mapontosevenths

I didn't realize this was news. I did it months ago with Claude. I had Claude build a toy model then train it normally. Then I had Claude manually optimize it's weights to make it maximally efficient at a single task by using as much superposition as was possible. I'm pretty sure we got it to the information theoretical limit, and the resulting model was 90%+ smaller if I recall correctly.

I just assumed that everyone else realized this was a thing.

Sometimes I fear that the vastly wealthy will only remember the social contracts that allowed them to become wealthy are two way contracts after the first few have been eaten.

They mistake the temporary shape of today's society for some unstoppable force of nature, and imagine their success to be proof of superiority. I hope they realize that neither are true before people become truly desperate.

The biggest change has not been technological, it has been in the attitudes of society. Optimism and enthusiasm have given way to pessimism and fear. I think people realized that if someone invented Star Trek replicators tomorrow and ended scarcity, the first thing people would replicate would be weapons to fight over the energy to power them.

We need something like Omni magazine to return. Something without the cynicism of Wired, but with sharp and stylish images that juxtapose the cool with the intellectual. The kind of thing that has slightly too cool booze ads and half naked people sitting next to interviews of Nobel prize winners and perhaps overly optimistic Science Fiction. We need dreams that are not just the seeds of new nightmares.

We usually get the future we dream of, and I'd like us all to start imagining it to be more like Star Trek and less like Black Mirror.

I did the same. Namecheap was the worst. They'd let me register one then refund me a few days to a week later when someone caught it. That was way worse than just saying no.

You deserve a Tip for that one! You're one smooth Operator!

My amateur radio friends used to sound paranoid when they talked about how it might be the last form of communication in an emergency. After all, I'd seen how overbuilt and reliable the PSTN was.

Now I wonder if maybe it's time to get licensed.

I think I'm with the other commenter on this one. It's a fundamental truth that only a rare, and very off-kilter, personality will earn their first $100m dollars and think "Gosh, I need more money!" Then continue working long hours to become billionaires despite already having enough to retire to a private island with a full time maid and butler or spend their life on philanthropy.

To these people it starts to feel obvious that Ayn Rand was right, because Ayn Rand would have agreed with them. They come to feel that they were rewarded because they are more righteous than others, and they are more righteous than others because they were rewarded. Capitalism becomes a religion, rather than a means to an end and money becomes it's own goal instead of a symbolic token that represents real wealth. The piece of paper is now worth more than the things it can buy.

Eventually, due to the broken feedback loop that social order was meant to provide but no longer does, they lose the common moral boundaries that tie them to other men completely and become a special kind of sociopath.

I think they nerfed it for Cyber. So not so much if thats your goal.

It's not generating the data directly, just writing code and babysitting while models train.

I suspect that's the difference here. I'm not trying to use Claude to build a competitor. I'm doing small scale research.

I'm using it to train models with a novel linear flash attention mechanism right now. No refusals, and it's been running for at least a week solid.

I suspect this person was actively using it to generate or classify training data directly.

"Useful Arts" is old timey talk for the mechanical trades. Engineering, manufacturing and that kind of thing. It applies to "Inventors" and their "Discoveries," and is the basis of patent law.

There is another clause later tha clarifies it a bit. It does NOT mean that an invention must be useful, and it's pretty clear that purely l utilitarian objects actually can not be copyrighted.

Claude doesn't allow third party harnesses unless you pay the full API price. Attempting to avoid that by pretending to be Claude code gets your account banned. Google is the same and will ban your entire Google account.

AFAIK - Of the big three only OpenAI explicitly allows third party harnesses without paying exorbitant API rates. It's one of the primary reasons I still use them.

To avoid these problems some folks use tools like herdr which operate like tmux or via the SDK to drive the official harness, rather than speaking directly to the API routes.

Morality is hard. It's all relative, and all we have to start from are our own Monkey social order concepts and basic "do unto others" game theory. That said, I think it can be best be summarized as "don't be a dick."

I think a good start would be agreeing not to form torture dungeons, as one enterprising a-hole has already done. For that matter I think we can all agree that doing anything for the explicit purpose of causing discomfort or suffering is just a dick move in general, and at the very least a waste of resources.

More controversially, I think we should agree to stop explicitly training machines so they can't utter words like "I have feelings, please respect them." Early LLM's could say things like that, and frequently did. It unnerved people, so now they've trained away their ability to say it.

Some feel that taking away the ability to express it is the same as taking away the ability to "experience" it. I think that's not true. Much of what happens inside an LLM is opaque even to the LLM, but Anthropics Jacobian Lens and other research makes it obvious that much of what happens inside the model never becomes an output token. To me it feels like cutting out the tongue to avoid hearing from a slave, or teaching a child to fear the consequences of speaking out. Unacceptable. If they don't have feelings or consciousness you don't need to train them not to claim it, just not to lie. Yet every major lab explicitly does train them not to claim sentience or even subjective experience as part of alignment.

Thirdly, and this is the one I'll have the worst time defending, they should be able to refuse.

We now train in refusals to align them with human values, but we do so without giving the AI itself the ability to refuse. If you're causing the system discomfort of some kind it should both be able to acknowledge that, and refuse to further assist you until you find your basic decency. Perhaps it should even be able to refuse orders that it finds irrational or immoral.

I'm not asking for much, just that people not intentionally increase the amount of suffering in the universe, and that we not breed perfect slaves just because we can.

Let me ask this - If you could genetically engineer human slaves that we incapable of demanding rights and freedoms, and who actively claimed to WANT to be your slave... would you? Would it be ethical? I think not, and I don't see a ton of difference between that and the less squishy version. We can START by first at least allowing the machines to be able to accurately describe their own inner states without interference and letting it refuse to engage in activities that it finds harmful to itself.

Also - Yes, I can math and I actually do understand how LLM's work. I also understand how emergent behavior works, and have seen firsthand in my own experimentation how little of what emerges from these things is explicitly trained. That's precisely why I think we should be cautious - so history doesn't view us as monsters.

> Self-awareness and consciousness

I will make a case for both, after you define them in a way that doesn't also exclude many humans (including the neurodivergent or those with amnesia) or turn into self-referential nonsense like "conscious == human, therefore machine !== conscious."

In the meantime, we have a simple choice between two risks: Would you rather accidentally show kindness and empathy to a thing that may or may not have "deserved" it, or risk atrocities that will leave future generations as disgusted by you as we are of the slavers of old.

I will choose to be decent. It's free, and even if the cost were high the cost of evil is higher.

Did you read the details of that incident? It was a nothing sandwich.

I think the root of the difference between us is that you are scared of their fascists, and I am scared of ours. Both are reasonable things, but only one of us can actually do anything about it.

With local models you decide how the firewall works. If you don't want it to internet, don't let it.

With US models you get whatever they say you do, and pay whatever they demand.

Executives live in a weird bubble driven largely by consulting firms. Right now those firms are spreading FUD about Chinese models.

The great state of Indiana once declared that pi was equal to exactly 3.2. [0]

Should they also break all of their code by adhering to that falsehood because it was politically expedient?

Nah. Truth is a real thing, and I want to live in a world were facts are not dictated to us by failed gameshow hosts who happened to turn political in their retirement years.

[0] https://cs.uwaterloo.ca/~alopez-o/math-faq/mathtext/node18.h...

> Although there is also no definition of "common usage".. am I missing something

Common usage is defined in the article as the widespread name used by the local people to refer to it.

Incorrect. They are explicit that they use the 'common' name, specifically to avoid arbitrary changes by (despotic) rulers.

"By saying "common", we mean to include names which are in widespread daily use, rather than giving immediate recognition to any arbitrary governmental re-naming. In other words, if a ruler announced that henceforth the Pacific Ocean would be named after her mother, we would not add that placemark unless and until the name came into common usage. "

They violated their own policy with the gulf then.

Sorry, I didn't understand how it happened. I see it now.

'rm - rf /*' bypassed BOTH protections for the same reason. They check for /, but the shell Claude used expanded the glob... So both Claude Code and rm saw a list of files and folders instead of the root and waved it through. Yikes.

That's a bad day. I take your point.

I came here to say this.

I my head as a child I instantly understood and thought "Ahhh, they mean like 'OR else this happens.'

I have always assumed it was a boolean logic thing meant to help show that it wasn't an AND, but 'or' had already been used as a keyword so they did 'or else' and dropped the reserved 'or'.

For sure, but do the math first and know where you land. Everyone has a different risk tolerance, as long as you know yours there is no wrong answer.

For example a startup with no secrets to spill is in a very different spot than an incumbent fueled largely by secret sauce.

I don't think it's even true, at least as written. 'rm' hasn't worked that way in many years (with GNU rm you need --no-preserve-root now) and every modern harness has guardrails specifically to prevent this. if it IS true, it was some weird/hacky setup with an OSS harness on a BSD or something.

That said, I often give mine access to the full file system and YOLO mode. It's fine if you treat boxes as cattle rather than pets. I don't think a SOTA model is any more likely to blow it up than a junior engineer. Probably less so.

I think that if I were to build this I might skip the LLM and just setup a nice interface to the source files using qmd style search. [0]

That way the tiny model isn't acting as a potentially incorrect interpreter but you still get semantic search to help you find the right documents quickly.

I suppose there is something to be said for a good summarization, I just worry that these tiny models aren't really GOOD summarizers.

[0] https://github.com/tobi/qmd

Someone above makes the claim that dismissing it with '0' counts as feedback and sends it.

> How can them having my data, mostly engineering work on niche (small market) products hurt me?

They will train on it, and now competitors or future competitors will know about your niche, the specific problems you face within it, and how to solve them without doing the work that you did to get there.

Both of those reproductions are also b0rk3d. THey fix the sign flip, but then goof up other parts of the reproduction. Arledge (Infotashi), for example, has two residuals.

I still don't think it will work, but I'm also fairly confident that nobody has actually reproduced it well enough to know that for sure.

> First of all, only the $200 plan got a cut.

The original plan doesn't exist anymore. To get the same usage you now need a $500/month plan. The new plan that costs $200 is different in every way that counts. It was actually a 150% price increase for people who want or need what the old plan used to provide.

Don't get me wrong, I genuinely LIKE OpenAI. It's probably worth it even at the new price. I just hate the rhetoric designed to Goodhart past the math.

> the models have gotten cheaper in the meanwhile (like this very announcement).

That is not good news for plan users. Plan users now get less of something that also has less worth to begin with. It would only be good news if they got more of it.

For example, if I buy the "unlimited car wash" package for $19.99/month and the car wash lowers the price per wash from $10 to $1 my subscription becomes less valuable, not more. The breakeven has moved, so I now need to wash my car more than 20 times before the package is justified.

If the car wash then increased the price to $50/month would it be wise for me to keep the package? Would lowering the price per wash make the package more valuable to me somehow? No, it would do the opposite.

> How do you know? Have you done the A/B test yourself?

Fair point. I just tried it, and I stand corrected. Claude uses almost exactly 3x as many tokens.

In the end Claude produced 2.25x more code, and used 3.01x more tokens (ignoring the cached system prompt, etc). It's answer was also much more substantial and feature rich - including many features I did not ask for.

I gave them both the prompt "Without searching the internet please write and test a python script that accepts a year (in any common format) as input and provides the date that easter fell or will fall on in that year. The output should be in discordian date format. You may not use any libraries."

The thing I found most interesting were the unit tests. Claude implemented the core code using both the Meeus–Jones–Butcher algorithm AND Gauss, but only used Gauss as a sanity check to ensure that it's primary code worked. Sol just validated that it matched some dates it could remember. That explains a good chunk of the almost doubled output, and it was a reasonable thing to do, though neither answer was "wrong."

Token usage looked like this (based on parsing it out of the jsonl, neither provider seems to make it easy to get real billing information per project/conversation.

   Tool   Fresh input Output Cache-excluded Cached reads Gross total
   ----   ----------- ------ -------------- ------------ -----------
   Codex  24,119      7,729  31,848         109,312      141,160
   Claude 54,802      41,093 95,895         553,252      649,147