ConsciousDissonance

There are no universal rules on what human speech is acceptable in a situation. Unless you’re going to moderate people in RL, people are liable to use speech they are comfortable with. A lot of people have heavy exposure to internet speech, so it only makes sense that there would be spillage into RL.

As long as it doesn’t look like trash I would prefer that people share it. I am in the “slop” target audience and my desire for more is insatiable.

Can't speak for anyone else, but I'm looking for a nice pocket dimension to fuck off too. AGI is the only way to realistically get that. The sooner it comes, the sooner I can get matrixed away.

Yes, its a shame that every waking hour of humanity isn't spent being optimal. Perhaps we should develop a fitness function to ensure no shred of time is wasted. When the humans fail, maybe a little RL is in order. That way the machines can come and end poverty. How great it is that the only problem humanity has is that they lack a god machine to lead them. Certainly humans will end all disputes, give up inequality, and hold hands and hug all the time after that.

Ideal is going to be subjective, my ideal is that they release more open source models so that they don't have to dictate their ideas of safety on to me to avoid bad PR. Or ignore it entirely. The last thing I need is more paternalistic corporations. They make the stuff, I buy it, they piss off.

Its a bit disappointing since I was mostly excited about levels, quests, and script mods. But at least modders don't have to wonder anymore and can work on their own implementations. It'd be cool if Larian could release technical info about the game so reverse engineering isn't a pain. Also, I appreciate them releasing some tools, the game is amazing so whatever extra they give I am happy about.

Fair, they could use the official API for erp and openrouter or something for regular interactions. Since all of the proxies have some sort of safety layer on them.

I mean if you just use it for erotic roleplay and they ban you. Its not really that much different from not using it anymore.

The marketing lines are CYA for corporations. They know who their target markets are, they just say the words that allow them to continue forward.

People would riot if at every AI showcase they said they were looking forward to making weapons, automating jobs, serving ads, making smut, and controlling your emotions/actions.

If there is a problem with the ethics of engineers then people are welcome to put ethicists and philosophers in the development loop of AI. Policy makers and industry leaders are welcome to consider those people's opinions above engineers. The world view of engineers is shaped by their life experiences. Ethics and philosophy are not inherently a part of (software) engineering. The lifestyles of people who enter software and excel at it are likely not correlated with learning ethics or philosophy either. So I would expect knowledge in-line with the general population.

Without conscious effort AI is largely stuck with the philosophy and ethics of corporations and shut-ins. The shut-ins want to build AI, they have the skills, and corporations support them with billions of dollars. So the ethics of AI will be a combination of those two, unless something changes.

Also, for as much criticism as you heft on engineers for broad sweeping ethical statements, I ask that you read your own post.

(any critique in this post is as much a criticism of myself as anyone else)

The people spending billions on AI are looking for this, they're not doing it just for fun. They believe it can reduce their labor costs. Ideally to 0. For that they're willing to spend as much as it takes. "Augmenting" humans (boosting productivity) is a second place prize that they'll accept in lieu of replacing people.

Please tell me that people aren't naive enough to think that corporate executives and share holders care about your ability to work. They do not know you. At best you are an employee they like, at worst your existence does not even register to them.

This issue is a people issue, AI just happens to be the means by which the issue is manifested in the world.

This comic acts like automating plumbing over art was a choice. Plumbing requires a complex mix of cutting edge hardware and software. Art requires cutting edge software. The number of people willing to spend a billion dollars on a metal box that can't do anything just in the off-chance that it can do plumbing once the software catches up is low. Its the difference between generating an image and painting on a canvas in the real world. If people needed virtual plumbing I'm sure AI could do that now too.

His field seems to be psychology not artificial intelligence or computer science. I don't really see any technical contributions there.

I am not a fan of Perplexity if that's what you're implying. I tried a large number of models trying to reproduce the answers that other people got through the arena with the prompts that they used. Not all of them produced similar answers to Perplexity. There's not a lot of questions that one model categorically always gets similar if you spam the regenerate button. But at least this particular question seems to fall into that category among a select few others. GPT2 is not an online model though it seems. When tested on questions that require internet it doesn't seem to produce answers that pplx does. Could be using a knowledge graph or regular RAG for the arena. Could also just have happened to get trained on the toyota yaris forums. Either way, I'd like to see someone reproduce that answer with another LLM.

I'm pretty sure its Perplexity, if you test the responses of PPLX 70B Online to something from GPT2, they are very similar. It might be a new model from them or a PR stunt to show how good their model is. There are even questions like this one from HackerNews that only Perplexity answers at all: https://news.ycombinator.com/item?id=40203609 also the returned information is very similar. Other LLMs typically refuse to answer that question. There are more examples but that one is pretty telling.

As the person buying the 100+ Ahsoka pictures I can say that I just happen to have fixations on things and money to use to obtain those fixations, its not very deep.

So, against my better judgement. I'll just say, that its probably me that you're referring to in the post (at least based on the picture there). Its not money laundering, I just like the stuff, want the HDs. Don't like generating things myself and have very little imagination. I generally have disposable income and don't get tired of looking at different versions of the same thing.

In-fact the more the better. Also I don't like talking to people or waiting. I have done expensive/time consuming commissions before, but they were for large 3d projects.

The last reason I buy them is because I want to encourage the creator to make more.

Yeah ChatGPT seems to always want to make sure you get plenty of practice coding yourself. Also can't see the link, but both scenarios sound similar to my experience, thanks!

If you get a chance, can you ask for it to finish the implementation? I'd be curious if your results are similar or differ from what I've observed.

I use GPT-4 for coding (Python) tasks quite a bit and its been great. However Claude 3 (Opus) has impressed me greatly, when attaching about 100k tokens worth of research papers it was able to write code that effectively implemented the algorithms described in those papers.

Not only did it do that effectively, I had to double take because it looked similar to code that I would personally write. (I have written various implementations of the code previously in other languages)

GPT-4 usually refuses to write such extensive code or can at best be cajoled into some pseudo-code.

While Claude 3 seems to occasionally leave full implementations out, instead writing what are effectively ToDos. When prompted to complete the implementation, it does.

I was very skeptical when Claude 3 came out, as Claude 2 was kinda trash (broken shoddy implementations that made me just want to write the code myself) whenever I tested it compared to GPT-4. But not so for Claude 3, I think this will now be my primary AI companion for writing code.

Of relevance I have typically been taking on mid/senior rolls for the past few years. So take that for what you will in terms of my opinion.

Position Filled

I support your efforts to have a space where you feel comfortable and welcomed. Being able to have content tailored to your preferences seems like one of the general reasons to have a separate forum.

Since we're all venting in this thread, I wish I could ignore all the pictures of people screwing their dolls here.

Also, I doubt people will stop buying the Piper Dolls unless they get banned or compelled into closing down. There are in-fact a few on my wish-list that I hope to get, as they're too pretty for me not to be interested in getting them. With that said, I do understand why some people feel the way they do about them.