I would argue that Alexa+ absolutely qualifies as GPT4 class. It can answer complex questions and, more importantly, has tool use and self-programming capabilities (“Alexa, make it so that the outside lights turn on at sunset and turn off before midnight, but keep the Christmas lights on until dusk”).

They don’t disclose which models they’re using, but they’re using Amazon nova at worst, with contemporary Claude models used on more complex questions.

Even though it’s obviously tuned to be a consumer product (shorter answers, bubblier personality), it’s clearly a jump up in terms of answer quality when compared to something like Meta AI on the Ray-Bans.

Alexa with LLM integration still cannot create essay-length reasoning or complex narrativisation. It's still mostly for one-off questions and smart homes, it's just better at the former now.

@Panfilo

i've put up YES limit orders. at what price would you be willing to buy with volume?

@Bayesian I don't have a ton of liquidity because I spent my mana the morning of the first and it didn't count for leagues because of the unannounced turnover change. But I'll do some more research and see if I'd be willing to at least throw 10k at this if it still looks promising.

@wispyartichoke Tagging again although you never got back to the guy from nine months ago, do you consider benchmarks to be what's important here, or the complexity of reasoning tasks actual users can effectively leverage the product for?

@Panfilo hey, sorry, not active on the site these days, but i’ll try to resolve this fairly. I think that if benchmark results make the answer super clear, we can use those ideally and be more objective, but otherwise i’m happy to engage with arguments about the actual quality of reasoning

Can @wispyartichoke clarify what GPT4-level means?

I'd vote yes, because Google announced they would completely switch their assistant to Gemini. However, I don't like that it says "GPT-4" level, as it's not clear that anybody is going to measure it and also it would likely be more restricted just by nature of being new and being a voice assistant (i.e. more limited and focused in responses).

Why are people so confident about this?

So far, they have not even expressed any interest in doing this.

@DavidBolin I agree. I’m betting NO due to the high cost of GPT-4 inference. I think the unit economics would be prohibitive for Alexa.

@DavidBolin haha 9 months later... It got cheaper

Related