digital creator | GTM @elevenlabsio | bringing talented people together @soti_house
Dublin City, Ireland
Joined February 2018
- Tweets1.5K
- Following290
- Followers907
- Likes2.6K
the interesting part imo is what this could do to training costs (if the efficiency gains hold more broadly)
more compute gets the headlines and its what we all see. but needing less of it might be the better story?
I've always believed that intelligence should come from simple, general building blocks that learn their own abstractions directly from raw data; in other words, the bitter lesson. I think we're finally seeing a real step towards AI that can learn from all data.
Language modeling on raw bytes is a microcosm of this ideal, and to me it's a central problem in AI: if we can't build LLMs that learn end-to-end from bytes, how will we handle the rest of the world's data that we understand even less?
Last year we introduced hierarchical networks (H-Nets) as a first step. Our new version finds up to 3-5x efficiency gains in controlled scaling law experiments, and at larger scale matches an open-source model trained on 11T tokens with ~10x less data and compute. The posts below walk through what we found and why we think it works.
cool motion graphics tools / APIs to check/use/invest in?
@fildotai @craviottx Hyperframes? quiver?
I remember yapping about super personalised AI content a few years ago (even before joining ElevenLabs)
it's here. in the next 3 months it's going to be decided who are the main players in terms of tech and distribution.
if you're thinking about getting into content, do it before 2027 or you're cooked
yesterday: event on building ai agents, with a guy who mentioned that a travel company charged 5€ extra for the agent interaction to @heycesr
today: travel data platform that allows you to adapt your website, recommendations and sales motion if the visitor is an agent
I'm on the edge of the edge
new edition is out. building got cheap. attention didn't.
so we went looking for who was operating.
valeraloop.substack.com/p/we…
new edition is out. building got cheap. attention didn't.
so we went looking for who was operating.
valeraloop.substack.com/p/we…
btw I posted about it on Linkedin MANUALLY. I think I hadnt post it like this >1 yr ago
a voice people can understand isn't the same as a voice they want to talk to.
the model gives an agent capability. natural voice and distribution turn it into a habit
a reflection for my thesis after attending this panel, with @pepmartorell @IgnasiCostas @pareras, last month
a voice people can understand isn't the same as a voice they want to talk to.
the model gives an agent capability. natural voice and distribution turn it into a habit
a reflection for my thesis after attending this panel, with @pepmartorell @IgnasiCostas @pareras, last month
today is the beginning of a Marcos Valera era. I recorded a video.
thats it. my agents have decided where, how and what to post. I'll give it instructions in the future but as of now this is gamechanger for me
I have videos recorded since 5 years ago. now I can post it with @zernio and LLMs + my harness
the video: xitter.gq/_MarcosValera/status/2…
nice subtitles, watermark for free and @DJIGlobal didnt pay ✍🏼
in this case: 10 min cut in X, full video on Youtube (it attached it in the newsletter, as it's private), and created 10 trial reels.
marcos valera retweeted
Aníbal, an engineer and professor, is losing his voice to ALS. With ElevenLabs, he’s recreated it so he can keep teaching.
We’re partnering with @AsociacionELA, Spain’s leading ALS association, to give people with ALS across Spain free access to our voice restoration technology.