They are still purposely writing a paper, whether that is with the help of an LLM or not. They are instructing the LLM to do the task of finding citations. It's no difference to googling for a paper that explains a specific point. You would still double check Google's output.
I understand how it is supposed to work, but myself and many other academics do consider wanting to write a certain point and then look for a paper that backs that up. I am not saying for the main subject matter that should be done; that should be read prior to starting, say when conducting a literature review on the current research. However sometimes you find yourself writing up at the end and you found a result in your experiment that had a surprising output and you want to have a reference to explain it or back that, that you hadn't considered before. It is especially done that way as it is a time saver as you can't nowadays realistically read every paper on an area as there are too many. The above method works just fine, but you do actually have to check that it covers what you are wanting. You can not just use an LLM and not check if it is correct. Likewise I can't just ask a colleague and not check it is correct.
I think they're saying that frontier LLMs may be usable to spot citations that are correct by shape (a real citation) but incorrect by usage (unrelated to the text)
I kind of hate the idea, but you probably could do a lazy LLM check of every paper and every citation and have it flag possible wrong (second sense) citations for human review
But you'd need a LOT of tokens and a LOT of human-hours
> have it flag possible wrong (second sense) citations for human review
And then what, we're done? How have we avoided the need for the same exhaustive human review? It only saves human review time if you trust the LLM not to miss things.
If the goal is to review every citation fully with 100% accuracy, then, sure, exhaustive human review is needed. But I suspect human review of a random sample would add value, catching some fraud, missing others, but having zero false positives (or as close to zero as human review can get).
An LLM could replace the random sampling. It doesn't need to be particularly good for the approach to provide value. I would worry about LLM bias though.
Another thing to consider is that readers can detect fake citations after publication, report to arXiv, and the author gets banned.
Right, that's what I'm saying. The LLM can identify and prioritize possible cases of academic fraud (or serious incompetence) for human review. As the cost of tokens drops it will become practical to go back and do AI reviews of every scholarly journal article ever written.
One of the issue I face is a blocking function that takes a generic constrained by a `trait` and its async version takes a generic constrained by an `async trait`.
They are saying web based solutions often out perform LÖVE, even though you would expect the opposite because LÖVE doesn't have the bloat of a browser engine.
Love2D uses Luajit and directly calls established game libraries. The CPU usage should be far better for 2D games, luajit is faster than a browser's javascript jit. You can also create single exe games that are a few megabytes and not a few hundred megabytes.
explain that to my webgl TypeScript browser game running at 180+ FPS while rendering a large RPG tiled world with infinite procedurally JIT generated biomes, with heavy processing delegated to webworkers.
As you aren't posting code or stats I can't say much, but I'd bet a native app would still be smaller and more efficient, since you have to wrap what you're doing in an entire Chromium instance and deal with a web stack designed for documents, which is definitionally less efficient than a native alternative. Tiles aren't exactly cutting edge technology.
"Heavy processing delegated to webworkers?" That just sounds like threads but worse.
The first post in this subthread was literally a statement that "A web-based solution is usually better performing, despite all the bloatware necessary." And you literally joined in to support that assertion against "the Electron haters."
And it isn't trauma, it's literal fact. Electron isn't used because it's technically superior to native applications, it's used because web devs are a dime a dozen. It's popular for business reasons, not technical reasons. It works "well enough," but only because computers are really fast but there's only so much slack an OS can take up when even parts of it are Electron apps, and probably vibe-coded to boot.
1Password has really been bugging me recently, all the emails they send have giant link buttons they want you to click without verifying where you're actually going
I feel like that's a result of prior bad experience. Noone goes into university, or even primary school a blank slate.
It's very easy to fall into a trap where education is transactional if that's what was modelled to you or that's how previous teachers treated you
reply