Pinned Tweet
A pleasure to be featured in Forbes — tl;dr we need to verify what model lies behind an agent, link below!
Sentient AI researcher @edoardocontente was featured in Forbes for his perspective on AI agents.
His point: capability doesn't always equal trust, which is why his work at @SentientAGI focuses on building verifiable and auditable models via open-source AI.
forbes.com/sites/digital-…
3
0
16
4,358
This is incredible work and super rare and valuable data. It’s still early but @basiscompany is one of the most GOATED companies out there
We’re open-sourcing emotionally-steered speech across 108 countries/languages - by far the most of any steered audio ever released.
10,420 unique voice actors + 38 delivery styles (disgusted, flirtatious, whispering, singing, …)
525 hours of exceptionally curated, r expressive TTS. All audio reviewed by native speakers!
We include all the usual higher-resource locales but also extend to a critically underserved long-tail of languages (Malagasy, Pashto, Oromo, Somali, Khmer, …) and dialects (Cypriot Greek, Kosovo Albanian, Libyan Arabic, …) where expressive speech training data is nonexistent.
Reach out if you're interested in more hours / locales! (Link below)
1
0
4
150
Edoardo Contente retweeted
AI is the first copyable medium with no model provenance:
So when distillation and copying happens, nobody can prove the model belongs to them ↓
So when distillation and copying happens, nobody can prove the model belongs to them ↓
Lexicographers planted words that did not exist.
Mapmakers drew streets that led nowhere.
Artists placed hidden signatures inside the work itself.
Long before AI, creators had ways to catch copying to prove provenance ↓
13
7
34
10,925
Edoardo Contente retweeted
Lexicographers planted words that did not exist.
Mapmakers drew streets that led nowhere.
Artists placed hidden signatures inside the work itself.
Long before AI, creators had ways to catch copying to prove provenance ↓
Mapmakers drew streets that led nowhere.
Artists placed hidden signatures inside the work itself.
Long before AI, creators had ways to catch copying to prove provenance ↓
29
13
52
15,747
Edoardo Contente retweeted
AI is the first medium that can be copied without leaving proof of origin.
When one model distills another, no one can prove it happened.
On October 14, that will change.
Catch up on the story so far: youtube.com/playlist?list=…
When one model distills another, no one can prove it happened.
On October 14, that will change.
Catch up on the story so far: youtube.com/playlist?list=…
24
8
50
7,912
Never understood the emphasis on “matrix calculus” — it is not elegant because of how matrix multiplication is defined, it is reducible to “scalar” calculus if you can manipulate indices and it has nearly no practical use because the differentiation is always done by a computer
"Matrix Calculus for Machine Learning and Beyond" is a free set of MIT lecture notes focused on the mathematics of derivatives involving vectors and matrices.
The notes cover gradients, Jacobians, derivatives of matrix-valued functions, differentials, derivatives of matrix determinants, Kronecker products, optimization, and automatic differentiation, with several connections to machine learning.
I think it is a useful resource if you already know basic linear algebra and calculus and want to understand how differentiation is extended to vectors, matrices, and modern optimization problems.
ocw.mit.edu/courses/18-s09…
0
1
5
169
Unfortunately I cannot tag Caixa Geral de Depósitos on X — but among far more serious reasons, the fact that it takes over an hour just to open a bank account with them is one of my top three motivations for ending all my dealings with CGD, effective today
@JNegocios @ECO_PT
@JNegocios @ECO_PT
2
0
3
128
I suggest all customers who are considering opening an account with them to refer to other banks if you care about speed, efficiency and trustworthiness
1
0
3
53
Crazy that “Jev” coined these “decision” models’ modality more so than Jev’s original nomenclature
Today we introduce MedDecider - an open weight family of Jev style decision models for medicine.
MedDecider achieves SOTA results on variety of medical tasks including knowledge questions, ICD codes, trial notes.
9B scores higher than Decider from Perplexity that is 3 times larger.
1
0
1
144
Edoardo Contente retweeted
These results don’t matter in any practical way, but imo it does expose a fundamental messiness in the world that our human preference for neatness and order obscured.
AI may show us that a lot more things are like packing n=17 squares than n=4. In this sense, it is blasphemy.
AI may show us that a lot more things are like packing n=17 squares than n=4. In this sense, it is blasphemy.
oh yeah btw guys we can do integer multiplication faster than n log n lol
I was definitely very surprised when this one came in lol
github.com/openai/math/tr…
35
65
961
77,186
Edoardo Contente retweeted
i surveyed a party, 6 people know what a tapir is, 8 do not. i feel very vindicated for not knowing what a tapir is (@Some_Jackie)
3
1
5
471
I remember distinctly having this very precise thought in 2021 when I saw this thing called Dall-E that made “intelligent” images out of prompts
there are many many people who studied physics and came to the conclusion that working on machine learning would probably create more scientific progress than studying physics including many of the original chatgpt crew like john schulman, liam fedus, amodei & jared kaplan at ant
0
0
6
149
Edoardo Contente retweeted
@edoardocontente Dizem que o homem quando ama
Não espera por ninguém
Não espera por ninguém
0
1
6
64
I have seen @crestonbrooks work day and night for this incredible release — go check it out!! The largest duplex dataset ever released
We’re excited to release @basiscompany’s first open source dataset: 1,500 hrs of track separated, multi-speaker conversations across 22 languages. The largest ever of its kind and first for several languages.
Basis has built up a network of over 1 million contributors submitting ns, ratings, transcriptions and will keep open-sourcing interaction data to cover all the weird/ambiguous/natural modes of human interaction (more soon!).
This release includes 70+ of hours with 3+ live speakers and 2645 native speakers meeting, sharing, arguing, laughing, trolling in English, Spanish, Japanese, Korean, Georgian, Mingrelian, German, French, Italian, Hebrew, Russian, Hindi, Arabic, Ukrainian, Dutch, Xhosa, Zulu, Portuguese, Chinese, Turkish, Polish, and Kazakh.
For some of these languages, this is the first ever open release of duplex conversational data.
100 hrs are densely labeled with 113k human judgements on subtleties that models struggle with (was this backchannel sympathetic or frustrated? Was this silence awkward or turn-holding? Was this utterance directed or general? …)
Reach out if you’re interested in more!
ჯგირი მორაგადეს ჯგირი მარჩქილე ოკონია
1
0
8
340
Love this quote
stealing from @natfriedman: 'we are all tied down by invisible orthodoxy'. all of Jev's capabilities were possible with existing models, yet just slower/expensive/tedious. we've acclimated to AI's limitations and forgotten what products are possible without them. much alpha here
1
0
1
99