Alexa AI: Why Is Its Brain Stuck in the Lab?

Craig: alexa's ai brain is stuck in a lab.

Jon: yeah. this is weird to me because they were first out of the gate with well, were they before siri? well, they were they were so, anyways,

Craig: they definitely won the home voice market.

Jon: yeah. so you would think that they would still be ahead or at least in the race. and right now, they're being lapped. so what happened here is, they're trying to upgrade alexa with an advanced ai capability, and they're facing challenges. i don't know what those challenges are, but they're, it's currently stuck in develop the development phase. so in the summer of twenty twenty three, andy jassy, the ceo of elect of of amazon, he personally tested the prototype of ai's enhanced alexa, focusing on its abilities to handle complex sports related queries. so while the prototype, i it seems like it demonstrated some improvements, it still fell short of his and they're saying he has lofty expectations, but who knows? so for instance, when asked about a recent game outcome, alexa actually fabricated a score, highlighting the ongoing issues with accuracy and reliability. rightfully so. so they're having a lot of difficulties. there's a lot of competitive pressure. the prolonged development phase likely is gonna require significant investment in terms of time money. so i think what they've been doing is kind of just i could be wrong. i'm i'm making an assumption here, but, maybe they're not investing as much as they should into this, if that makes sense.

Craig: yeah. it does definitely make sense. you know, is amazon alexa really losing its edge with chat, ggp, and google as it's now sprinting ahead? because google was so far behind when these things first started. it feels like alexa has been benched, but why is the once mighty alexa ai still stuck in the lab while others are, you know, out playing the game. so, you know, let's break this down. you know, the picture when alexa debuted in two thousand fourteen, it was like inviting a clever, obedient assistant in your living room. it could play music, order groceries, even tell corny jokes.

Jon: yeah.

Craig: but now we're in two thousand twenty four, and users want more than basic commands. you know? think about it. we talk to ai all the time now expecting smart responses that understand nuance, tone, and context. each enter chatgtp you know, for its ability to hold conversation with complex questions. unlike alexa, it doesn't just follow instructions. it converses. like, you've played with, chatgtp voice, and a lot of people have.

Jon: yeah. yeah. and and that's the thing is alexa was great when it first came out. it was great to kind of do various tasks and things, and kids were using it to order stuff off amazon, which was funny. but, yeah, it's it's way too far behind. and i think they're the ones that initially like, kind of solved the nlp natural language processing kind of issue on a large scale. i don't know. like, i know there was dragon naturally speaking, but i don't think that ever came close to what they were able to do with alexa.

Craig: definitely, i think alexa was a lot ahead. i think, yeah, dry they have two different functionalities. i think some of the dragon speech was better for long term data input because i knew a lot of people who, you know, just talk to it. and it would create, like, ten page documents from their conversation rather than, what alexa can handle the snippets.

Jon: yeah. yeah.

Craig: so yeah. apparently, alexa can't keep up because it has a language problem. gets, it's great with straightforward commands like turn on the lights, but it throw in an idiom or ask it multilayered questions, and it's suddenly scratching its digital head. imagine asking your fan, could you make it feel like christmas in here? alexa would probably just respond, i don't know that. chatgp, on the other hand, might tell you to set the mood or suggest some holiday classics to play or something like that.

Jon: yeah. who knows? yeah. yeah. they definitely have to work on the accuracy. it's funny, and and under and under like, the problem that they're having is the same problem that these llms are having with hallucinations. they're generating, they're generating fake data. they're generating fake links, fake resources. and, again, i'm gonna repeat this because this is what i'm worried about is that data that's being generated is gonna be used to re train these llms, and that's worries me.

Craig: yeah. and wonder where, you know, things like the strawberry model come in where it's it's using synthetic data versus real data. but, yeah, the ai start training themselves, and we talked about this before. it becomes collapse, collapse, collapse. and hallucinations are really real thing. it hallucinates for me all the time. it's getting better, but it hallucinates for me all the time, you know, where we'll be talking about we'll be planning an event. i'll be getting a description. like, here's my description. make it better. and then it throws in speakers, dates, things that, like, a bunch of stuff that just don't exist. it just makes it up. and i'm like, you know, why would you do that?

Jon: yeah. did you ask it why i did it?

Craig: not at that point in time, but i do now. yeah. it but it usually just apologizes.

Jon: yeah. it just apologizes. yeah.

Craig: and i'm like, stop apologizing. you know, we have this thing where you produce something that was not what i wanted you to do. let's figure out how we could be more clear together. if you don't understand or i'm unclear, you know, you need to prompt me more. i need to, you know, tell me how i can give you better information. that's usually the conversations i have.

Jon: yeah. that just brings up this prompt that i shared with you this morning to or a sentence to add at the end of your prompt. and use, for for those who use prompts, this is great. so what you do is you say, ask me clarifying questions until you're ninety five percent sure of your answer. and what it does is it will ask you some really interesting questions. some you might not be able to answer, to be honest with you. but it will, it will reduce the number of back and forths that you're gonna have to do with it. and at the end, it gives you a much better output.

Craig: yeah. i'm gonna have to definitely try that one.

Jon: yeah. the other one too, did you know i we're kinda digressing, but have you used the temperature between one and ten?

Craig: i use the temperature, usually at point seven.

Jon: oh, so you do use the temperature, input or prompt?

Craig: yep. well, i use it, for certain things. i definitely use it when i'm calling apis. it's a thing you can do. i don't use it every time with my prompts.

Jon: okay. i didn't think about using it for calling apis. that's cool.

Craig: well, i've been doing a lot of that lately. so you could control temperature and a bunch of other stuff when you call it. so

Jon: okay. interesting.

Craig: so here's the kicker. amazon has been trying to bring alexa up to speed with features like alexa conversations, which was supposed to help it better understand context. but these upgrades, while impressive on paper, just haven't been any game changers. competitors like google assistant and even siri have benefited from years of linguistic training and contextual improvements backed by google and apple with their rich data troves. with every update, amazon's assistant, gets a bit better, but these improvements are barely noticeable to the average user, and the ai race is brutal. and if you're not keeping up, you're getting left behind. it looks like amazon's getting left behind.

Jon: yeah. it does look like amazon. and even, like, google, i know they're gemini, but, like, we haven't heard much about gemini, have we? like, we've heard a lot about perplexity. we've heard a lot about openai. we've heard a lot about, claude, but not much about, about

Craig: there's a lot of there's a lot of, google models out there. gemini, there's a bunch of others. but i think they learned their lesson around, let's not make a big noise because they got beaten down pretty good with the first couple of releases. hey. look how good we are.

Jon: yeah. but i will say there is a lot of talk around there. notebook l m, which is really cool.

Craig: it is cool, but it's very limited. yeah. but it's cool. i definitely, will continue to use it. i'll continue to use it in some of my workflows for personal education and whatnot. i will probably use it to add audio files to websites that are, complicated, topics and whatnot. so there's a lot of things you can do there. yeah. so some insiders say that amazon has been hesitant to take big leaps with alexa because it's tied so heavily to amazon's other services, which makes drastic drastic updates risky. alexa's job isn't just to be smart. it also, prime product pusher, as you say, getting everything to order. hey. will you, order this or add this to my invoice? and nudging you to buy things the, through its voice if, you know, it looks like you're short or you're asking around. so the split focus, may be what's slowing alexis down.

Jon: potentially. but i would think that it wouldn't be too difficult to integrate that their, into their new, ai functionality. no? or am i wrong?

Craig: well, it says it's hinted, that it's working on new voice recognition features and nlp and that, i've seen some stuff around like, it's trying to bridge the gap between rivals like chatgtp, but i've seen stuff around where they're trying to make a deal, with openai right now to, connect, alexa to chat g t p. but there's some issues around pardon?

Jon: that's what apple did.

Craig: yeah. and i think they're they're dumping this. so the big question is, do you think alexa will catch up, or is it time to pass the mic to chat gtp and disconnect chat gtp to it and let's go? is this a story of slow and steady comeback or too little, too late? you know? what do you think?

Jon: i don't know. i, it's it's tough to see. i, to be honest, i think they're they're too little too late. but who knows? maybe they gotta check up their sleep. who knows? but we'll have to see.

Craig: indeed. we love to hear what you think about the future of ai bots what this means and, you know, the ai models and who's gonna leapfrog who. drop us a comment, and we'd love to hear back from you.

Jon: absolutely.