Googe AI IO 2024

Man I find I weird I can’t speak so vaguely conversational and meandering to AI what a waste of time. I couldn’t pair Bluetooth with my headphones despite following the "hold by button for 4 seconds that was in the official and unofficial responses for 3 pages of search results. What if someone hasn’t done the proper work and published it for the AI to steal? The 4th page buried in a reddit comment was some guy discovered you had to hold it much longer than 4 seconds.

This uhh was really lame.

There were some AI tools that seemed useful, like “go through my emails and store receipts in a folder and create a spreadsheet” That seemed really neat, and actually doing tedious work that I wouldn’t want to bother with.

The “tell me on this desk what makes noise” is like… who is this for? What does this accomplish? The whole trick of “AI assistant explains what you are looking at” is so played out at this point. It is one of those features that feels like a party trick, but isn’t actually going to be that useful in day to day.

I loved that their video generation examples were very small on the screen, and also didn’t really match the prompts they were showing that well.

Google’s AI model itself has a ways to go before I would trust it with any important tasks. They showed some features that felt like… yeah, based on what I have seen with your search AI suggestions, I think you are a long ways away from it being accurate enough to be useful.

Like, would you really trust Google’s AI summary of an email chain to pick which bid and contractor you want to go with for an expensive home project? Yeah… gonna want to read through those emails myself.

I’m not impressed by OpenAI’s demo yesterday too:

lol, and if they put that out there, it was probably the best demo they managed to put together.

Their other demo where ChatGPT helped solve a mathematical equation was pretty impressive, I thought.

I have Gemini and I thought it might be able to do useful things. I tried:

“I have a flight to Spokane in June. Can you put the flight times on my Google calendar and set up reminders for the day before?”

Nope. Didn’t even try.

"OK, can you tell me what time my return flight is?

Nope.

“OK, there is one morning non-stop flight on Alaska from GEG to SAN that morning. What time does it leave.”

Here’s a Google Flights result for round-trip flights from GEG to SAN.

“So, that’s a no and you’re useless.”

That’s pretty cool as a tutoring app (that of course you got “for free” as part of model training).

The voice is a little over the top, and would get tiring really quickly.

Yes, it would be cool if it could do those things.

I have still yet to see Google’s offerings be able to do any of the cool stuff they are supposed to be able to do now, which does not give me much confidence in what they showed in I/O.

I just love how much these presentations push the AI assistant describing stuff in front of you. Like… it looks cool, but when you think about it… how is it useful? Most things are designed so you know what they are, if you are traveling, you have looked up the landmarks, and also… doesn’t google Lens already do a lot of this stuff anyway? It is a parlor trick that looks impressive, but when you think about actual use cases… how much will you use it?

I think some of the gmail integration could be good though, anything to better organize the multiple gigabytes of mess my gmail is would be appreciated.

No one’s going to use that, it’s not really meant to be used. It’s a demonstration of multimodal capabilities, that could be leveraged in other practical applications.

And honestly, building up a mental representation of a scene from a live video feed, and then being able to answer questions like, “where were my glasses?” is impressive.

Now, based on prior cases, I’m not taking that demo at face value, but the capability it’s purportedly demonstrating is in fact pretty hard core.

On my face, because I am looking at a screen.

Also, even if it looks impressive, how is that good if it isn’t something that is actually useful?

Show us the “other practical applications” then.

The way to look at it is that “where are my glasses” isn’t a feature demonstration. Its a demonstration of a fundamental requirement of a useful AI, the ability to understand its environment. That understanding enables responses that have the appropriate context.

I feel like maybe you aren’t following what happened in the demo.

Low level technology is often not inherently “useful” itself, but enables useful things.

An internal combustion engine is not useful on its own, but it enables tons of useful applications, like cars.

Having an AI that is able to just watch a live video feed, and understand the feed in a deep manner where it’s building up a model in real time that it’s able to reflect upon could be useful in tons of ways. Are you really not able to imagine how that could be useful?

I guess on some level, it doesn’t really matter, because other folks will be the ones who use such things… but I feel like you’re kind of going out of your way to discount stuff needlessly. It’s not really going to impact the development of any of this stuff, but it seems like you’re kind of screwing yourself over by biasing your perspective so heavily.

Yeah, the “describe what you see” thing isn’t useful, but that sort of capability is what is needed to do what everyone wants eg smart glasses to do, ie parse the environment and provide useful contextual information. It’s necessary but not sufficient.

My feelings exactly.

As I mentioned on the other thread, this felt very much like a keynote intended for suits, rather than showcasing technology. A “please don’t tank our stocks” presentation. I realize Google I/O has been trending in that direction for a while, but it struck me forcefully yesterday.

I’m glad to find out I’m only one weirded out by the interaction this new stuff wants us to jump through to to accomplish less.

Yeah, the demonstration for an engine is to move a car. You know, the thing that it does.

Google did not do a demonstration of that level. This was a tech demo, akin to showing an engine that starts up and runs and moves a piston. Show us the car! Oh wait… maybe not.

Having an AI that is able to just watch a live video feed, and understand the feed in a deep manner where it’s building up a model in real time that it’s able to reflect upon could be useful in tons of ways. Are you really not able to imagine how that could be useful?

Name a way that it is useful in which I don’t have a brain that also saw that stuff too? I can write down notes. Microsoft Hololens had some cool demos of how it could be useful, like overlaying schematics onto things, or keeping a checklist of parts needed in a repair that techs could wear. Show me a demo like that!

Why even go to any PTA meeting when I can have my AI just run down the bullet points of the meeting? I mean, do I really care about my kids? Oh, why even have these PTA meetings when people aren’t showing up to them?

Let’s talk about what would have been a useful demonstration. Show the AI assistant taking “minutes” of the meeting and sending out an auto-generated email to the attendees on the discussion. They showed something similar to this later, but stuff like that really should be the focus. Like I said, I thought that some of the gmail stuff was impressive, better categorization of emails, receipt tracking and cataloging… actually useful.

The stuff Google showed was the baby stuff, which inherently leads me to believe they are a ways away from an actual practical use for the tech. If they can build the engine, it isn’t that useful to google shareholders if they don’t have the car too. This was all smoke to be blown up asses

This was meant to show off tech to investors, and honestly, it wasn’t that impressive at all, and if anything it showed how far off we really are for this stuff to become more useful. We get beta access in 2 months? How much of that beta access is going to be pulling data to better the model? I would think that investors were not all that impressed by this.

A bunch of the stuff they showed was stuff that Google photos or Google Lens can already do. Wow, you can search by video now? I mean, I can just take a photo and search now, does a video really make it much better? The example they showed of the tone-arm swing issue was hilarious, because as a record enjoyer, the answer google spit out was not actually useful to someone who didn’t know what was going on. (She was trying to engage the needle without the player being set to “play” so it was attempting to swing the arm back to the cradle) It was not “unbalanced” and if she went to go try and rebalance the tone arm, it would probably fuck things up.

Show me photos of my child swimming? Google already creates albums automatically for my pets, I guess this is slightly more complicated than that? I mean… google photos is already really useful.

Google, where did I leave my keys? Oh wait, I have a tile device that shows they were left in my car in the parking lot.

Anyway, the entire I/O just didn’t show of much that was exciting, and stuff that we already saw Microsoft and OpenAI promise in their tech.

AI should be able to do the tedious tasks we humans don’t like to do, not make the tasks we have to do more tedious. I don’t think that Google showed much of that in the presentation.

It’s a good thing that humans have perfect recall and memory of everything their eyes saw!

What if you didn’t have to place tracking devices on things, and instead wore AR glasses that observed your environment while you were doing things, and then you could ask it about where your keys were WITHOUT you using additional hardware for them? Or about anything that had happened to you?

I mean, are you really not able to imagine any ways this kind of thing could be useful?

20$ Tile versus $$$? AR glasses?

Isn’t that “additional hardware” as well?

They are showing too many examples of stuff that has already been solved. Big “Silicon Valley” energy, solving things that already have solutions.

I think a big part of the reason some of us are very leery of what is going on now, is the same reason many of us were leery of blockchain some years back.

The technology we’re doing is cool … yeah, so was/is the technology of the blockchain. But - very much like the blockchain - what Google showed yesterday was very much technology in search of a problem to solve.

And I don’t say that because I think the current generation of AI - as a technology - is in any way similar to blockchain. LLMs et al are actually genuinly useful, and have a lot of interesting applications (as we’ve discussed previously in the AI thread). But I suspect (as Doctorow also says in many more words), that the reason we get presentations like these - is that the actual practical uses of current AI technology simply cannot justify the investments that have gone into developing them. And thus you end up with (once again) the “cool technology looking for a money-making problem to solve phenomena”.

A personal gmail assistant that categorizes my emails is cool. Is it so cool that I would pay $20/month for it? Maybe I should just set up some e-mail filters instead. As @JonRowe notes… these things already have solutions. Trying to get people to pay for something they can already do with a little bit more effort for free (or cheaper) is rarely a good business plan.

Yeah, but it’s one piece of additional hardware that can track literally any object you ever encounter, vs. having to attach a hardware tracker to literally every object.

“Who needs GPS, that problem’s already been solved. I got my atlas right here!”

The thing is though, the kinds of AI we’re seeing here has already demonstrated actual functional uses. Like, I’ve already used these sorts of models to process natural language communications on a number of actual projects. There is actual tangible utility, even if their current form.

Blockchain never had any demonstrable utility to anyone.