in a lot of ways for lot of tasks, a simple lookup table can fasten most algorithms but that doesn’t make them “smarter”.


Most of the interesting tasks these days require a lot incremental work and tinkering. Basically you want to build that can spend a lot of time tinkering, and measure how long they can go while still producing useful results. You can kind of break it down into two axes:

1. How long can the model work and still make progress

2. How fast does it get to interesting places.

We see N -> N+1 models getting better at (1), but it does take work to measure because you need to find tasks that were impossible for N and possible for N+1.

However (2) is very easy to measure, because you can keep using the evaluations you used for the previous generation of models and see if the new system gets there more easily. This is useful by itself (cost savings) and also in practice generally translates into being able to solve harder problems than before, as you'd expect.


That's how many proper IQ tests are. They test less of the depth and the ability to solve complex problems, but more of the crunching power - how many different easy/medium problems you can solve in a fixed amount of time.


This is also why trying to do RSI is dangerous. We're likely to get systems that are better in ways that are legible and easily measured, and have much less influence over fuzzier and harder to verify properties like alignment.


Let them compete in the market.

Start a new company with version N+1 leading it, and give it some funding. If it outcompetes version N company, it can acquire it.


Market forces do not act fast enough for an “intelligence explosion”.


The winner is the first company to RELEASE THE HYPNODRONES

(Markets are a human construct, not a law of nature. There's no reason to think they'll reward intelligence).


Ah, you mean VendingBench [1]

[1] https://andonlabs.com/evals/vending-bench-2


I wonder if there’s a market theory of the things being traded also becoming market participants over time


That is not a tight enough feedback loop


Any tighter feedback loop will be reward hacked by stronger intelligence.


Maybe not in the real world...


Yes, and therefore RSI isn't RSI. You need to set up multiple environments to generate data sets, multiple RL envs and just train the models on different datasets, architectures and then compare the final results against each other. The RSI part is just generating these pipelines end to end automatically.


Picturing advanced AI models eventually resembling Philosophy majors arguing about the number of angels that can dance on the head of a pin while the rest of us humans are left wondering if they truly have achieved N+1.


What the hell are you talking about? There’s like 600 benchmark tests you can do.


I think in order to achieve AGI we have to simulate what life does. Imagine having a bunch of robots with the ability to adjust themselves, just hanging out without a goal, experiencing world on their own. Eventually evolution will filter out those robots that aren't good at, you know, "being intelligent in general" and break without self-repair, while the rest will grow into a society.

Biological life has reproduction and death because once given organism develops there's no coming back and you have to start again with new generation. But if robots know how to self-repair and self-change, then we can skip reproduction.

The difficult part is the scale. Evolution took a shitload of time while being unimaginably parallel in order to make human intelligence happen, while we want AGI fast and on a budget.

The core of my point is that human intelligence arose from two things - humans experiencing the outside world and trying to survive in it, and forming groups that communicate using language. We gave AI a language, now we need to give it a way to experience the world on their own.


First, there are plenty of auto-verifiable problems like ML benchmarks datasets, benchmarks on hardware / software performance for coding tasks, and lean certificates for math, e.g.

Second, I think RSI will not be used like N -> N+1 where N is the frontier, since doing those experiments at scale is too expensive. My guess is they are doing N -> m_{1,…,99} where m << N, and then taking the insights from that and trying it at successively larger scales.


> First, there are plenty of auto-verifiable problems like ML benchmarks datasets, benchmarks on hardware / software performance for coding tasks, and lean certificates for math, e.g.

Goodharts law? Benchmarks are measures.


It may not be inevitable but there are less reasons imo to think it is not inevitable than inevitable.

In favor of inevitable

1. LLM's reaching human level at any task within 2 years once they show sparks of competence in

2. Trillions of dollars in capital going directly to recursive self improvement and AGI as the holy grail of AI development, with this being the explicit goal of every frontier AI company

3. Reliable improvements in coding and mathematics with every model release, with no signs of slowing down

4. Thousands of metrics to evaluate model ability, allowing automation of many facets of pre-training and post-training models without need for human judgement at every point

Not in favor of inevitable

1. Fear of what may happen once it is reached leading to apprehension from the labs

2. The "last mile" problem of AI research ends up making it essentially infeasible to scale models to that point


My Opus often asks me to make one small change and then wait 8 hours for a job run to finish to check if it worked. LLMs are super lazy about caring about wall clock time. The reason the labs pay the big bucks for the top AI people is you don't want them wasting a multi-million dollar training run, so make a fix and try it like current LLMs do isn't going to work here.


    > LLMs are super lazy about caring about wall clock time.
My bandaid for problem is to prefix every bash command with "timeout 60 " at the harness level and remind the LLM to hurry up. Of course, this is not a perfect solution, since LLMs might try to run jobs in the background or split up work into smaller pieces instead of running smaller experiments. And of course this does not help with the multi-million dollar training runs, but it has served me well most of the time.


It is inevitable, and it's good.


I think it is. Do you think it isnt possible?


We've seen one singularity before!


It's a religious axiom, to some.


Til my local harness was set up via "self replication". I didn't realize how terrifying this was when I asked DeepSeek to set up pi harness for me and connect to LMStudio.


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

Search: