I know someone who works in law and deals particularly with an area of US benefits and healthcare law. One of their workflows for lower-level employees at their firm involves taking in documents from healthcare plans and organizations, analyzing them for certain kinds of data, and then importing that data into an internal system they use to analyze and provide guidance on plans. The internal system can contain hundreds of documents for an individual client. All of the documents have the same information (roughly) but in totally diverse formats and styles. Once it's in the system, it's easy to compare and analyze across documents and the research process is much faster.
They recently bought a Claude subscription and began using Claude to do the initial read of the documents and output JSON they can import into their internal systems. The work still must be reviewed by an attorney - Claude is nowhere near making the kinds of judgments a lawyer would make about this content - but it has increased their throughput from 2-3 documents an hour to 8-10 documents an hour by killing the busy work.
LLMs have great advantages for this kind of work - but not for decision-making. I just don't see OpenAI ever admitting that.
(I've left some details intentionally vague because this is a very specific area of law and I don't want my friends to be identified without their consent.)
I just realized how refreshing it is to read an honest take like "from 2-3 documents an hour to 8-10 documents an hour" instead of "it's doing the work of a month in 5 minutes!!!!1".
People will eventually realize that this is exactly what AI does for most workflows. The propaganda is "synthetic humans", "obsolete mathematicians" and so forth, but what we will likely see is a fantastic time saver for things we didn't like to do to begin with.
Sure, nobody knows the exact future. But you still plan for what's probably going to happen. I don't know if social security is going to be around when I retire, and all the experts are saying it's probably not going to be. So, I'm planning a retirement without. Similarly, all the experts are saying AI (<- that word is not "LLM") is probably going to improve. I personally think it's irrational to ignore experts.
I think that a person (me) who has worked directly with frontier AI for over a year now for many hours a day may have gained some expertise in what this technology is capable of and thus what is dangerous about it, but I wouldn't speculate too hard on what superintelligent AI would or could do because we have not yet seen such a thing nor is there even an established criteria as to what that would look like or how it would be valued and in what capacity (hint: humans are still the final evaluators of what is valuable)
The simple fact that I have to write every control scheme in the book to keep the thing on the goal path (and not lapsing into laziness or dishonesty) is itself possibly a safety feature, if you look at it that way
> The simple fact that I have to write every control scheme ...
Look three years ago for a reality check of the progress. The LLM you're using now isn't the LLM you'll be using in 5 years. It probably won't even really be the same architecture, just like those from 3 years ago. What you're doing was literal science fiction not long ago.
People with experience in every other field have a better handle on what's coming than others, and yet experience with AI is apparently the sole exception to you LOL
See e.g. <a href="https://scholarlycommons.law.northwestern.edu/facultyworkingpapers/144/" rel="nofollow">https://scholarlycommons.law.northwestern.edu/facultyworking...
ivraatiems · · focus · HN ↗
They recently bought a Claude subscription and began using Claude to do the initial read of the documents and output JSON they can import into their internal systems. The work still must be reviewed by an attorney - Claude is nowhere near making the kinds of judgments a lawyer would make about this content - but it has increased their throughput from 2-3 documents an hour to 8-10 documents an hour by killing the busy work.
LLMs have great advantages for this kind of work - but not for decision-making. I just don't see OpenAI ever admitting that.
(I've left some details intentionally vague because this is a very specific area of law and I don't want my friends to be identified without their consent.)
ManuelKiessling · · focus · HN ↗
glimshe · · focus · HN ↗
meowface · · focus · HN ↗
pmarreck · · focus · HN ↗
nomel · · focus · HN ↗
pmarreck · · focus · HN ↗
The simple fact that I have to write every control scheme in the book to keep the thing on the goal path (and not lapsing into laziness or dishonesty) is itself possibly a safety feature, if you look at it that way
nomel · · focus · HN ↗
Look three years ago for a reality check of the progress. The LLM you're using now isn't the LLM you'll be using in 5 years. It probably won't even really be the same architecture, just like those from 3 years ago. What you're doing was literal science fiction not long ago.
aswegs8 · · focus · HN ↗
pmarreck · · focus · HN ↗
aswegs8 · · focus · HN ↗
See e.g. <a href="https://scholarlycommons.law.northwestern.edu/facultyworkingpapers/144/" rel="nofollow">https://scholarlycommons.law.northwestern.edu/facultyworking...