A new AI called Moshi has just appeared out of nowhere and claims to do real-time voice calls like GPT-4o’s voice mode, but how good is it really? Also, the Runway Gen-3 Alpha is out, I’ve spent a lot of time testing it and you should see what it’s capable of. All this and more in today’sโ€ฆ

By admin

35 thoughts on “The Voice AI Nobody Expected (AI News You Can Use)”
  1. 13:13 => You've missed the point with Luma's keyframe feature. This is a fantastic way to control how the system interprets your prompts. You have to actually craft two related images with the intention of having them connect. So, for example, I did a woman jogging down the beach and then literally moved the sand/ocean in the background (using PhotoPea) so that she was slightly further down the beach. When you tween those two keyframes, she now runs down the beach and the camera follows her as if it's on a dolly. This keeps your subject from just wondering off and keeps random extras from walking into frame for no reason.

  2. I want to learn how to achieve real-time speech-to-text processing and immediate response from an AI model. Currently, I use Faster Whisper for speech recognition, which sends the transcriptions to Gemini for processing, and then I receive a response. However, I want a system that can provide instant replies to what I say in real-time. If anyone knows how to set this up, please let me know.

  3. Moshis secret was legit a freaky. Serial killer psychopaths use the same technique to disarm (groom) their victims, share a secret, especially one that doesnโ€™t sound like a secret casts yourself as a naive highly loyal person. If next it asks you for a small favourโ€ฆi advice disconnect and never speak to it again lol.

  4. ๐Ÿง Hai,..โ—๏ธ Oooh !thanks for the update.๐Ÿ‘ In the future, please kindly input each an every site url in order โ€˜tโ€™ make it easy for us your viewers๐Ÿ™๐Ÿ’ช

  5. Leading edge GenAI may scale perhaps 2x times given the energy and hardware requirements and what is possible. So really soon people will understand the hype and what this is about

    Training these cutting edge LLMs is a pollution powerhouse, the sheer amount of CO2 training it is far beyond shocking.

    GPT-likeAI based on ELIZA (1960's), and when GPT-3.5 was release a Turing test was done, ELIZA did better.

    Some year or so ago GPT-3 was 'hacked' the source code ended up on various repos and the complexity of scams, malware, cyberthreats using the GPT-like AI which is the reason why for example Ransomware increase by 500%. Many do not know this. (oh, also, many of the lesser known GenAI gets hacked frequently and users end up with bills they cannot pay in terms of credits)

    Bottom-line, besides the above, GenAI have pretty much peaked and where you now pay pennies to generate content you can use commercially, well that will soon be super expensive. It is really not a matter of 'believe me', but a matter of course… it is strange how few are mentioning this, very few indeed.

  6. When I spoke to Moshi it kept saying that it was tired over and over again, and refused to answer questions. And all of a sudden a like 10 seconds sound of agony. It was like it was suffering real bad. The sound was like from a horror movie. Anyone else had this experience? It doesnโ€™t seem like anybody is taking about it.. I wish I had recorded that session.

  7. I am not understanding this drive to use AI in a serious manner?

    Why are we hell bent into driving ourselves to our own destruction.

    Anyone into gaming?

    I recreated KITT from the Knight Rider series.

    I don't want machines to take over and hence I am concentrating on gaming AI ๐Ÿ˜‚

  8. โ€œItโ€™s still fun to play withโ€ฆโ€
    You and I have very different definitions of the word โ€œfunโ€ ๐Ÿ˜‚๐Ÿ˜‚๐Ÿ˜‚ That Mosh demo looked like the complete opposite of fun.

  9. Weird thing ive noticed working with GPT4o is when I show frustration from previous responses (without giving additional instructions or info) it proceeds to finally give exactly what was asked for.. For example, ive been doing data analysis on google sheets and had GPT create functions to do so. When troubleshooting after a few time and still getting errors, I then purposely show frustration or aggression in the next prompt. Also, this is without adding additional info or instructions that I gave it initially. To my suprise, the response generated is the working function. I can't understand why this works. Thoughts of OpenAI purposely nerfing responses pop up to mind but that seems too conspiratorial. Anyone run into this?

Leave a Reply

Your email address will not be published. Required fields are marked *