4 ms·

This was already posted here: https://news.ycombinator.com/item?id=43221377 https://news.ycombinator.com/item?id=43221377 but I’m really surprised at the lack o

This was already posted here: https://news.ycombinator.com/item?id=43221377 https://news.ycombinator.com/item?id=43221377 but I’m really surprised at the lack of attention this model is getting. The responsiveness and apparent personality are pretty mind blowing. It’s similar to what OpenAI had initially demoed for advanced voice mode, at least for the voice conversation portion. The demo interactions are recorded, which is mentioned in their disclaimer under the demo UI. What isn't mentioned though is that they include past conversations in the context for the model on future interactions. It was pretty surprising to be greeted with something like "welcome back" and the model being able to reference what was said in previous interactions. The full disclaimer on the page for the demo is: " 1. Microphone permission is required. 2. Calls are recorded for quality review but not used for ML training and are deleted within 30 days. 3. By using this demo, you are agreeing to our " edit: Actually this has been posted quite a few times already and had good visibility a couple days ago: - https://news.ycombinator.com/item?id=43200400 https://news.ycombinator.com/item?id=43200400 Others: https://hn.algolia.com/?q=sesame.com https://hn.algolia.com/?q=sesame.com

hn_user82179 2y ago
It was genuinely startling how human it felt. Apparently they are planning on open-sourcing some of their work as well as selling glasses (presumably with the voice assistant). I’m very excited to have a voice assistant like this and am almost a bit worried I will start feeling emotionally attached to a voice assistant with this level of human-like sound.
- jofzar 2y ago
  I still feel like they don't have the right amount of human to them, maybe it's because I'm Australian and it sounds like I'm hearing an American robot? Edit: well I asked the "male" model to speak more like an Australian and yep, getting way more uncanny. If it had an Australian accent I think it would mess with me more
- igleria 2y ago
  Maybe the ability to personalize the voice so it is more... robotic or based on a fictional thing like Knight Rider would help to change the attachment to something more... healthy?
  - MarcelOlsz 2y ago
    Yeah this is straight up creepy, and I also can't stand chatgpt saying "Lmao" and "Yeah". Keep it formal & robotic.
    - WesolyKubeczek 2y ago
      What ever did you tell ChatGPT so it responded with "lmao"? I told it that it should behave explicitly like a computer in the system prompt, sort of worked.
      - MarcelOlsz 2y ago
        After multiple prompts and utterly garbage output: https://i.imgur.com/5aOARCV.png https://i.imgur.com/5aOARCV.png I'm almost positive that some AI systems have a backend that analyzes the sentiment of your messages and if you threaten to cancel billing it will notice your defcon-1 sentiment and spin up some more powerful instances behind the scenes to tide you over. This is actually much more stressful than working without any AI as I have to decompress from constantly verbally obliterating a robotic intern. I'll try with the system prompt. Also love your username.
        WesolyKubeczek 2y ago
        > After multiple prompts It generally maintains the tone you set. Remember that it outputs most likely tokens based on the system prompt of its owners + your system prompt + the whole conversation. If OpenAI and default system prompt tell it that it's a helpful cheerful secretary/assistant, you get best results if you talk to it "professionally". I heard you could make Claude say "kurwa" a lot while helping you program in Go if you convince it that you want a conversation with your ziomek Seba from your backyard with whom you like to share kebab and browar, so there goes.
anon373839 2y ago
It really is an astonishing technological feat! Also note that the largest model they trained is only 8.3B parameters (8B backbone + .3B decoder). It's exciting to think that they're going to be releasing this model under an Apache 2.0 license.
Mistletoe 2y ago
Just realizing how uncanny valley it is to talk to AI and it never remembers anything you said in the past. Imagine if a human did that. It’s like you are talking to Tom Hanks’ Mr. Short Term Memory from SNL over and over. https://youtube.com/watch?v=C6ufImch00g https://youtube.com/watch?v=C6ufImch00g
- ekianjo 2y ago
  That can easily be fixed if you attach it to a RAG system
- micw 2y ago
  I does remember but you have to ask for. Try to say "make a bookmark at this point" and later ask for that bookmark. You can even give the bookmark a name or ask it to do so for you.
huijzer 2y ago
> This was already posted here: https://news.ycombinator.com/item?id=43221377 https://news.ycombinator.com/item?id=43221377 but I’m really surprised at the lack of attention this model is getting. I'm surprised by the lack of attention that Gemini 2.0 with native audio output got. They have a demo at https://youtu.be/qE673AY-WEI https://youtu.be/qE673AY-WEI, which I think is really good too. The main problem with Google's model is that this audio output is not supported by the API, but you can try it at https://aistudio.google.com https://aistudio.google.com. In general, text to speech is pretty good nowadays I think. For example, this is a little math video that I made a few days ago: https://www.youtube.com/watch?v=G1mvLrCfjFM https://www.youtube.com/watch?v=G1mvLrCfjFM with the (old) Google text to speech API. Honestly, I think the narration is better than I personally could have done. It's calm, well pronounced, and sounds relatively enthusiastic.
- smusamashah 2y ago
  How do I get to this in aistudio.google.com?
  - huijzer 2y ago
    I think the one under "Stream Realtime" should be similar to the demo. It's only Gemini 2.0 flash though and not the full one.
- moralestapia 2y ago
  >They have a demo at https://youtu.be/qE673AY-WEI https://youtu.be/qE673AY-WEI That's not a demo, that's a video. Anyone can make something like that in an afternoon with a couple friends and a microphone. Also, Google is known for putting out fake "demos", remember the Google Duplex scam?
  - underdeserver 2y ago
    Scam? Duplex worked.
    - deleted 2y ago
znpy 2y ago
> 2. Calls are recorded for quality review but not used for ML training and are deleted within 30 days. Sounds (pun intended) reasonable.

hn_user82179 2y ago

It was genuinely startling how human it felt. Apparently they are planning on open-sourcing some of their work as well as selling glasses (presumably with the voice assistant). I’m very excited to have a voice assistant like this and am almost a bit worried I will start feeling emotionally attached to a voice assistant with this level of human-like sound.

jofzar 2y ago
I still feel like they don't have the right amount of human to them, maybe it's because I'm Australian and it sounds like I'm hearing an American robot? Edit: well I asked the "male" model to speak more like an Australian and yep, getting way more uncanny. If it had an Australian accent I think it would mess with me more
igleria 2y ago
Maybe the ability to personalize the voice so it is more... robotic or based on a fictional thing like Knight Rider would help to change the attachment to something more... healthy?
- MarcelOlsz 2y ago
  Yeah this is straight up creepy, and I also can't stand chatgpt saying "Lmao" and "Yeah". Keep it formal & robotic.
  - WesolyKubeczek 2y ago
    What ever did you tell ChatGPT so it responded with "lmao"? I told it that it should behave explicitly like a computer in the system prompt, sort of worked.
    - MarcelOlsz 2y ago
      After multiple prompts and utterly garbage output: https://i.imgur.com/5aOARCV.png https://i.imgur.com/5aOARCV.png I'm almost positive that some AI systems have a backend that analyzes the sentiment of your messages and if you threaten to cancel billing it will notice your defcon-1 sentiment and spin up some more powerful instances behind the scenes to tide you over. This is actually much more stressful than working without any AI as I have to decompress from constantly verbally obliterating a robotic intern. I'll try with the system prompt. Also love your username.
      - WesolyKubeczek 2y ago
        > After multiple prompts It generally maintains the tone you set. Remember that it outputs most likely tokens based on the system prompt of its owners + your system prompt + the whole conversation. If OpenAI and default system prompt tell it that it's a helpful cheerful secretary/assistant, you get best results if you talk to it "professionally". I heard you could make Claude say "kurwa" a lot while helping you program in Go if you convince it that you want a conversation with your ziomek Seba from your backyard with whom you like to share kebab and browar, so there goes.

anon373839 2y ago

It really is an astonishing technological feat! Also note that the largest model they trained is only 8.3B parameters (8B backbone + .3B decoder). It's exciting to think that they're going to be releasing this model under an Apache 2.0 license.

Mistletoe 2y ago

Just realizing how uncanny valley it is to talk to AI and it never remembers anything you said in the past. Imagine if a human did that. It’s like you are talking to Tom Hanks’ Mr. Short Term Memory from SNL over and over. https://youtube.com/watch?v=C6ufImch00g https://youtube.com/watch?v=C6ufImch00g

ekianjo 2y ago
That can easily be fixed if you attach it to a RAG system
micw 2y ago
I does remember but you have to ask for. Try to say "make a bookmark at this point" and later ask for that bookmark. You can even give the bookmark a name or ask it to do so for you.

huijzer 2y ago

> This was already posted here: https://news.ycombinator.com/item?id=43221377 https://news.ycombinator.com/item?id=43221377 but I’m really surprised at the lack of attention this model is getting. I'm surprised by the lack of attention that Gemini 2.0 with native audio output got. They have a demo at https://youtu.be/qE673AY-WEI https://youtu.be/qE673AY-WEI, which I think is really good too. The main problem with Google's model is that this audio output is not supported by the API, but you can try it at https://aistudio.google.com https://aistudio.google.com. In general, text to speech is pretty good nowadays I think. For example, this is a little math video that I made a few days ago: https://www.youtube.com/watch?v=G1mvLrCfjFM https://www.youtube.com/watch?v=G1mvLrCfjFM with the (old) Google text to speech API. Honestly, I think the narration is better than I personally could have done. It's calm, well pronounced, and sounds relatively enthusiastic.

smusamashah 2y ago
How do I get to this in aistudio.google.com?
- huijzer 2y ago
  I think the one under "Stream Realtime" should be similar to the demo. It's only Gemini 2.0 flash though and not the full one.
moralestapia 2y ago
>They have a demo at https://youtu.be/qE673AY-WEI https://youtu.be/qE673AY-WEI That's not a demo, that's a video. Anyone can make something like that in an afternoon with a couple friends and a microphone. Also, Google is known for putting out fake "demos", remember the Google Duplex scam?
- underdeserver 2y ago
  Scam? Duplex worked.
  - deleted 2y ago

znpy 2y ago

> 2. Calls are recorded for quality review but not used for ML training and are deleted within 30 days. Sounds (pun intended) reasonable.