AI for developers and Google Workspace
blog.google
blog.google
If you really can't be bothered to write the welcome email maybe just don't, rather than throwing some ML generated drivel as the first impression a new hire gets of your management. The direction this is going in is so focused on automating away the hassle of email, with the end state surely being that both sides of an email exchange are ML models churning away at each other.
Workspace saves you the time and effort involved in writing that first welcome reply email.
Use AI to solve problems caused by AI.
It's just more efficient now to grab the same starter email that you then customize. Admittedly, google shouldn't be pretending otherwise, either. But what they also did was make it so you can get the starter email without having to see their ads.
Edited to add: I think a better argument on this could he made about the "draft a thank you note to the team" example they give in the video. That's more the type of thing that should be personal and probably not from a starter template, googled or AI-generated. https://www.youtube.com/watch?v=6DaJVZBXETE But even here, the reality is people were googling for these starter emails already.
So, the best ones will still write their own and the others will save time.
And by definition, most people aren’t the best.
So, we all agree the automation is welcome?
Can we skip this phase of technology?
Edit: Maybe this will inspire the next wave of tech burnouts turned mushroom farmers
If you're growing so fast you're sending hundreds of emails in a short time span, then maybe you should replace a pseudo-personal ai-written "welcome email" for new hires with something that scales better and feels personal.
Besides that, it's not about how rapid you need to send these emails out. It's about how often it's done. If you had to write birthday cards to people you barely know every day, maybe it's only one a day, it's going to get rinse and repeat very quickly. All the ways of saying "have a great day" are going to get exhausted and whoever is writing it is going to feel insincere in a short amount of time. So why not just have the AI write this sort of thing?
Slack bots welcome you to slack channels, when your email is set up there is a default welcome message. No one really cares, they just want to feel welcomed by something to break the ice. I don't think anyone really gives a stuff about how the message is written as long as it's honest. If I joined a company and the welcome email was signed off with "You friendly welcome bot", I'd be happier than some sort of boilerplate cut and paste message signed by a real person pretending they aren't writing the same thing for the 5th year in their job.
Person A: write a thank you note to the team
AI: [long text]
Person B: summarize the text
AI: Person A says thank you.
I would honestly prefer blunt, straightforward, de-GPTified communication style. I hate the idea of wasting my time on reading substance-free filler, regardless of whether it was written by an AI or not.
Basically endource the AI and be clear that it is the thing welcoming you. If you hide the fact an AI is pumping out an automated email it just comes across as if the business thinks of you as a drone.
This is why when a cashier at the supermarket tells me that instead of waiting in line I can use the self check-out lanes, I tell her "I don't want to help a robot take away your job."
Usually they don't understand. But sometimes you can see them start to think about what's happening.
In the local grocery store at rush hour, cashier lines are long and self-checkout tends to be quicker for buying 3 items. Taking a whole cart into self-checkout is slower than a cashier, because the self-checkout has like a 5% chance per item of going into "assistance needed" mode.
I just want to pay for my items and get out of the store, in whatever way works.
I find places with scan as you shop apps way better than waiting in a checkout line. I get a running total of what I have in the cart, and when I'm done, I just leave without a line.
As pxl97 said, I preferred being able to scan things as I shopped and not have to stop at the checkout except to pay, but the one place around here that had that discontinued it a few months ago.
We need to push back hard against the idea that automation is "stealing" people's jobs; the promise of automation has always been that of more money for less work. We need to remind everyone of that, and reverse the trend of automation simply putting more and more money in the pockets of the already-wealthy at everyone else's expense.
get those good old shopkeepers to do it while you stand with your list out front.
Of course, this gets into UBI territory and making sure that as a society we are not just replacing menial jobs to then abandon the people who would have had those jobs.
Managers are supposed to work on the system, at least in theory. 1on1’s are for the manager like server metrics are for the SRE. An early warning system of future problems. Ideally you can fix issues before the problem blows up into an alarm.
I'm not sure that is accurate. As a manager, I have a bunch of production metrics on my people. I can tell who's getting work done, who isn't, who has high defect rates in their work, who makes fewer defects, who is positive ROI, and who is not paying their keep. Where the one on ones are useful is when they are informed by the metrics, and those conversations don't take anywhere near one hour every two weeks. They take 10 minutes - 15 minutes, max.
Sometimes you gotta kick people to go take a vacation lest their mind/body takes an unplanned vacation for them. Even if the raw numbers are looking great. You’d prefer to have a planned outage wouldn’t you?
And hey, if you don’t need a full hour, it’s okay to make 1on1s shorter! Few people will complain about getting more time back in their lives.
But be warned: The good stuff usually comes out 5min before the end of a meeting when an employee says ”Oh yeah one more thing that’s been on my mind …”. It takes time to relax into that area of concern that’s a little scary to bring up.
No matter how explicitly you state it, ask for it, overcommunicate that you're approachable even if seemingly busy... people are averse to the impression that they're interrupting their management even when they have something important to discuss. When you create the space and have even that quick 10 to 15 minute chat every couple of weeks, you're creating space for your people to have your attention without worrying about getting your attention. And people often take advantage of that path of least resistance.
So many of the most important conversations to have with the team are less about their output and more about things they've got going on or need to know _despite_ their output. All the standard sprint metrics are just one piece of the puzzle.
"so, last week you had a -23.5% contribution... You usually are on the positive side, so what's up?" and the employee tells me that their lead stuck them on a project where they hand no experience, and didn't have time to learn how to do x or y before getting thrown at a ticket... or, "I'm having some issues with <personal>..."
Stuff as a manager you can sometimes help a lot with... and other times, well, you can be supportive. Just because numbers are involved does not mean we have to be cold and heartless... At the same time, I've found managing without numbers to just lead to cliqe-y political culture where people's ability to schmooze is rewarded more than their contribution to the team and that drives away the most effective contributors.
> And hey, if you don’t need a full hour, it’s okay to make 1on1s shorter!
Agree... I think I just realized that the exception should be a long 1 on 1. Just because I schedule 15 minutes doesn't preclude taking more time if there's something in the 1 on 1 that should take longer.
I've been doing engineering management for ten years and I've never seen a set of metrics that can tell you these things.
What software engineering metrics tell you who has positive ROI?
Or are they working in an environment so steeped in insincerity that they don't even know what being genuine is, or that one of the most important things for a manager to do with things like a "welcome" is to be setting the tone.
Nothing says welcome, and bring your genuine self, and trust colleagues, and feel intellectual safety, and invest in being a part of the team that's investing in you... than a new hire's manager can't even be bothered to say welcome themselves, and offloads it to a chatbot button instead.
OTOH, this fits the sociopathic culture you'd expect, when you look at their hazing rituals, compensation levels, and promotion criteria. It just doesn't fit the culture some of us techies envisioned when that company was started.
But for the less experienced, AI is a starting point. And depending on the kind of person you are, the kind of upbringing you have, and the kind of education you got, you will be motivated to edit and improve what the AI gives you.
The other thing I think this will help improve upon are those idiots and scumbags who give commands that are not very well thought out and then yell at talented people for not following instruction because they filled in the blanks with their own ideas. Giving it to an AI will show ridiculous their commands were. (sorry, end rant)
One can just hope that the ML entities won't start insulting each other. :)
They would essentially tell me some facts that I would then re-write for them, so why not just use their own words.
If they want to use generative AI, that’s on them.
Humans are amazing in breaking the system for convenience and profit.
The end-state is an ML model running on a client device in isolation. Why bother using SMTP?
If the content of your email could have been generated using an AI trained on publicly available information, it necessarily must not include anything private.
Which means you are spam messaging your new hires with messages that contain zero actual information!
Noise is a word that does not even properly capture the uselessness of such communications. I swear, many humans seem to just want filler text to put places, and they just keep shoving it there.
Anyone read blindsight by Peter Watts? Spoiler alert: an entire science fiction book written around the fact that sending useless information to hyperintelligent beings could be interpreted as an intentional distraction and therefore a threat.
You use ai as a tool, you prompt it, change its results and feed it back.
A little work gives you impressive results way faster than if you wrote the whole thing yourself
This is fine if you never want to ever write something yourself, ever again.
I'd much rather retain that skill.
In a similar manner, using AI tools like GPT-3 or GPT-4 can be likened to having an AI editor that assists you in the writing process. By working together with the AI, you can maintain your writing skills while also benefiting from the AI's ability to generate content quickly and suggest new ideas.
Just as a writer doesn't lose their skill by working with a human editor, collaborating with an AI editor can enhance your writing without diminishing your abilities. The key is to find the right balance between using AI for assistance and retaining your personal writing skills, just as a writer and editor strike a balance in their collaboration at a newspaper.
This comment was written in collaboration with ChatGPT.
For example, I typed this message. But I'm sure with careful massaging, I could get an AI to write the same thing, word-for-word. The laziness you're referring to probably happens even if the manager is typing the email out themselves.
not everyone has good english and corporate writing skills.
I can’t help but think that an example like this shows how far Google has become like Microsoft in the 2000s. A bunch of rich, middle aged managers that have spent too long in the alphabet soup.
We're going to have AI help us generate long pieces of text from very small blurbs of information.
Then the reader is going to have AI read the long pieces of text, and condense it down into just small blurbs of information.
Will my self-driving car also take unneeded detours through pretty places, making it take longer to get where I'm going?
Self driving cars might have set destinations that are cheaper to visit, or have free rides that are ad-supported, or recommend stops along your route.
I think you mean set destinations based on an auction of what business paid most for your visit, based on your demographics & ride history.
heck I might like that service, if optional
I guess that depends if it's _your_ self-driving car or Google/Uber/Teslas self-driving car that they're letting you use. It might even be so that the physical car is yours, but the self-driving software is licensed and controlled by a third party.
Just look at the "I'm on it" example from their announcement video and what nonsense their AI makes out of it: https://youtu.be/6DaJVZBXETE?t=31
It reminded me of an old article [1] about Howard Schultz where his reply of "On it." was praised by another CEO because it demonstrated that he was much more focused on getting shit done rather than overcommunicating or over-delegating.
I just dug it up:
> Friends and colleagues agree that he is as fanatical as ever about Starbucks. Millard Drexler, the chief executive of J. Crew, recently e-mailed Mr. Schultz to complain that the coffee lids at a Starbucks on Astor Place in Manhattan kept spilling coffee on his shirt. Mr. Schultz’s reply: “On it.”
> Mr. Drexler, who has a habit of e-mailing C.E.O.’s with complaints, says: “I can give you many more examples when they say, ‘I’ll send this to a research department or a gatekeeper.’ ” But, he says of Starbucks, “to have that kind of quality control they have around the world is pretty extraordinary.”
[1] https://www.nytimes.com/2011/03/13/business/13coffee.html
No, but it may take a detour through the local McDonalds drive through because they sponsor the self driving car company.
So it's
* small explanation
* initial proposal of what you probably want
* fixes and approval
Not really any different than asking a person to do it for you.
You don't tell your car/mapping software the exact route you want, you tell it where you want to go and then review the proposed plan, and that's with a very restricted setup (exact destination, limited routes).
Edit
More importantly, short messages that convey everything you want with the right tone are hard. Taking my mashed thoughts and converting them to easier to understand prose is a benefit even if it's longer.
Just like a larger function can be better than a code-golfed one. It's longer but it doesn't take longer to read.
I hope so, when you want it to. "Unneeded detours through pretty places" is a kind of being present in the world.
An AI that directs you to explore places it deduces that you'd enjoy would be pretty cool.
Not sure though if it's mostly stuff that's being beta tested as we speak, or stuff that's not even built at all as of yet.
I don't want to think too hard about the alternative, of large groups of people not even bothering to ever try to learn to write. Probably become better megacorp output consumers.
Another part shows the AI creating personalized marketing messages. For a brief time they will work, and then every company will be producing them by the millions, and then customers will stop reading emails at all -- again, the AIs will be writing messages consumed by other AIs.
The generated 'thank you for your leadership' message was amusing. If it's all fake, then what's the point?
I can't see how AI isn't going to make the problem of falling social interaction 1000x worse.
You can still walk down to the creamery and drink a milkshake. There are plenty of flesh and blood people there to talk to. The birds are chirping and it's a sunny day.
There are some bad/weird stuff happening on parts of social media, but you can literally just turn off your phone and go outside.
AI is just another step in the arms race.
Not mocking. This is a really hard space. You can start a whole company and build a product and overnight goog/msft/openai can tweak an existing product or API to make you obsolete. I'm sure some folks will figure out moats eventually... Always rooting for the upstarts over the incumbents.
Seems like the opportunity is applying LLMs in the background to accomplish previously difficult tasks, as opposed to just building text generator tools.
The acceptance that you've been doing corporate busywork as your career seems at odds with your confidence that you'll be the one to outsource that busywork to an AI. Doesn't the rapidly increasing capability of automated systems to pull off this busywork imply that careers built on doing that busywork are increasingly at risk? Am I missing something?
Most of us techies, at some point in our career, have managed to automate some boring or tedious part of our job. We don't tend to get let go for doing this, it just frees us up to do other, possibly higher-value things. Or, if you are less lucky, your employer just comes up with other not-yet-automated tedious things for you to do to demonstrate toil.
Our days are all filled with a combination of useless, low-value tasks and useful, high-value tasks. I'm personally a believer of finding the low-value tasks and automating them, so that you can spend more of your time on the high-value ones. I've worked with people who have the opposite view: Spend your time visibly toiling on whatever is easiest, because most employers can't tell the difference between spinning your wheels being busy and providing actual value.
"Available for select developers. Stay tuned for the waitlist soon"
When the order to pre-empt comes too quick to get your waitlist landing page approved.
I hope we realize how insanely stupid this is soon.
I started using Brave.
you've actually switched browsers to have access to it? why???
But yeah, Bing Chat is only available in Edge.
Also, Edge on Android allows me to block ads.
Now Microsoft has all of your data, is it actually any better?
> I dislike one company having control over web.
Edge is also based on Chromium, developed and controlled by Google
> Edge on Android allows me to block ads.
You could try Brave (also Chromium-based) or Firefox (Gecko-based) with uBlock Origin, both will block ads and tracking
That verbose email is going to end up a weird machine interchange format (which only looks like formal writing for human consumption), just to send a message like "I'm on it."
Excellent!
goes through the terrible UI of google cloud console
Nothing there. I see it's not actually today it's a sometime soon with some people in a private preview. Any info on what model size they're actually releasing?
Time to try the demo they showed in google docs.
> Now, we’re excited to take the next step and bring a limited set of trusted testers
Oh.
It's good that they're actually planning to ship stuff, but when? I've seen private previews sit around certainly for 6+ months, probably a lot longer.
They do this all the time launching something in the US but not mentioning the limits leaving me struggling to get something to work.
It doesn't help that their approach to ai is often "don't worry a magic thing will tell when you want it and will make it appear" and I don't know if I'm not making the right incantations or it's broken or it's not available.
I wish these would all come with a "notify me when available" button at the very top.
They're going to open up PaLM & an image generation model (imagen?) through Google Cloud. It looks incredibly easy to finetune PaLM in their environment:
Video (Build, tune, and deploy foundation models with Vertex AI): https://www.youtube.com/watch?v=yg2yHIKQ7oM
My understanding is that Google has always been ahead (at least until recently) - but the difference is that one of their competitors ships.
OpenAI has a bias to action. DeepMind has a bias to grand-challenge stunts. Google has a bias to perpetual validation.
> Earlier today, we announced the PaLM API, a new developer offering that makes it easy and safe to experiment with Google’s large language models. Alongside the API, we’re releasing MakerSuite, a tool that lets developers start prototyping quickly and easily. We’ll be making these tools available to select developers through a Private Preview, and stay tuned for our waitlist soon.
Overview: "Summarize my tech feed"
In an informative article: "Didn't I Star something similar last week?"
In a news article: "What does website X think about this event?"
The future is bright!
Nature goes the path of least resistance, and nowhere is this more obvious than in human interactions. If your signal consists of three words, write three words instead of generating blobs of decorative noise masquerading as content, that serves no purpose but to be decoded by the same ai, but client-side.
Compare with Microsoft for Startups, which isn't seen as hip and edgy but can get you like $5k+$25k in Azure credits (and some OpenAI credits) without having to go the VC route.
Seems more entrepreneurial to me. Venture-backed is just not the right path for everyone.
I'm like you, bootstrapper. Went with GCP for 4 yrs until they shutdown my production servers last fri.
Crazy story. Thanks for sharing and yes, suffice to say, this informs our tech stack choices, hah.
If that's the pace at which they're planning to compete with Microsoft/OpenAI, they already lost.
Instead i want
- Intuitive interfaces to respond to the email with a few clicks and a short laconic email (I don't want to waste other people's time)
- Specialized text generators for legal purposes, which create provably verifiable and consistent documents
- Expert systems that doctors can use and trust
[AI: Expand this to a 70-page diatribe]
AI assistants that summarize and enrich information for you, fine. That actually saves time. But this feeding of generative output to humans is way off the mark. It feels like the product people are not sitting down to think what they're doing and just riding the GPT hype wave.
HR/PR approved corporate communications are already crappy. From a practical viewpoint, I don’t want to wade through pages of bullshit boilerplate and illogical images. The future is bleak and boring…
https://arstechnica.com/gadgets/2023/03/google-dusts-off-the...
Now every Google team needs to shoehorn "AI" into its products, no matter if it makes sense or not.
“So if you’re a manager onboarding a new employee, Workspace saves you the time and effort involved in writing that first welcome email.”
Some great announcements, could have given more pertinent examples. You save your time elsewhere using AI tools, so you spend time with your valuable newcomer.
So you are getting a gelding. I really miss the old SV - make it fun and powerful.
Usually the most interesting things come out of unsanctioned uses of technology.
It's intentional.
Again, this maybe bad timing since GPT-4 will (probably) gate crash Google's announcements this week, hence why Google has done this today.
But for now, AI copyrighters are going to be destroyed, quickly by the same tech giants: Microsoft (OpenAI), Google, and possibly Meta.
This is a very grey area. A prompt, for example, could be copyright. An image you create by drawing a base layer and having the AI fill the details in could be copyright as well.
The government copyright office doesn't really decide these things. Courts do.
- “easy and safe”
- “all supported by robust safety tools”
when the only things users really care about are quality, latency, and cost. Not sure where their head is at but unless they are oriented around these three things they’re never going to catch up.
But ChatGPT is doing exactly the same. There's more energy spent trying to manipulate the output to be politically correct than actual energy spent making the AI more powerful.
If they had just put up the same model without that attempt to make GPT wear a politically-correct mask, they'd just get another Microsoft Tay event, and only get a buttload of negative publicity for their investment.
For an internal application used for specific practical tasks, sure, focus on making it more powerful; but if you want a demo for general public as a 'pre-sales advertisement'? You bet that political correctness has to be there.