Aria is a cool idea but so slow that I'm defaulting back to ChatGPT

bryce126

bryce126

After I ask a question or interact with Aria, the wait time for a response is much slower than I'm used to with ChatGPT.

Here's a question that took 14 seconds for Aria to begin typing:
"how do I find someone who does really good illustrations for mobile app design"
(entered into ChatGPT, this took 1.57 seconds)

This one took 5.83 seconds:
"What's Beethoven's most famous symphony?"
(entered into ChatGPT, this took 1.27 seconds)

(strangely, ChatGPT and Aria disagree about Beethoven's most famous symphony...ChatGPT advocates for Symphony No. 9 in D Minor, Aria says Symphony No. 5, as an aside)

Once the question is asked, the speed at which characters are "typed" is far too slow as well. If you can speed both the data retrieval and the actual displaying of the answer, I think people will use it more.

My internet is not an issue here. Good luck!

Lemonkicker

Lemonkicker

Nothing substantiated here, just something to consider.

FAQ and other sources show it's using OpenAI models but no specifics, the prediction speed seems consistent with GPT-4 and it has live web access now. I'm guessing there's a few more layers of logic being wrapped around Aria that take some time to gather and feed into the prompt going to OpenAI.

Fetching live grounding resources takes time, going out on a limb that Opera doesn't have the same $$$ to match Bing or Bard speed for that. ChatGPT free is fast but dumb and offline, Aria isn't bad for higher quality answers and live web access baked into a browser.

The way it embeds search links isn't displayed the same as GPT-4 generation so that filtering could also be part of the orchestration.

https://press.opera.com/2023/05/24/opera-unveils-integrated-browser-ai-aria/

'Aria, which is based on Opera’s own “Composer” infrastructure, connects to OpenAI’s GPT technology and is enhanced by additional capabilities such as adding live results from the web.'

Getting consistent subjective information is difficult, they(Aria, GPT-3.5, GPT-4) all seem to do a decent job when asking for the top x symphonies. GPT training data will likely have multiple opinions on the topic and it picks the best prediction from that given current context, abstraction and math

https://help.opera.com/en/browser-ai-faq/

'Conversation history is stored on Opera’s servers for 30 days. OpenAI’s servers will also keep anonymized (not connected to your identity) parts of conversations for 30 days. Open AI can also use questions you provide to the chat to improve their systems.'

I'm happy using Aria as an option, pop quick questions when I need internet enabled AI not tied to a large ecosystem or company. Also using Aria for cross checking another AI chat output or claim - let it crank out information in the sidebar then read when it's finished

Jungstilling

Jungstilling

Got answers from Aria Chatgpt 3.5 failt,

wonghow

wonghow @bryce126

@bryce126 AI does not give same answer, and not necessarily correct. AI only learns from data. Data are defined by people.
Something like Music, even people cannot determine which is the best Beethoven symphony, it is not a fixed answer. If you do a search, you will get answer that no9 and 5 is not the top. no3 is, because can be votes from the music community

BestCodes

BestCodes @bryce126

@bryce126

Aria uses a slow API, because APIs cost money and with millions of users it takes a lot of server power to make Aria as fast as it is even right now.
Aria already uses GPT anyway, so you aren't really 'switching', except for the occasional times when Aria defaults to Bard (PaLM, thus the disagreements).

The typing speed is in place (it could all be typed instantly) so that users can't just ask questions and get responses so fast that they would bog down the server. Instead, they have to wait for the current response to finish so that they can ask another question. Aria is unique because it is built in to a browser and has some real time data access.

@wonghow said in Aria is a cool idea but so slow that I'm defaulting back to ChatGPT:

AI does not give same answer, and not necessarily correct. AI only learns from data. Data are defined by people.
Something like Music, even people cannot determine which is the best Beethoven symphony, it is not a fixed answer. If you do a search, you will get answer that no9 and 5 is not the top. no3 is, because can be votes from the music community

That is totally true, too.

I use Aria and other AIs (Claude-Instant, Llama, Bard, and GPT4). I don't really have a preference. (Although I attempted to make an 'unbiased' AI here:
https://poe.com/BestCodes-Unbiased). I hope this was informative!

Hi, I'm Best_codes! I'm a full-stack web dev who moderates on the makeblock forum. I have a website, https://bestcodes.dev. I love Opera, so I'm on here occasionally to report bugs and answer some questions.