If you're using AI to make content you're saving time and money but you're making shitty content. If you can't see the shittyness of the content, it's all the more reason you shouldn't be doing it.
3 Boosts
1 Like
@Noisecolor @grumpygamer I am not a scientist and thus cannot comment specifically on that.
That said, I am a programmer. When I look at some of the trash that people have vibe coded, it makes me wonder if these scientific papers are actually any good, or if they just look good.
I've seen "working" code come out of AI, but all too often the actual underlying code ends up being brittle, unmaintainable, and a nightmare to reason through. It may look to an outsider like good code, but any even moderately competent programmer understands the perils of the massive amounts of technical debt these things needlessly introduce.
Maybe it's different with scientific papers, but I doubt it.
@me @grumpygamer @Noisecolor It's the same with scientific papers. When there's an automated formal solution checker, these AI-generated papers pass it, but any human trying to decipher the proofs will have the worst time of their life.
LLMs can only ever generate mid-grade content, the absolute average of everything consumed in training. It's designed that way. To statistically choose the next most likely word/phrase in a string. It can never "discover" anything new.
If it's "good" that's because it's regurgitating something good a human wrote, that the LLM consumed.
Why would anyone be excited by this?
@Noisecolor Part 2: As you can see, people do not want to understand your point. AI is bad and there is no way to turn that belief around or have a more detailed opinion on good/bad topics. Intend to not discuss this with hardliners, waste of time. I discuss a lot with people that are open minded. AI is a great tool and a risky thing at the same time. We should use the first and try to avoid the latter.
One thing is for sure: it is here to stay.
@wacholderpolka @Noisecolor This is total nonsense, because it can be said about anything.
@Noisecolor I agree. AI is no magic, it’s a complex tool. It creates things that can be good or bad, like every tool. If a scientist does not check the tools’ output, it is a bad scientist, not a bad tool.
AI created great content and can give the user the power, to create something he wasn’t able before. Many people have excellent ideas but 99% will never be realised because of missing money, knowledge…
AI can partly fix that.
@wacholderpolka Except even in this case the "user" didn't create anything. The AI generates the things according to criterion that are partly obscured by design and partly controlled by a commercial entity.
There's nothing to fix here, if you want to actually create something, you have to do it yourself. Not everybody is meant to be author, visual artist, musician or coder, and we're completely unmaking what it means to create something as AI slop generated at scale overtakes human creations.
The "good enough" attitude is killing human culture for the profit of a handful of overvalued companies. This will not end well no matter what happens.
@wacholderpolka @Noisecolor Is it a tool though? A powerdrill where you can’t be sure that the diameter is right is not a tool. Even if it tells you „you‘re right, you needed 5mm and I drilled 8mm. That’s on me. I will update my skills.md to make sure I drill what you‘ve told me. Should I order screws for 8mm holes for you?“
@Noisecolor @grumpygamer Also, if a scientist is publishing 'something they don't even understand themselves', then they are no good as scientists, tbh.
@UkeleleEric @grumpygamer
Maybe there will come a time where this will become the norm. Human intellect will not be enough. Its not that hard to imagine, considering we have tons of scifi on the subject.
@Noisecolor Sci-fi is one of the worst predictors for what will actually happen because most authors knew what they were writing about was impossible but it did sound cool. GenAI is currently fooling people including you into thinking sci-fi could be real after all, but the physical and evolutionary constraints haven't changed, people are just mesmerized by a text extruder.
@Noisecolor @grumpygamer codswallop. Random joining of the whole of what the tool has been fed on (including misinformation, proven falsehoods, disproved hypotheses, flawed studies and opinions) is most unlikely to come up with anything usable, and the percentage of 'possible' 'good' stuff would be miniscule. Tell me you don't understand what AI does without telling me...
@UkeleleEric @grumpygamer
So unlikely that it had no chance of solving any mathematical problems? 😂
@Noisecolor It does, but so unhelpfully that several Field medalists had to take a stand against blind mathematical problem solving because the scientific benefit isn't the solution itself, it's the journey a human took to solve it.
@Noisecolor No good scientist is releasing a paper they don't understand. Doing so makes them not a good scientist. You thinking that is true kind of proves my point.
@grumpygamer @Noisecolor Exactly.
The entire point of a scientific paper is to communicate scientific results to (at least) other people working in the field. If not even the authors of the paper understand it then no one else can be expected to either, and it's a complete failure at the one job that a scientific paper has.
@spacelizard @grumpygamer Not if it leads to other discoveries.
@Noisecolor @spacelizard Not if you can't understand it!
@grumpygamer Dunning-Kruger amplifier?
@jbqueru How much people put faith in the Dunning-Kruger effect which has been shown to appear even with random data is a good example of the Dunning-Kruger effect.
@grumpygamer
There are cases when scientists publish papers that they have written with ai and don't even understand themselves. But the papers are good. That happens a lot as humanity has already produced a lot of knowledge and just hasn't connected the dots a lot of times.
That's is a good example why you are wrong with your absolutist statements. Good content can and does come from ai. You just hate ai.