Hacker News Reader

Comment by brookst | Hacker News Reader

> any sort of LLM writing is a tacit acknowledgement that people don’t really care what they are writing

Couldn’t be further from my experience. I work with a distributed team that has four different native languages, with substantial engineering staff for each.

LLMs have made our cross-geo communications so much better. I can get a summary of a standup that happened overnight (to me) and in a foreign language (to me) when I get up in the morning. I can write idiomatically and naturally without worrying it will be confusing or misinterpreted.

Plenty of people are not naturally talented writers, especially in your language of choice. LLM writing doesn’t mean they don’t care, it means they DO care about communicating clearly.

View on HN

> I can write idiomatically and naturally without worrying it will be confusing or misinterpreted.

Note that this depends on the language. LLM translations aren't that much more accurate than predecessors other than that it's more strongly constrained to grammatical English, and there's limit to how natural one could be through translations, depending on how much cultural co-development your own background had with the target environment - your framing and storytelling will sound off even through human translators if the cultures are too far apart.

LLMs tends to be a bit more robust than previous MTs, that's for sure.

How accurate are the summaries?

This is what worries me. I'd say that >50% of the time, Copilot-generated summaries of meetings in which I was presenting misinterpret what I said in some important way.

I'd much rather (and do) take a couple minutes to write my own take-aways. It's the golden rule in a business context: I'd prefer to receive three one-sentence bullet points that are actually accurate over a couple pages of AI slop, so that's what I give to my colleagues.

The cross-language factor is an interesting angle I haven't had to contend with, though.

I need to do that kind of thing all the time—-and it annoys me to no end when people post the summaries in chats to “catch up” on a meeting, because I know they’re wrong. As an European who understands five languages well and can take some solid hints in many others, nothing beats actually listening or scanning a transcript.

They are quite accurate but the problem is that they sometimes leave out a critical piece of information said in the meeting that doesn't have a lot of repetition but that turns out to be like a crucial factor in the meeting and the LLMs completely miss that.

I turned on Apple Intelligence to summarize notifications from Outlook, Teams, etc. I haven't found it to be very accurate yet, especially for MS Teams.

Based on my experience, it's very risky. A lot of times, it connects similar but very different things that should be seperate and changes the meaning of the message/discussion

Not OP, but I'll answer from my experience trying several different tools for this: the good ones are roughly as accurate as having a human note taker, familiar with the domain terminology, would be on average.

I’ve found some minor inaccuracies, but nothing catastrophic enough to make me think I’d be better off without them.

I like that--generally I agree that LLMs inflating text is not adding value, but it is a great point that LLMs can help bridge language barriers.

Do you use an specific tools to streamline the translation or format of your communication? I'm facing the same problem, but wonder how well an LLM would be at translating / understanding the domain specific terminology that we use.

Reply on news.ycombinator.com