CMay

28 ms·

241.
by CMay 1y ago
A consensus of confidence and self-consistency through consensus seem fine for certain kinds of tasks, especially ones that involve recalling training. This is a bit like scaling up determinism, obedience, collectivism vs individualism...
242.
by CMay 1y ago
Prediction: this means extra work for developers. it also means extra potential abuse and risk through agents. the agents.md will sometimes be outdated or inconsistent with the readme. even the readmes are sometimes outdated or inconsist
243.
by CMay 1y ago
it would be interesting to get GPT-OSS 120B and 20B responses to all of these questions to see how they compare.
244.
by CMay 1y ago
Maybe both? I tried using different animals, scenarios, solvable versions, unsolvable versions, it gave me the correct answer with high reasoning in LM Studio. It does tell me it's in the training data, but it does reason through thi
245.
by CMay 1y ago
The 20b solved the wolf, goat, cabbage river crossing puzzle set to high reasoning for me without needing to use a system prompt that encourages critical thinking. It managed it using multiple different recommended settings, from temperatu
246.
by CMay 1y ago
The paper does not seem to support that human speech has a universal transmission rate or that every spoken language has a universal information rate. They showed that information rate varied by individual and by language, just that it var
247.
by CMay 1y ago
Here's an example: https://youtu.be/y02xPRX6vCE?t=1376
248.
by CMay 1y ago
Here is one example: https://youtu.be/y02xPRX6vCE?t=1376 Basically anything absolutely genuine and authentic, especially where I have enough information to understand the context of their emotion so that I can relate better
249.
by CMay 1y ago
I've been collecting videos for years that make me shed tears. I don't easily shed tears if it feels a little too manufactured. Usually what works best for me is just seeing actual genuine unprompted emotion, though sad scenario
250.
by CMay 1y ago
Would it be possible to have 3-4 words and you have to make sure that each word you produce results in a valid phrase? So not only the final phrase makes sense, but you have to navigate through a path of valid phrases. This way you might r
251.
by CMay 1y ago
Even if smaller more specialized models outperform larger more generalized models for specific tasks with these architectures, that does not logically support your point or serve as meaningful evidence against my post. These systems can hav
252.
by CMay 1y ago
Ah, so you misread. I said conversational intelligence machines, but you focused on LLMs. I don't know where LLMs will lead, but I haven't ruled out the possibility of improvements continuing to surprise us. If anything is unsci
253.
by CMay 1y ago
Can you elaborate?
254.
by CMay 1y ago
This does kind of make me wonder. It's believable that we might either see an increase in the number of new programming languages since making new languages is becoming more accessible, or we could see fewer new languages as the proble
255.
by CMay 1y ago
So, I've interacted with Kaldaien a good bit over the years and while he can have strongly stated opinions that are at times a little at odds with the intensity of an issue, it's not generally that he's fundamentally wrong.
256.
by CMay 1y ago
I don't think they moved or were taken. It appeared that the rocket took off from the corner of the pad and landed in the center, with one camera angle for the corner launch and one camera angle for the central landing. So, I assume
257.
by CMay 1y ago
That would be a bad checklist and overspecialized. You want it to be as generally applicable as possible.
258.
by CMay 1y ago
> But that's not what I'm getting at. What I'm asking you is: what if the President of the United States, in some way shape or form aims to kill Congress. If he succeeds, is that a constitutional crisis? It depends on what
259.
by CMay 1y ago
> I don't understand how. You're saying the President could literally murder the entire Congress, and things would just go on fine? Who would ever run for Congress after that, knowing that if they ever defied the President he c
260.
by CMay 1y ago
He says crazy things to expand people's possibility space, increasing their tolerance for a less bad version of it that might have otherwise been harder for people to accept. This is in Art of the Deal. Same thing with Gaza becoming s
261.
by CMay 1y ago
No, because it doesn't work that way. What people thought January 6th was and what they thought it meant, are not what it was. That event might be meaningful in any other country, it was not meaningful here in the US except in whatev
262.
by CMay 1y ago
The US is unique compared to basically every other country on Earth, so an actual constitutional crisis would take more here than in other countries. A constitutional crisis would be if an entire generation of Americans across a majority of
263.
by CMay 1y ago
What you're talking about is the rhetoric and politics surrounding the current times as opposed to the reality. Trump says a lot of things and he is a man of action, so that makes him a credible threat whenever he says he's going
264.
by CMay 1y ago
Not going to leak my tests, but here's how you can create your own. - Think up a topic that's interesting to you, yet maybe controversial. - Look up primary sources and empirical information about it. - Then look at a relevant Wik
265.
by CMay 1y ago
One immediate observation I have about this model is that it seems to do a better job of filtering out or toning down some ideological disinformation that other models regurgitate from activist controlled Wikipedia articles, at least for a
266.
by CMay 1y ago
Everything is a constitutional crisis now, because nobody really knows what a constitutional crisis is. We're just numbing people down and normalizing the words until they mean nothing, because we aren't using them when they real
267.
by CMay 1y ago
In terms of reasoning models like QwQ or Qwen 3 I didn't waste too much time trying to improve their results aside from coming up with various ways to constrain their reasoning token output with prompts. Even though Gemma 3 27B QAT is
268.
by CMay 1y ago
Ok, so Qwen3 32B works now with the update. It seems much better than the Qwen3 30B A3B for quantized local use from what I can tell so far. Not sure yet how it compares to Gemma 3, but it's at least not clearly worse. It definitely
269.
by CMay 1y ago
Think I tried both the Unsloth and original yesterday, but it looks like the model got updated today so I'm downloading the new version. We'll see how that goes!
270.
by CMay 1y ago
The Qwen3 32B dense model just fails for me due to a template issue, but the Qwen3 30B A3B model does work. I think the more dense the model is at a small number of active parameters, the more sensitive a model can be to quantization. Onl

More ›