I don't use ChatGPT for those, but it is the epitome of what you are describing, people will take a single sentence, ask some LLM to blow it up the correct length and in the process make it a complete waste of time for everyone.

My guess is that with long-form text losing value due to LLMs, we will see a return of very succint 1-2 lines employee feedback.

This would be a great outcome in a lot of areas!

Having been on the receiving end of many of these, it absolutely is pure BS and I lose all respect for anyone who themselves have so little respect for their colleague's time as to subject them to the AI-written slop instead of actual genuine feedback.

The whole fucking point is to give them actionable feedback, both good and bad, for them to work on themselves, not some generic hallucinated summary of some bullet points you haphazardly threw together. I can copy/paste the review prompt into ChatGPT myself, thank you very much, I don't need you to do it for me and to pass it off as your own genuine thoughts.

As for the commenter you are replying to, you dont have any specific information on the review. Yet you declared its hallucinated, generic, haphazardly thrown together, simply copy pasted, etc. Consider that your conclusions are based off an idea in your head and not his actual review.

Be better. Someone respected your opinion enough to go out and ask for it. Take a minute to reflect.

But as the other comments mention it might just all be bullshit anyway.

I always put a lot of time into reviews before. Should I not use the tool to make something even better (within realistic time commitment)?

If I use an AI to create some cartoon graphics for a slide, should I have bettered myself by learning graphic design instead?

Slackers tend to repeat the same thing over and over in progress meetings... 'I am blocked because... <insert external cause>' or 'I helped that guy figuring out... they did not have a clue'

Vs the more curious, get it done attitude: 'I tried this and that and it still doesn't work, but I learned that... hence I will aproach it from following angle... '

Peer feedback is valuable, it is important, and it is expected of senior team members. I guess it is a fun game if you have a poor manager, but the whole argument around the strategy isn't even internally logical, and if you play it out it is a poor outcome:

1. Company expects Senior Engineers to provide input on their teammates. 2. Senior Engineer has a Bad Manager, and decides to intentionally withhold feedback because "that is their manager's job, and up the chain" 3. Senior Manager (skip-level of Senior Engineer) determines that their Manager is a Bad Manager, and replaces them. 4. Good Manager joins, determines who is performing, and asks "Why did none of the Senior Engineers identify this earlier?" 5. The supposedly competent, but intentionally malicious Senior Engineer in this hypothetical is (correctly) deemed either incompetent or not believable by the New Good Manager. 6. Good Manager finds a component Senior Engineer with any sense of character to replace them.

This post is such hogwash, it is so fully of toxicity, and it is a dumb strategy. When things "hit the fan" this person is being tossed out in the regime change as well.

I would truly hate my life if I worked with people even a fraction as toxic as this.

It's more complicated than this.

The short form isnt actually the best form. It's incomplete. The LLM is being used to decompress, because it can be difficult to do. Blindly using an LLM isn't the solution but it can be part of an effective workflow to write good feedback.

Also, I'm sure some people take a brief, complete idea and expand it into an entire paragraph because they have some warped perception. That's bad, but I dont think most people are doing that because most people dont see any reason to.

There's a cartoon going around where in the first frame, one character points to their screen and says to another: "AI turns this single bullet point list into a long email I can pretend I wrote".

And in the other frame, there are two different characters, one of them presumably the receiver of the email sent in the first frame, who says to their colleague: "AI makes a single bullet point out of this long email I can pretend I read".

https://marketoonist.com/2023/03/ai-written-ai-read.html

The biggest productivity boost I ever managed was using Whisper to convert meetings to text and then a big model to summarize what happened.

Then I can chat with the docs and meetings about who decided what, when, and why. It's a superpower that I could only implement because I'm in the C-suite and could tell everyone else to get bent if they didn't like it—and gave babysitters to the rest of the C-suite.

Having visibility and ownership for decisions is a huge deal when everyone has access to it.

> gave babysitters to the rest of the C-suite

What does that mean? That they got help, if they found the tech too complicated?

> Having visibility and ownership for decisions is a huge deal

Has this changed how people behave (yet)?

And baby sit means hire something between a secretary developer that makes sure that important meetings had the record bot invited, gave it a once over and then went back to the 70% of their job that was actual development.

But yes, phone screenshot is another strategy with much less grey area. I'm just becoming more and more paranoid of some potential defense trying to accuse my photo of being doctored, especially with more and more AI tools available.

(We are use zulip for chat which is better than everything else I have used since irc. But the search is too limited for someone who knows regexes.)

You can have an offline copy of emails and you can BCC them to your personal account if you want.

If you are willing to violate that rule or the message affects your work contract which you are of course allowed to archive at least in zulip chat that's very simple (for a software person). They have a straightforward REST API. IIRC you can even choose between markdown source and HTML rendered output.

Many reasons. First, chat doesn't exist. What exists is scores of incompatible chat apps.

I use WhatsApp but I consider WhatsApp messages throwaway because I keep losing them anyway. They are scattered across multiple phones with no way to merge them. Backups are platform specific. Exports don't contain any metadata and can't be imported.

"Chat" is a useless mess, not a paper trail.

For email, I have consistent backups with metadata across many email providers and email clients going back to 2008.

Chat messages tend to exist in one place only (vendors' servers), with maybe a transient local copy that gets wiped over time, or "for privacy reasons" (like Messenger switching to E2EE, effectively wiping cached history on any device that went through the transition). Chat message is an object, it's designed to exist in a single place, and everything else is a pointer to it, or a transient cache.

E-mails, in contrast, are always copied in full. You send an e-mail to me, you retain an independent copy, I get an independent copy, and a bunch of servers in between us keep an independent copy too, even if briefly. I forward your e-mail somewhere, more people and servers get their copies. I reply back to you, more independent copies, that also quote the previous messages, embedding even more copies that are even more independent. This makes it very similar to paper correspondence (particularly when photocopy machines are involved), i.e. impossible for a single party to unilaterally eradicate in practice.

And then chat vendors implement silly features like ability to retroactively unsend a message, force-deleting it from recipients' devices too (it may still exist in backups, but vendors refuse to let you access those, even with a GDPR request). In e-mail land, that's fundamentally not possible.

(Microsoft tried to bolt it onto their corporate e-mail software, but it only works in Outlook/Exchange land, and it's easy to disable (at least was, in OG Desktop Outlook - not the still broken New Outlook Desktop Web App). I discovered this when I once saw an e-mail I was reading suddenly disappear from my Outlook, which prompted me to find the right setting to disable honoring unsend requests.)

So, come discovery time, critical chat history may turn out impossible to find, and any deeper search will require forcing cooperation of the chat operator. E-mails, on the other hand, tend to turn up, because someone, somewhere, almost certainly has a copy.

I've skimmed maybe 50% of them, but not enough to consider them "read". It's 99% bullshit. Even legitimate email is spam these days.

I'm too busy with other fake work to need to additional fake work managing pointless email comms.

At this point I am thinking my Thunderbird should probably just unify the Inbox view and the Task view, since it would be a more accurate representation of how I view email.

For you: important things become tasks, useful things are filed, and everything else gets trashed.

For me: important things get opened and replied to. Useful things are starred (and opened). Everything else stays untouched.

And that pesky unread number is irrelevant because I mute all notifications. I'm not discounting your method, I am just now realizing the circle of it all.

I practice inbox zero also, the value for me is knowing that if it's in my inbox it's because it requires actioning, if it's not it's ignored (deleted or archived).

I also just generally like deleting things as much as possible, I don't like the cruft. If I have to search through old emails I don't have to filter by stars or anything like that, I like knowing that if it exists it's because it's important.

- sending "sounds good" even when the recipient hasn't, in fact, read the initial email => catastrophic alternative

- writing an elaborate email explaining in luxurious details why it in fact sounds good => not catastrophic, but costing time on the other side to read and understand, with zero added value

AI will not replace human thinking, even though many people seem to believe and put their brain on stand-by.

It feels like someone wants to transport water from A to B and transports it as steam, just because _we can _.

after all, what’s the point of a giving someone a bunch of cheap flashy gifts for Christmas (instead of say, socks) either?

As long as we can all pretend they were thoughtful and meaningful, and someone isn’t using AI when making it (or just picking random crap off the shelf, and they removed the price tags) or using AI when reading it (aka making a big show of opening it, and then throwing them in the trash immediately after the person leaves), then we all get along. It even looks like we’re doing a ton of work/spending a ton of money to make the other person happy.

Not that anyone does any of the things I’m describing, just being hypothetical, obviously.

I suspect it will be obvious enough shortly it will go the way of the ‘popcorn bucket’ fad or the like, but for now…

And what "normies" do with that is skimming it, ignoring majority of it and answering random part.

Exactly that. For me, a lot of effort in structuring e-mails goes into making it look like text instead of bullet points, because some stupid social expectations, but then still making it bullet-pointed in nature, because if I don't, the typical normie recipient will do exactly what you said: ignore majority of it and answer random part.

(And then they'll somehow screw it up anyway, and I'll still have to chase them after that one critical question they conveniently forgot to address.)

Making them feel good and "seen", obviously. This is perfectly expressible in "engineer thinking" (I won't say "quantifiable", because there's this meme that engineers see things in binary, whereas the reality is, math is perfectly fine with fuzzy ideas and uncertainty - it's the normies that can't handle those).

Hell, there are some game-theoretic approaches to maximize social ROI on gifts, but I won't go into those, especially that they tend to flip the sign on the return if the recipient learns about them.

/s, of course, but not that unrealistic.

AI or not, it would be better if they just sent their prompt instead.

Because that's literally what it is. Its an algorithm that is continuously asking itself, 'what is the most likely word I should say next?'

Whereas an author that is intending to communicate a point, will start with an idea, write a passage to explain the idea, and then edit their passage to the minimum number of words that most precisely, accurately, and succinctly communicates that idea.

It's the modern equivalent of LMGTFY. The OP could just as easily written the same prompt themselves. The difference is that LMGTFY was an expression of irritation, smugness and hazing. The ChatGPT reply is just garrulous laziness. I expect and hope we'll develop social rules that mean this type of reply will be seen as passe.

1: people will use ChatGPT to write their formal emails based on a casually written text 2: people will use ChatGPT to convert their emails from formal text to summaries\ 3: this will get automated by email providers 4: eventually the automation will be removed and we'll just talk in plain language again

Same when summarizing, just less frequently.

As someone who cares about precision and clarity in my writing, I do not use LLMs in the context of communication.

They specifically allow you a grammar/spell check and also change tone (formal/informal) and length. Length one I have never used but the grammar spell check is a godsend that I use almost always.

It's also really useful to for words that are spelt almost the same. Suit and Suite for example.

Also throughout my day, I'm constantly switching between 2 languages that have almost identically written words. Adress and Address. The normal spellchecks often don't mark it as an error because my computers and browsers naturally also have 2 installed keyboards and languages.

I worked with groups of tradespeople who had poor literacy and they had to write emails and some of them were very poorly written. AI would have helped these people a great deal in providing information but also being able to understand what was coming back to them.

> There's a cartoon going around (...)

Both frames of the cartoon represent a real perceived need: for the sender, the need to inflate the message to "look nice" because "people expect it", and then for the recipient, the need to summarize the nice-looking message to get the actual point they care about.

Hopefully the use of AI in email will make that cartoon (and the underlying message) widespread, and lead to people finally realizing what they failed to realize all these decades: just send the goddamn bullet point. We don't need AI in e-mails. We just need to stop wasting each other's time.

EDIT: and riffing off rpigab's comment downthread, https://news.ycombinator.com/item?id=42723756 -- I wish for the future people will feel comfortable, instead of sending AI-generated e-mails, to send the goddamn prompt instead. It carries all the information and much less noise.

I mean, using LLMs makes sense if you actually need to communicate in prose - for many, myself included, it's much easier to evaluate whether some text sounds right, than to write it that way in the first place, so LLMs are useful in evolving and refactoring your own writing (and learning how to write better from it, over time). But that is rarely the case in transactional or business communication - for that, just send the prompt.

Alice: Hey, Bob, I finished the job, pay me

Letter: Blah blah blah, Bob, blah blah blah, $$$, blah blah blah

Bob: Oh, Alice is done, hey Charlie, pay her

Letter: Blah blah blah, Charlie, blah blah blah, Alice, blah, $$$, blah blah

Charlie: Ok, Alice is paid

Letter: Blah blah, Alice, blah blah, $$$, blah blah, bank account, blah

Alice: kthx

Letter: Blah blah blah...

Sometimes I would spend 15 minutes writing a 3 or 4-line email of this kind. Not anymore.

These are human problems desperate for magical ways to do less work.

Turns out it doesn't work at all. It gave me a random selection of rides, was missing info in some of them, and worst didn't realize it was giving me bad info. Pretty disappointing.

I mean if I'm going to proof-read the full task output from the AI, I might as well do the task by hand... but proof-reading a script is much quicker and easier.

Even when meetings are summed up, which I think they should be, you frequently see that no real progress was made, someone did all the work before the meeting started and this is now just a one hour sign off, or everything is simply pushed to the next meeting.

If my experience with Microsoft Office Copilot is any indication, these features produce very confusing, low-quality content if they are not completely wrong and useless. Used it once and never touched them again. (My company is still paying for this and rolling this out widely despite many reports of how unhelpful they are.) I doubt Google Workspace can do any better.

At the same time, I tried the Gemini Research feature last night, via the Gemini webapp, and was resoundingly impressed. From a vague description, it can find the open source project I was looking for, provided ample links, and a pretty good summary of the project.

deets: https://news.ycombinator.com/item?id=42706997

There is no value for a bloated autocompletion tool.

There is value for concise drafts.

I wish Google would cut the PMs and bean counters, ressurect some of their better projects, and trim their fat instead of cut their sinews.

Note I do use ChatGPT pretty frequently, but I've found it much more useful to have a separate space for the kinds of conversations I have with ChatGPT.

In this week so far (first week back from Christmas / New Year leave) I've spent maybe half a day total on work that could be classified as "progress". The rest of the time has been meetings and the required meeting follow-up work.

There's no point in Sprint Planning or considering adding priorities to the current plate. It's full. But nobody has time to eat things off the plate because we're always in meetings to work out how we can eat off the plate more efficiently.

/rant

I've come back from holidays angry. Things gotta change.

Equally, instead of talking about meetings as detracting from your work, start talking about them as the work.

When your manager asks about your milestones, or accomplishments, or success stories, make meeting attendance front and center.

When discussing software development, bug fixing, etc in the meetings, point out that you won't actually do any of it. Point out that 20+ hours of your week is in meetings, 10 hours of admin (reading, writing, updating tickets), 5 hours of testing etc.

"This task will take 40 hours. At 1 hour per week I expect to be done in October sometime. If all goes to plan'

Yes, it seems cynical, but actually it has real outcomes. Firstly your "productivity" goes up. (As evidenced by your ticket increase.)

Secondly your mental state improves. By acknowledging (to yourself) that you are fundamentally paid to attend meetings, you can relax in your own productivity.

Thirdly by making your time allocations obvious to your manager, you place the burden for action on him.

If you convince your colleagues to do the same, you highlight the root problem, while moving the responsibility to fix it off your plate.

I was just thinking about how for the people requesting all of these meetings, the meetings are the work. If they don’t meet / waste everyone’s time, they are… unproductive.

For engineers, meetings are the non-productive part and are not counted anywhere.

Adding them to Jira and accounting for their cost is the way. Businesses understand money. Meetings are expensive.

Does your company log meetings as tickets?

My experience though is consulting to large organizations. They have lots more people, more layers, and hence need more accountability. I get the need for that, but also see that balance is required. I help both sides understand the requirements of the other party, and help them find balance so that both sides win.

Part of that is helping programmers understand what managers need, and part of that is helping managers understand what programmers need.

Managers, for example, are happy to add everyone to every meeting. Workers usually prefer one on one time.

Equally co-workers often benefit from set-aside time for team meetings. This helps with in-team communication.

Information flow is necessary. Doing it well is better for everyone.

This is a huge problem in all orgs of any size and one I battle with - misaligned incentives.

> For engineers, meetings are the non-productive part and are not counted anywhere.

Part of addressing the issue is to not be binary in your thinking. You'll lose the people you need to persuade. Some meetings are very productive and necessary for engineers. The goal isn't to get rid of all meetings as much as it's to only have productive meetings. When forced to only have productive meetings, fewer meetings naturally result.

(yes this happened to me before)

All the work to actually reach requisite agreement for the decision is done in the days / weeks leading up to the meeting via ad-hoc-ish one-on-one or one-on-very-few meetings (possibly including graft and corruption).

The "decision" meeting isn't organised until the result is known and guaranteed.

This maybe doesn't apply to Agile / Development-related meetings, but I'll keep trying to determine how to make it apply, such is my disdain for this (seemingly) waste of the team's time (he said, whilst posting on HN).

I do not miss development.

Unless you own the shop I don't see the issue - good money for a day's work a week?

Unless you own the shop I don't see the issue - good money for a day's work a week?

Yeah, except I have a visceral feeling of pressure to make progress and I don't want to be "one of those people" who don't work towards some kind of improvement. I had a bit of a rant today, and one of the leaders agreed with basically all of my points, although they said that there's a limited amount that can change in the immediate due to existing priorities. However, I'm still going to dedicate some time every day to map out how to improve on the status quo - this will further inhibit my actual task progress, but in the pursuit of a loftier goal (so, yes, potentially making it worse, but it'll feel like I might make things better...).

Sure, and that's an actionable solution if you can control the actions of everyone else who emails you.

Notably, in my experience there is a high correlation with that background and being curmudgeonly. Mainly because that means someone has been responsible for outcomes, regardless of feelings. And something often has to give, and it’s usually feelings. It’s also hard to not be cranky or even angry if someone has to constantly be the one ‘not having fun’ or cleaning up messes so the whole thing doesn’t fall apart.

There is huge market demand exactly for what you’re complaining about, which is faking things happening as convincingly as possible, precisely because being clear/concise, etc. helps with seeing the root cause of problems, and if someone is worried (or is legitimately) a root cause of the problem, of course they’ll consider that bad.

For example, a good sign of a badly led organization is that it’s always busy, but never seems to get anything done. Everything is an emergency, so nothing really gets fixed, etc.

Or there are constant meetings and emails, but nothing gets decided.

People will pay good money for the right kind of wallpaper that makes that ugly wall look pretty again.

These outputs still require editing for sure, but each one can easily save me half the time to write these things.

I probably spent 20 minutes doing this and got value for my 20 minutes.

I don't see what I get out of 80% of these products. It's just more noise.

These features are just so rudimentary you just know a bunch of MBAs from McKinsey came up with them over a 7 month and $25m

I’m not a native english speaker, but working at US subsidiary I must produces reports in english etc - and having an LLM proofread my texts for me is great.

LLM:s are new modality to computing. If you need it, they are great. But just like excel/sheet have limited applications a LLM with data has limiited use as well.

That worked, but to be honest I have tried similar things more recently that didn’t work. Perhaps there is a routing model up front that decides whether or not to use a lot of compute for any given query?

Google also plans on charging more money for APIs for code completion plugins for IntelliJ IDs, etc. this year.

I would like to see AI pricing models be sustainable, not give things away for free, and have lots of control over when I use a lot of compute. I actually have this right now because I usually use LLM APIs and write my own agents for specific tasks.

I tend to agree, except these two things are kind of the same thing. It can make going through the noise easier by intelligently filtering out the noise or finding you the signal. Search. It doesn't necessarily need to eliminate the noise.

Maybe AI would be better if it prevented the noise, and its definitely going to add noise (expanding a few basic thoughts into an email with lots of fluff), but it can also solve it.

I think this really encapsulates something that I hadn't been able to put my finger on in regards to LLM summarization. What it seems to indicate is that, if you need a computer to summarize a large amount of text that someone has sent to you, there are two likely possibilities:

1) The information is incredibly dense/important/technical/complex. This necessitates the extreme length of the message - (think: technical documents, research papers, a rough draft of a legal notice, or your will.) For these sorts of things, you should not rely on an LLM to summarize it, because it may miss key details of the message.

2) The person sending it to you is bad at communicating, in which case the solution is help them learn better communication, rather than "de-noising" their clumsy wording into something comprehensible.

"But what if its number 2, but it's coming from your boss?"

Then I see two obvious points to consider:

First, you should absolutely be telling them about the problem, regardless of the position that they hold. You can phrase it in a way that isn't rude. "Hey boss, I saw (message) but I'm not 100% the intent. I've actually noticed that with (other time)...I usually try to front-load the action items up front, and put the specifics lower down. Anyway, to make sure I'm tracking, you're talking about (action) on (thing), right?"

Second, until (or unless) their communication style is de-noised, then part of your job is being able to "translate" their instructions. Using an AI to do that for you is a bad idea because, at some point or another they're going to be trying to speak to you in-person, or by phone.

Not having dealt with their mannerisms in an unfiltered way might lead to you being "out of practice" and struggling much harder to figure out what they're trying to convey.