[xen2xen1]

I've worked with legal software professionally a few times, and kept thinking software won't replace lawyers, it'll replace paralegals. Not all, but a great deal of grunt work could be replaced with AI and good forms. Now if we could get them not to lie?

[ssnistfajen]

There still needs to be a human manually proofreading them, even if it's as trivial as looking up whether the cited cases actually exist. Same goes for code outputs from currently available AI tools: some still needs to debug them to make sure it works, and fix any troubles that spring up during deployment and production.

Of course after some number of iterations, AI may gain the ability to self-check and self-correct to the point of achieving 100% accuracy, but I think that milestone will quickly lead to AGI and job security will quickly become an obsolete topic for every industry anyways.

[xen2xen1]

Maybe. But I keep thinking it's not going to replace the single paralegal that a single lawyer has, but it might reduce the 9 paralegals to 6, and then 4, and then..

[ssnistfajen]

That's a fair point, but AI tools could also unlock higher potential for them to take on more workloads. Perhaps by maintaining the same number of paralegals or even hiring more, they could take on an exponentially higher number of cases than before. There's infinite room for new litigation as long as human society persists.

[xen2xen1]

Well, then either it will empower the "little guy" to do more (more your example) , or empower big business to capture more big business (more my example). Wanna guess which one I'm betting on?

[worrycue]

> Now if we could get them not to lie?

This reminds me. Whatever happened to IBM’s Watson? It’s not a LLM right? But it seem similar enough function wise.

[droopyEyelids]

Watson wasn't just one thing, it was a marketing term to combine all IBM's efforts to do early AI like efforts. They didn't have any real success in making a product out of anything and stopped marketing 'Watson'

[SoftTalker]

Do we even know why AIs "hallucinate"? Is it possible to prevent it?

[qayxc]

> Do we even know why AIs "hallucinate"?

It's because, just like human memory, they aren't databases or search engines. Generative transformer models are basically next-word-prediction machines on steroids. They take the input and try to "guess" the most likely reply based on their training data.

These machines have no way to distinguish facts from fiction, only probabilities of combinations of words that would make the most plausible reply.

> Is it possible to prevent it?

There are methods to prevent this by incorporating specialised knowledge databases into the training material of these models. This, however, only works with models that have been finetuned on very specific tasks and topics [1].

Other approaches use AI to transform human questions ("bag of words" inputs) to queries into structured knowledge bases, match the results (e.g. tree-like structures of context and facts) to the question and turn them back into human language [2]. The downside of these methods is that they're currently limited to simple QA formats and won't feel as "natural" as talking to chatbot and requires carefully prepared and curated knowledge databases.

[1] http://jens-lehmann.org/files/2019/iswc_bert_simple_question...

[2] https://arxiv.org/abs/2303.13284

[droopyEyelids]

I think it's because AIs aren't actually in touch with reality as we know it, they're generating text from a sort of graph of concepts that comes out of the training process.