golol

17 ms·

301.
by golol 3y ago
It's not like you can hide the web from OpenAI. They could just use a secret crawler. Or buy the data from a third party company.
302.
by golol 3y ago
But OpenAI is not selling the rights to any images, or are they? When I pay for Dall-E, does the contract give me any rights for a work? If not then there is no issue.
303.
by golol 3y ago
If you release a media with copyrighted content it is IMO first and foremost your problem. Now if you have some contract with the guitarist that specifies that he produced a sample he has the rights to and sold it to you, but he clearly was
304.
by golol 3y ago
Very bizarre, the implication is literally reversed in the analysis of the problem versus the actual problem.
305.
by golol 3y ago
I find this article in quite bad taste, and a look at the front page of the website explains.
306.
by golol 3y ago
I really idolize Greg Egan, I feel like he has staked such a unique position for himself in the cultural-academic landscape. I like to say that he writes science-fiction while many other authors really write engineering-fiction or sociology
307.
by golol 3y ago
I don't feel like writing a long text so I won't really give an appropriate reply, but for me my point is that all this talk about American history and American sci-fi authors is not actually relevant at all to the common themes o
308.
by golol 3y ago
IMO the author is wrongly characterizing a lot of fundamental themes in sci-fi as specific to some conservative 70s American culture when they are actually fundamental ideas related to economy and technology, more like econ-fiction rather t
309.
by golol 3y ago
The oroblem is that gaining this responsiveness fundamentally makes your task much more difficult. Instead of just creating a picture you're now writing code that has to be maintained. In my philosophy arxiv is for documents which are
310.
by golol 3y ago
IMO pdf and HTML optimize for different things. pdf is easy and pretty. HTML is easy and responsive. But making pdf responsive is impossible and making HTML pretty is not easy. I think having arxiv for well-polished pretty documents, not re
311.
by golol 3y ago
If you build a complex Chain if Thought style Agent and then train/finetune further by reinforcement learning with this architecture then it is not a band-aid anymore, it is an integral part of the model and the weights will optimize t
312.
by golol 3y ago
Yea but some people seem to expect that GPT-4 should follow soon after GPT-4 because GPT-4 followed soon after ChatGPT.
313.
by golol 3y ago
In my opinion the field is not that rapidly advancing. The major breakthroughs, where something was really much better than everything before were the following: GPT-2 February 2019 GPT-3 June 2020 CPT-3.5 December 2022 GPT-4 February 2023
314.
by golol 3y ago
Tierification of AI models is not some business strategy, it is a necessary consequence of the reality that AI is massively compute constrained right now. The size of a model is extremely important for inference time and cost. It just doesn
315.
by golol 3y ago
Well yes, thr Turing has to be defined exactly. That is not mvoing the goalposts because I never considered the Turing test to be a quick 10 minute chat. I am sure that Turing in his original publication soecific the conditions manner under
316.
by golol 3y ago
Alternative version: Start a long distance relationship with a chatbox and see if it is engaging.
317.
by golol 3y ago
The Turing test is never dead: Just deploy the curret best chatbot as a remote consultant in a technical field and see how well it works. This is a Turing test.
318.
by golol 3y ago
Just wait until crash test results come out before speculating wildly. Hard shells can also crumple.
319.
by golol 3y ago
Listen to what he says in some interviews, he is moderately right wing but there is no reason to think he would be a nazi.
320.
by golol 3y ago
Musk may be an asshole boss and he may post cringy right wing opinions on X, but trying to boycott his businesses based on those actions is quite disproportionate compared to the behaviors whole industries are engaging in regularly. Elon an
321.
by golol 3y ago
I disagree. Dunning Kruger is not a statement about predicted score correlating with actual score in some way. It states that predicted score does not correlate well with actual score. This can be rephrased as the prediction error having a
322.
by golol 3y ago
Yes, my point is that the Turing test has not been invalidated as a tool to detect AGI.
323.
by golol 3y ago
It's not nihilism, I just expect absolutely nothing productive to come out of an attempt to regulate and control a technology that doesn't exist yet. Please show me a single result of alignment research that has real world applica
324.
by golol 3y ago
>Gary Marcus
325.
by golol 3y ago
AGI is general intelligence at human level, which GPT-4 does not have. Passing the Turing test is ill-defined as it entirely depends on the timscale. I'm sure Turing gave some time limit originally but we do not have to stick to that.
326.
by golol 3y ago
No, then you start worrying when the technology is there. There is no point in worrying about it now when it is fiction. If AGI/ASI does not arise in some kind of uncontrollable "foom", you can always examine what you are dea
327.
by golol 3y ago
There is foom as in decently fast self improvement to super human level, but that is far from a dangerous foom scenario where that AI then escapes. You can just turn it off. It can not spread through the internet or something like that, tha
328.
by golol 3y ago
Look, there is ASI/AGI safety, and "please don't swear and scam and misinform, and be unbiased" safety. ASI/AGI safety is only a concern right now if you believe in foom , i.e. that an AGI would rapidly self improv
329.
by golol 3y ago
LLMs are built that way, with prompting this behavior can certainly be achieved. It's not going to work oerfectly and jailbreaks will still be possible, but not so easy.
330.
by golol 3y ago
LLMs are the first instance of us having created some sort of general AI. I don't mean AGI, but general AI as in not specific AI. Before LLMs the problem eith AI was always that it "can only do one thing well". Now we have so

More ›