#Discussion
1 messages · Page 1 of 1 (latest)
They should add emotional voice like elevenlabs
Elevenlabs emotion is way too unstable, too many artifacts and lacks coherence. Sometimes switches voice mid sentence.
There's already other models coming soon or that's already out that's comparable and better
That's weird it never happens to md
Me*
I use it all the time and it's perfect?
Maybe they've fixed it, I don't know but when it came out it was kind of ass
I'm more excited about this tts
Omg you should try the v3
Like is sooooo realistic
They fixed it
Aah nice, I'll give it a try again. If it's as good as the videos it might be huge but man do I wish it was open source
Either way on topic, gpt 5 is kind of idk not that impressive.
I think google will soon release something better, grok is already training a new model and meta has yet to show what their billions of dollars in buying up ai researchers will lead to
But meta has announced their building agi and releasing it in due time while sam Altman falls flat on most promises he makes
Gpt 5 is nice but not impressive enough for what should have been a huge update
Does Gemini 2.5 still compare to gpt-5? Or is google behind
It's better in some cases
And that's just 2.5 pro and not the new deep think
they're announcing 2 models to run on consumer GPUs
I'm not super in the know about AI, but meta wants to push open source stuff, right? ^^
Not an expert either, but LLaMa was big on the local scene for a while, and that one came from Meta
Yes, used to but in the future they will probably stop with that for the most part because it's "too dangerous"
Wouldn't it be more dangerous to only let billion $ companies use it? If everyone has it, at least there's no advantage. If you find a machine to run it on, that is.
Yes and no, both has it's benefits and consequences. On one hand once something is opensourced it's out, it's like opening pandooras box. It will aid others to do their own research and development and bad actors can finetune a quantized version that's easier to run with some amount of money and if it's "good enough" or what true AGI would be in this case it could prob help you develop a nuke or whatever else you wish. On the other hand it's also great having it opensource because it also democratizes access to intelligence to everyone which is prob one of the biggest positive socioeconomic things that could happen in my opinion.
Having it be closed source still means people could probably use it if they decide to give access to it which is front and center in Metas plan and directly said by mark zuckerberg, now if you trust him or not that's up to you to decide, I think he is overall a good person. But the other benefit about it is that it's self contained and they can decide to sunset the model if it's rouge and they can also limit bad actors.
the pandora's box is the dataset used to train which may originate from some copyrighted sources, people's personal biases & misinformation, unsafe contents, etc.
I think that's dumbing down the concept of AGI. By the time we have proper AGI it would be able to come up with novel ideas outside of it's dataset, which AI is already able to do to some extent. We're talking about general intelligence, not a Wikipedia copy paste machine in a compressed format. People can already seek all of that information without AI so it's not much of a pandooras box to have a dataset in my opinion.
Since this came out a few weeks ago, been trying out IndexTTS2. This is like, back when 15.ai was out, except you can use your own clips and control the emotions with far less effort, and the length of audio matters less if the sentences are very clearly spoken!
