golol

32 ms·

391.
by golol 3y ago
I think the sad reality is that indeed (1) and (2) are extremely unlikely and that doesn't mean that extra dimensions is more likely. Extra dimensions is just even less likely.
392.
by golol 3y ago
100% chance the guy is just lieing for attention or because he is delusional. He said that UFOs could be coming from extra dimensions. He might as well say that the UFOs are coming from atlantis or the second circle of hell. It's just
393.
by golol 3y ago
Why does there have to be one specific combination of these aspects that yields the one, true "intelligence"? The truth is that it is simply an extremely extremely broad spectrum. Any agent that performs useful computations can be
394.
by golol 3y ago
Intelligence has a pretty clear meaning. Of course you will have a hard time nailing down a definition but that's also true for simple words like "chair" or "car". Artificial intelligence is also a pretty good name
395.
by golol 3y ago
Similar question actually: If I fill my private notebook with hateful tirades against Johm Oliver, and someone finds my notebook and publishes the content online, am I liable for slander or insult?
396.
by golol 3y ago
Yea but the spreading around is the important part. Should just creating the content be illegal? Right now it kinda is but I don't find that to be quite consistent.
397.
by golol 3y ago
I agree that distribution is the key. I am not really comfortable with creation or viewing of such content being illegal. It feels like thoughtcrimes. Distribution is where you are really being immoral. And if you create something in a form
398.
by golol 3y ago
> People viewing explicit images of you without your consent — whether those images are real or fake — is a form of sexual violence. What if you imagine it? What if you describe it to your friend? What if you sketch it on your tablet and
399.
by golol 3y ago
If you connect a bunch of LLMs with a hard coded structure (Like sensory module -> planning module -> action agent or something like that) then for me that is like assembling a symbolic AI system out of non-symbolic components.
400.
by golol 3y ago
Yes, I agree. Symbolic AI is for hardness, which we can use for the skeleton of the agent. LLMs are the fuzziness that surrounds and interfaces the skeleton. Everyone who writes some simple code that assembles a promot is already doing that
401.
by golol 3y ago
Well my whole point is that you have to build up a large codebase of formalized mathematics first, and you have to do it in an elegant way so it is easy to work with. Then verified programs is something we can start to realistically conside
402.
by golol 3y ago
The formalized proofs are useful because they allow you to formalize more difficult proofs. That's circular but let me explain. Formalizing all of mathematics in a formal, computer verifiable language is theoretically easy. The problem
403.
by golol 3y ago
Symbolic AI could never become great because it was missing the connection to the intricacies of the real world that you can only get from data. I think a symbioses of symbolic techniques with LLMs or generally multimodal autoregressive fou
404.
by golol 3y ago
While "perfect third" may not exist as a name for an interval, it does make sense to apply the adjective "perfect" to the interval "third" in order to distinguish a just intonation third (4/5) from a well-
405.
by golol 3y ago
Yes, at least decoder-only transformers afaik.
406.
by golol 3y ago
Most problems can be solved with increased funding, training and recruitment. But there are just not enough people. Yes, we could surely employ enough emergency call reaponders, but in many many governmental institutions you find a lack of
407.
by golol 3y ago
People always only think about the downsides of AI deployment and completely neglect the upsides. > Callers to Portugal's 112 emergency line sometimes wait five to six minutes to get a response. Improving this would be great! > I
408.
by golol 3y ago
It shouldn't be surprising though. If a human makes an error parsing JSON, what do you do? You make them look over it again. Unless their intelligence is the bottleneck they might just be able to fix it.
409.
by golol 3y ago
Just weigh your training data by the engagement it has obtained online. High quality output will have more engagement.
410.
by golol 3y ago
Honestly, I suspect asking GPT-4 to fix your JSON (in a new chat) is a good drunken JSON parser. We are only scraping the surface of what's possible with LLMs. If Token generation was free and instant we could come up with a giant sche
411.
by golol 3y ago
I despise whoever wrote this article. It is almost completely devoid of any real information. Trying to gain an understanding of the situation based on the on the information presented here (Total crash and fatality numbers all time and las
412.
by golol 3y ago
At the time I was a /sci/ regular and great Greg Egan fan, so that Matt Parker video that referenced both of them felt like an incredible crossover episode.
413.
by golol 3y ago
You know I agree that superintelligence is totally possible, but what I disagree with is that it follows immediately after AGI. Suppose you reach the level of AGI. Then your AGI can not just write down the weights for a smarter AI. It has r
414.
by golol 3y ago
The paper states that hallucinations are logical reasoning errors... that confused me. For me a hallucination is the conjuring up of facts, things, references, characters etc. as is convenient for the given context.
415.
by golol 3y ago
It is. I'd argue that alignment is the difference between GPT-3, which was a neat toy for nerds, and ChatGPT.
416.
by golol 3y ago
I don't claim they are impossible (although I personally believe 5 is extremely unrealistic and 4 is difficult), I just have not seen any mathematical or experimental evidence. Experimental evidence would be an example of n accelerated
417.
Different Levels of “AI Safety”
2 points by golol 3y ago|5 comments
418.
by golol 3y ago
If you prompt step by step thinking but the model writes the explanation after the solution, then you have not actually had the benefit of step nx step thinking. The model has to generate the explanation first so that the solution is genera
419.
A BackwardsGPT would be interesting
3 points by golol 3y ago|0 comments
420.
by golol 3y ago
I believe "Can an LLM do X?" is often the wrong question. One should ask "How much prompting does it take to get it to do X?", and more specifically "How specific to the task does the prompting have to be" and

More ›