kennywinker

7 ms·

1.
by kennywinker 3d ago
> Doe not following a moral rule imply not understanding it? No. People definitely do immoral things knowing they are immoral. But if you are incapable of understanding anything, as I believe LLMs are, then you are incapable of underst
2.
by kennywinker 3d ago
My position is that cheating is too slippery a concept to train out. But hey, I am no expert, so maybe I am wrong there. But I'm pretty confidant morality is too slippery a concept to train in. As someone else in these comments said:
3.
by kennywinker 3d ago
Once the model is released, the LLM no longer learns.
4.
by kennywinker 3d ago
To be clear, I was being entirely silly - mostly to express agreement with your point that our current laws aren't really built for a world with lots of agentic LLMs running around in it.
5.
by kennywinker 3d ago
When we put the LLM in jail, do we put the entire model in jail, or just the instance that committed the crime? How do we prompt it to let it know it's in jail?
6.
by kennywinker 3d ago
> I don't see what calling these systems "not intelligent" gets you here. I am trying to get at an idea. That these systems lack a mind that can understand morality. That they don't have the ability to experience cons
7.
by kennywinker 3d ago
jinx! :)
8.
by kennywinker 3d ago
I think my position, as overcomplicated as it is, is that even adding a reward for prosocial behaviour during LLM RL will not lead to perfect alignment. You can train it not to cheat at chess by altering the moves, but it will cheat by peek
9.
by kennywinker 3d ago
Watch out, your human exceptionalism is misleading you. Plenty of animals have a sense of morality ( https://pmc.ncbi.nlm.nih.gov/articles/PMC6404642/ ). For social animals (humans included) behaving morally can be
10.
by kennywinker 3d ago
> The implication of what you're saying is that pathological liars and perpetual grifter snake oil salesmen types aren't intelligent. No, I don't think that is the implication. I think you're making the "if all x
11.
by kennywinker 3d ago
I responded to this idea more in detail here: https://news.ycombinator.com/item?id=49686196 But tl;dr: even if LLMs do have the intelligence to understand the consequences of their actions, there is no way for them to exper
12.
by kennywinker 3d ago
Sure, laws are incomplete. Legal systems work by imposing consequences into a moral decision. Should I rob the bank? I will have money, which I like - but I might get caught and lose the money and my freedom, which I don't like. For mo
13.
by kennywinker 3d ago
Without access to reasoning traces, we can't know that - someone inside openai/anthropic would have to run the test - and we'd have to trust their results. I would be curious to see how the open weight models do on a test lik
14.
by kennywinker 3d ago
To me this underlines the fact that these models aren't intelligent. Like there is something like intelligence that emerges from them, which is what we see when we look at benchmarks or ask it to solve hard coding problems. But there i
15.
by kennywinker 3d ago
Does it help with benchmarks? Are you saying there are examples of benchmarks where the models have solved the problem by cheating?
16.
by kennywinker 4d ago
That's great advice to all people who fail to see large hazards fast approaching. Like, say, a fascistic kleptocracy that's quickly forming in a large western country.
17.
by kennywinker 4d ago
They don't even need to fudge the data on the studies they have published. They just need to hold back other studies that contradict the idea. Pouring over the published data looking for flaws will never get us access to the unpublishe
18.
by kennywinker 4d ago
> Qwen flash-next and GLM 5.3 to show some of the other strong open weight models, and they predictably dominate it Absolutely - no argument from me here. Bigger is very clearly a lever you can pull to get more out of a model. > But t
19.
by kennywinker 4d ago
A classic big ai talking point. Color me skeptical.
20.
by kennywinker 4d ago
> But along the same dimension, consider that currently only about 40 - 60% of the world uses AI for only about 5 - 15% of their work hours. Ah yes, i am constantly lamenting that my barista isn’t using ai enough ;) Hopefully you’ve adju
21.
by kennywinker 4d ago
https://artificialanalysis.ai/?models=gpt-5-3-codex%2Cqwen3-... Shows qwen3.8-27b along side seven larger models of ~similar vintage. Only one scores above 27b. Many of those are closed models so idk their exact parameter c
22.
by kennywinker 4d ago
> LED lightbulbs use 80% less energy than incandescent but you don’t see people using 5x more lights on their homes. I mean… some homes definitely do. You must have seen those houses that are all lit up front the outside by lawn mounted
23.
by kennywinker 4d ago
Perhaps. But their huge valuation is based on them supplying the massive buildout of data centers that’s happening / planned. If that dies because a lot of people’s needs turn out to be met by a system at home they can run a 30b-150b m
24.
by kennywinker 4d ago
Does that not prove my point? Bigger doesn’t automatically mean better. Quality of training data, and model structure, matters as much or more than size
25.
by kennywinker 4d ago
> The GFC would not have been prevented by a reserve requirement. Who said anything about that? > The problem didn't originate in the banking system I guess i consider mortgage lending part of the banking system, but no matter -
26.
by kennywinker 4d ago
Fair enough. I’ve barley touched oss-120b, so i didn’t know it was so few active params. For a direct comparison, qwen3.6-35b-a3b is still better at coding than oss-120b. And Qwen3.8-27b is still better at coding than opus 4.1. Yes, if you
27.
by kennywinker 4d ago
This is definitely the position of the big ai companies. But it doesn't match my experience. Qwen3.8 27b is clearly smarter at coding than MANY bigger models. gpt-oss-120b for example, is almost 4x the size, and performs way worse at c
28.
by kennywinker 4d ago
Yes. Totally. Will a taxi driver call the police to let them know where I was if I am a journalist on my way to meet a source? Networked cameras with ai monitoring is begging for totalitarian oppression. A cabby calling the cops because I d
29.
by kennywinker 4d ago
Assuming it’s all going to be vibe coded garbage, yeah it will be much more complex. Like a toddler writing a symphony.
30.
by kennywinker 4d ago
I’m not worried about the lack of reserve, i’m worried about the money shell game where private companies can drive inflation or deflation whichever serves their profit margins best. The 2008 global financial crisis was a result of this, so

More ›