Mustafa Suleyman says people that are unimpressed with AI's capabilities are mind blowing to him, after recent backlash around Copilot and Windows as an agentic OS.
“The fact that people are unimpressed that we can have a fluent conversation with a super smart AI that can generate any image/video is mindblowing to me."
That’s your killer feature? “But you can converse and it can make any image!”
He says it like no one has seen the party trick yet. Like, yeah dude, we were all impressed with it in 2022. Then we learned that it’s all smoke and mirrors.
I was not impressed in 2022 either when it was frequently wrong, didn’t back up anything it said with facts, sources or evidence, and business people kept hyping it immediately as some sort of revolution.
I spent a lot of time having an existential crisis before I realized I was not alone.
Plus, the “you shouldn’t expect exponential improvement” gaslighting has begun. Like, for the amount of imaginary dollars being thrown at machine learning, where the fuck’s the Brave New World being sold to investors?
But I’ve never had any desire to have any meaningful conversation with it, and it’s still a fight to make AI do valuable things.
I work with AI, just today I finished another AI tool integration (it’s actually very cool, but proprietary). The “trick” to making ai useful is to create useful tools for annoying tasks then let AI deal with those tools for you. And even that’s sketchy.
It is so easy to get hyped by the benchmarks but there are only two real benchmarks that matter: do people actually want to use it, and can it do what you promise it can.
For 99% of AI integrations the answer to both is a resounding no. Maybe when models are smarter that’ll improve, but so much AI crap is just devs trying to check a box, and not actually what users would want.
Yes, and no. It’s not all fake, there’s stuff going on, it’s just not what they’re selling it to be, and highly pushed into places it needs to stay away from, for safety and for inability. My takeaway on him being surprised as how people aren’t impressed isn’t the LLM factor of what it can do (well or not), it’s that HE isn’t aware that other LLMs are doing better than Microsoft’s version. He really is deep if he doesn’t know what the competition has. That’s why there’s a lackluster interest (as well as burnout of AI “solutions” for every damn thing, often worse than just doing it like before).
My coworkers use Co-Pilot. When they have downtime, just for amusement, just to see how badly it mangles things it ought to be good at doing. Never mind the fringes where an LLM isn’t suited at all.
That’s your killer feature? “But you can converse and it can make any image!”
So can any toddler.
He says it like no one has seen the party trick yet. Like, yeah dude, we were all impressed with it in 2022. Then we learned that it’s all smoke and mirrors.
Some of us weren’t even impressed in 2022 because we don’t want to talk to a computer or create shitty images.
I was not impressed in 2022 either when it was frequently wrong, didn’t back up anything it said with facts, sources or evidence, and business people kept hyping it immediately as some sort of revolution.
I spent a lot of time having an existential crisis before I realized I was not alone.
This is exactly it.
Plus, the “you shouldn’t expect exponential improvement” gaslighting has begun. Like, for the amount of imaginary dollars being thrown at machine learning, where the fuck’s the Brave New World being sold to investors?
I am still impressed with it.
But I’ve never had any desire to have any meaningful conversation with it, and it’s still a fight to make AI do valuable things.
I work with AI, just today I finished another AI tool integration (it’s actually very cool, but proprietary). The “trick” to making ai useful is to create useful tools for annoying tasks then let AI deal with those tools for you. And even that’s sketchy.
It is so easy to get hyped by the benchmarks but there are only two real benchmarks that matter: do people actually want to use it, and can it do what you promise it can.
For 99% of AI integrations the answer to both is a resounding no. Maybe when models are smarter that’ll improve, but so much AI crap is just devs trying to check a box, and not actually what users would want.
Yes, and no. It’s not all fake, there’s stuff going on, it’s just not what they’re selling it to be, and highly pushed into places it needs to stay away from, for safety and for inability. My takeaway on him being surprised as how people aren’t impressed isn’t the LLM factor of what it can do (well or not), it’s that HE isn’t aware that other LLMs are doing better than Microsoft’s version. He really is deep if he doesn’t know what the competition has. That’s why there’s a lackluster interest (as well as burnout of AI “solutions” for every damn thing, often worse than just doing it like before).
My coworkers use Co-Pilot. When they have downtime, just for amusement, just to see how badly it mangles things it ought to be good at doing. Never mind the fringes where an LLM isn’t suited at all.