If you've got a prompt, along the lines of: given some references, check their validity. It searches against the articles and URLs provided. You return "yes", "no", and let's also add "inconclusive", for each reference. Basic LLMs can do this much instruction following, just like in 99.99% of times they don't get 829 multiplied by 291 wrong when you ask them (nowadays). You'd prompt it to back all claims solely by search/external links showing exact matches and not use its own internal knowledge.
The fake references generated in the ICLR papers were I assume due to people asking a LLM to write parts of the related work section, not verify references. In that prompt it relies a lot on internal knowledge and spends a majority of time thinking about what the relevant subareas are and cutting edge is, probably. I suppose it omits a second-pass check. In the other case, you have the task of verifying references, which is mostly basic instruction following for advanced models that have web access. I think you'd run the risks of data poisoning and model timeout more than hallucinations.
For references, as the OP said, I don't see why it isn't possible. It's something that exists and is accessible (even if paywalled) or doesn't exist. For reasoning hallucinations are different.
Yeah especially as in places like London there have been many explosions and house fires originaying from cheap foreign e-bike batteries.
Some Chinese companies care about a long-term brand and place high standards on themselves but it's not true that anything online has passed safety standards. It's hard to differentiate the two due to the amount of fake reviews also.
People who cite interest rates and economy are coping hard. It's all of a sudden a factor now but never was before. Apart from dotcom specific crash, tech has seen relentless growth, until recent years where it was shown firing people had no negative effect.
That service does a lot of the matching that you need and provides all roles from PM to designers to programmers. It's unquestionably better as you're dealing with volunteers. Unless you're picky with who you select and then lucky for them to have enough continued time for you.
Companies just want a resume and straight to interviews.
First registration is $2 but any renewal is $20 after. A lot of first year domains use different pricing and they are charged differently I think by the TLD holders, so it's not just stupid marketing by like GoDaddy who hopes you don't transfer to another registrar after them taking a loss the first year and huge hike after
A lot of super cheap ones have expensive renewals.
There is already socialcoder for non profits, which also provides project management and liaison. This thread would not do better than that service for non-profits, simply because it's more organized.
.io is simply the new .net, what you get when you can't get .com but don't want to rename. Although .io and .ai are really expensive so I wouldn't be surprised if something new comes up.
I have set my OS to dark mode on windows and android.
It doesn't work without extensions like DarkReader etc if you don't provide the theme/CSS as the website. I uninstalled DarkReader because it just guesses how to map things over and some websites became unreadable or in subtle ways things don't display properly.
Maybe I will use whatever is suggested here but you really shouldn't have to for a site as big as HN.
Damn, poor little Google seems really scared. I'm not sure if that info is true. You have to provide the whole source of the entire product and not just a segment of code (eg library) you forked which they do often?
It's a research paper. You can write a theoretical paper and let others apply it practically, which others can figure out the practical aspect and report results of benchmarks, or others can also build on the theory.
This paper only has 2 authors. The other solvers are probably applying technique specific tricks and speedups, and you're working with approximate optimization, it's not that easy to move everything over.
Your first sentence is describing the web based LLMs. There are others that operate against the project structure and has read all the files. But a lot of CRUDs also involve a much easier modular domain structure.
LLMs amplify you quite a bit in productivity if you're doing CRUD which is what most people are doing. You just clean up the code and clarify potential improvements.
Only if the rate of newer services/companies keep up and outweigh the ease of development we should eventually reach. People are currently still maintaining jQuery PHP and other tech that makes doing anything much more complicated than it needs to be.
A good game developer provides entertainment at least. Contrasted to Twitter who under Dorsey&co shipped NFTs and 1 redesign in 10+ years. They had thousands of engineers sitting doing nothing. Imagine firing 90% of a school staff or 90% of a McDonalds staff.
Not to say that moving forward with development constantly is worthwhile either. Zuckaberg spent billions on a Metaverse which is somehow objectively much worse and buggier than VRChat.
The fake references generated in the ICLR papers were I assume due to people asking a LLM to write parts of the related work section, not verify references. In that prompt it relies a lot on internal knowledge and spends a majority of time thinking about what the relevant subareas are and cutting edge is, probably. I suppose it omits a second-pass check. In the other case, you have the task of verifying references, which is mostly basic instruction following for advanced models that have web access. I think you'd run the risks of data poisoning and model timeout more than hallucinations.