Jeremy Baum

@Jackyhuang @ChristinaPhili5 so great to (finally) meet!!

I ❤️ NYC MofNTS! @ChristinaPhili5 & I hosted a special edition of Members of Non Technical Staff Dinner in NYC this week! There's SO many incredible operators here esp the design talent. It was so fun bringing everyone together & none of them wanted to move to SF but I tried (,:

@justinsliao dm'ed!

anyone looking for a 1br in north williamsburg? moving and have a gem of an apt under market price. flexible lease take over. comment or dm for details.

Making a model useful means making its outputs trustworthy: verifiably true, or at least grounded in sound reasoning. But for many important domains (non-verifiable or long-horizon tasks), correctness is hard to define when there's more than one right answer and no feedback loop tight enough to verify it worked. Training to produce truth (instead of the appearance of truth) is hard.

thank you to @amytam01 @_katzdaniel @danielmend_ and everyone who helped review 🫶

the best ideas are the ones that come when you are eyes closed & in bed. naturally, these are the hardest ideas to write down...

About to annoy so many people

@rtwlz i can't wait

I'll be at NeurIPS this week from 12/2 to 12/7! Would love to chat if you're excited about data, multimodal LLMs, or evals (across modalities). DMs are open 🙂

talked about voice ai + emotion with @lilyjclifford and linda sheng from minimax :)

@tszzl i stand by that parisians are misunderstood

whenever i visit paris I have whatever the opposite of Paris syndrome is. it was shocking to me that such a place could exist and not be the most important city in the world and not incessantly talked about

@heyaytac @hume_ai what voice were you using? feel free to dm details and we’ll look into it!

@hume_ai Honest opinion it is not great when it comes to german with EV4 mini. It lacks a lot of nuances, switches to english when repeating a phone number and cuts words at the end of a sentence. I have tried with different voices and latency is far more then 200ms.

Introducing Octave 2: our next-generation multilingual text-to-speech model What’s new: - Fluent in 11+ languages - 40% faster (<200ms latency⁠⁠) & 50% cheaper than Octave 1 - Multi-speaker conversation - More reliable pronunciation - New voice conversion & phoneme editing capabilities For the month of October, we’re offering 50% off our Creator plan - use code OCTAVE2 at checkout!

@sdianahu @pipecat_ai @GoogleDeepMind @boundaryML @covaldev @langfuse @tavus super exciting 👀 would love to sponsor from the Hume side!

Excited to host a real-time voice AI hackathon at YC's office on 10/11 with @pipecat_ai @GoogleDeepMind @boundaryML @covaldev @langfuse @tavus Sign up here: events.ycombinator.com/pipecat-gemini… We'll offer guaranteed YC interviews to the winners

@mMoriLiu midnight grind pays off

another midnight slack thought came true thanks to @_jeremybaum

our logo ↑ is soo cute wth🥹

@hume_ai i kinda hate this

did you know that GPT-5 is basically da Vinci?? try below, if you want...

@BrennerSpear @hume_ai EVI 3 is a speech language model too, like GPT-4o realtime voice 👀

Some models use a speech-to-text → LLM → text-to-speech pipeline: - Qwen2.5 Omni (open source, with API!) - EVI 3 from @hume_ai Hume’s EVI 3 is unique: - Low latency via their tightly integrated API - Choose your LLM - Use any voice (even your own!)

The future of silicon intelligence is here: truly omniscient models that process text, audio, image, and video, just like our brains handle multiple senses But how close are we to real, seamless multimodal AI? I built a demo to let you try a few speech-to-speech models:

2024: Voice Cloning 2025: What about personality cloning? Hume’s voice AI can now not only mimic your voice but also speaking style and language. It’s now available via our TTS and new speech-to-speech model, EVI 3, which is also launching today.

@hume_ai @JamesLiounis_ perplexity runs EVI 3 + web search :)

@Presidentsegun @BrianRoemmele voters on huggingface.co/spaces/TTS-AGI… say it’s much better 😊

@BrianRoemmele Is this any close to being better than elevenlabs?

Just got early access to Hume! A hyperrealistic voice cloning system with speech-to-speech model. With just a few seconds of speech you can fully customize your conversational AI. Early demo is limited but you can try it now! I just did Carl Sagan! demo.hume.ai