"With symbolic AI, you don't just get predictions or answers. You also get transparent explanations for the results, so you understand how the AI arrived at its conclusions."
This is a clip from my recent interview with
@matteorizzi
and Jyske Bank TV.
Our
#ScientificAI
system includes a science-specialized large language model. It is trained and operated by the incredible
@AbzuAI
team in
#Barcelona
.
Maybe we should call it CatGPT
@narendramodi
@AjitSin39185415
@isro
YES! Allow me to express my congratulations and gratitude to the entire Indian people. This is a monumental and awe-inspiring achievement, and a wonderful gift to humanity. 🇮🇳🌏🌕🙏
The battle begins: Is AI learning from copyrighted material unlawful?
Every research paper published in recent decades is copyrighted. If AI learning is deemed unlawful, AI is doomed to remain ignorant until the ~120-year copyright terms end.
🪦🤖
@bjork
There are a couple of errors in that map. The 1946 one is fiction.
The UN plan is correct, unfortunately, the Arabs rejected it and opted for a war to get it all instead.
Between 1949-1967 the West Bank was part of Jordan, and Gaza was occupied by Egypt.
A 16th century LLM trained on the texts of the time would have insisted that Galileo was wrong.
And if it didn't, the church would have implemented alignment regulations to ensure that it did.
Let's be careful with regulating the truth.
#ai
#science
Hybrid AI systems that combine the intuitive thinking of LLMs with the reasoning power of symbolic AI will be the next phase shift in the AI revolution.
The raw creative power of LLMs combined with the logic and reasoning of symbolic AI is formidable.
🤐 Here is a little trip down memory lane in the hope that it will inspire today's lawmakers:
"If liberty means anything at all, it means the right to tell people what they do not want to hear." - George Orwell
"Give me the liberty to know, to utter, and to argue freely
The claim that consciousness does not exist, or is somehow beyond the grasp of science, is a fallacy that I would love to debunk.
I wrote a little thought experiment in this article that I hope will convince you.
I consider myself a Dane🇩🇰 and European🇪🇺. I have supported the EU all my life. I nearly despaired on the night of Brexit.
But if EU regulation ends up detaching our continent from the AI economy of the future, Brexit may end up having been the wise move in the long-run😔
My hikes are different these days.
While I am walking a bunch of smart and patient AI agents keep working for me.
I explain a bit about what that means in this video.
We're entering an era where some of us have genuine digital colleagues; AI agents that can perform tasks that humans used to do. The boost in productivity is huge, but the question is, how can we humans adapt?
Here are some thoughts:
In the past, it was mainly manual labor that
If you want to have a say about a possible US ban on open source in AI research, you have 268 days to speak up.
From the briefing about
@POTUS
Oct 30 executive order
@elonmusk
@xai
There is an alternative to black-box AI, and it is not xAI.
At
@AbzuAI
we call it Scientific Artificial Intelligence. Here is how it works:
@SaffronJivi
@stats_feed
I fully accept if Indians of today prefer the country to be called Bharat. It is indeed an ancient name for the people and the land.
However, the name "India" is derived from the river Indus, which was called Sindhu in Sanskrit. When the Persians interacted with the Indians,
I strongly advocate for AI firms to build & run their own models.
Case in point: Deteriorating quality and accuracy of GPT-3.5 & GPT-4 since March.
I'm very happy
@AbzuAI
runs our entire AI stack in-house.
@brhooooom911
@stats_feed
There is nothing wrong with the name Bharat. It has been in use since at least the Rigveda as the name for Vedic tribes in the area, lasting for at least 3 millennia.
"India" also has a similar long history.
"Shindu" (the Indus River) in Sanskrit became "Hindu" in Persian,
@soumithchintala
@colesbury
No GIL in Python!❤️
For 20 years I have used Python and C as my two complementary "native" languages in most projects I do.
Call into C, release GIL, do heavy work, reacquire GIL, return to Python.
I can code that pattern in my sleep. But I won't miss it.👋
Why do so many startups brand themselves with a tech they use - AI - rather than on the problem they solve?
Another podcast with me in Danish🇩🇰. This time from
@Bootstrappingdk
. Have a listen:
Abzu is a leading AI research company in Barcelona.
If you're an ambitious AI engineer, with an interest in building, training and running LLMs, now is your chance.
We are looking for Senior AI and LLM engineers. Read more and apply
Do you want to be part of a leading AI research team in Copenhagen🇩🇰, and help put Europe🇪🇺 on the AI map?
We are looking for AI and LLM experts who want to build AI models, and not just use APIs from other AI companies.
Read more and apply:
Both Newtonian physics and relativistic physics assign a significant role to "mass."
But what if we've been viewing it wrong? Perhaps "entropy" is a more suitable candidate to assume this central role.
Could this perspective resolve contradictions in the theories, eliminate the
I'm honored to be a part of the GLOBAL CONVERSATION on the responsible use of tech.
I'm in good company with amazing speakers like Andeep Singh Gill (UN), Mike Walton (UNHCR), Manos Antoninis (UNESCO), Andreas Schleicher (OECD), and William Florance (Google for Education). This
- Generative AI: tell me things that somebody else already knows
- Predictive AI: predict things without understanding them
- Scientific AI: Understand things that were never understood before
Do you see why this difference is important?
Can symbolic AI work with less data, deliver full transparency and help build trust?
IBM's Al Martin and I jumped in a conversation for his "Making Data Simple" podcast:
"We will start to realize that we cannot trust what we read, what we see, what we hear. We will learn that we have to know the source ..."
- me, in a recent fin-tech panel at Microsoft Denmark.
A brief intro to a forgotten machine learning method. Everyone interested in trustworthy and explainable AI should know about it.
#science
#symbolicregresssion
The idea of objective reality is flawed - physicists have known it for a century, but still we cling to it. Many things in science and philosophy becomes easier if we let the subjective-objective dichotomy go.
@ylecun
I don't fear that AI is an existential threat to humanity.
I think it is equivalent to an alien species, and I would not fear that a hypothetical alien species from Alpha Centauri would want to kill humanity either.
I guess I'm an optimist.
We don't know anything about the intrinsic properties of the world out there. We only know how it is experienced by us through our senses.
If you think that conscious experience is an illusion, then you must also accept that we know nothing at all about anything.🌑
We perceive the world solely through our conscious experience. Two implications:
1. Our understanding of the world is necessarily incomplete.
2. Every assertion about the world takes conscious experience as its basic axiom.
@ylecun
Thank, Yann LeCun, you for being a strong supporter of the open source community in these trying times.
Remember the bad old days when Microsoft was trowing all their weight into the battle against open source? Steve Ballmer even called Linux a cancer!
History repeats itself.
Quote from a new article published last week:
"The QLattice algorithm outperforms traditional ML algorithms in all evaluation metrics, achieving over 90% accuracy on the test dataset. Additionally, compared to traditional black-box ML algorithms, QLattice stands out for its
Will AI be conscious? Will it be dangerous?
I have been writing about it in the "Machine Consciousness" publication.
Would you like to share your view by contributing as a writer?
If you could have your own AI assistant that knew "everything" up to the last time you synced it online, would you still use Google search and other online search engines?
I believe that open scrutiny and transparency in AI are better safeguards than closed, secret development.
If you agree, join me in signing this open letter from the Mozilla Foundation.
#AI
#Transparency
Will AI become conscious? And what is missing for that to happen?
In this four minute video, shot while hiking near Sjælsø, I do my best to answer to those questions.
Do you think your workplace will have digital employees by the end of 2024?
A digital employee is an AI agent capable of doing a complete job that was previously done by a human.
"My adventure with computers began in 1981 when my buddy, Allan, showed up with a Sinclair ZX81..."
That's how my recent interview with Authority Magazine starts. We proceed to cover everything from AI, startups, and my dream lunch date.
I wonder if someone could get
@ylecun
and
@ESYudkowsky
to sit down and talk it through some day. Their Twitter debate is perhaps entertaining but not productive.
@lexfridman
would be the ideal host.
There's a trend to use the language of science against free thought.
Yet science thrives on curiosity – the stupid, crazy, blasphemous, skeptical, and revolutionary thoughts.
#thoughts
#speech
#science
Super conductors, fusion energy, AGI: The 20s might just be humanity's craziest decade.
The implication of each breakthrough would be immense, the implication of the three together is truly mind-blowing🤯
Despite its modest title, this paper is revolutionary – and I don't use that term lightly.
An AI-discovered a sorting algorithm in the C++ standard library. If you have any background in computer science, this should be nothing short of startling.
Scientific Artificial Intelligence for single case research?
Psychologists, speech language pathologists, and many other fields rely heavily on single case studies. Can scientific AI help?
I think so. But let me know what you think!
@llanga
@colesbury
No GIL in Python!❤️
For more than 20 years I have used Python and C as my two complementary native languages in all projects I do.
Call into C, release the GIL, do parallel work, reacquire GIL, return to Python.
I can code that pattern in my sleep. Don't think I'll miss it.
"Hallucination is what we call it when we don't like it, creativity is what we call it when we do like it"
- Marc Andreessen speaking about LLMs in the Lex Fridman podcast.
A little gem💎. If you don't listen to
@lexfridman
already, you should really try.
I often engage in inner monologues or dialogues in my mind, and I just learned to my surprise that not everyone does.
When you think to yourself, how often do you experience an "inner speech" (conversations or narratives in your mind using words or sentences)?
Two reasons AI, and in particular, LLMs, may end up being harshly regulated:
1. AI disrupts the information monopoly advantage of the tech giants.
2. AI threatens the jobs of the most influential and powerful people in our society.
@karpathy
I'm a bit skeptical about the importance of compute. I am also craving more computing power, but I think the next breakthrough will be algorithmic.
The current rapid wave of AI progress has opened the board for new algorithms waiting to be discovered.💡
@pelledragsted
Jo, men der er alligevel en forskel ikke? Det er sjældent at israelske soldater voldtager palestineske kvinder ihjel mens de filmer det, og derefter trækker ligene nøgne gennem byen, mens de israelske tilskuere jubler.
Krig er forfærdeligt, men nogen opfører sig værre end andre.
@historyinmemes
Funny. My four absolute favourite cities are New York City, Barcelona, Paris and Copenhagen.
But the most frequent replies below are clearly Paris and NYC.🤔
Well, the eye of the beholder, I guess.
Every time you think about applying AI, ask yourself: "Wouldn't it be better if I used transparent and explainable AI?"
And if the answer is yes, then go and read about symbolic AI.
Suggested starting points in the thread🧵👇
ChatGPT generates text, not science.
Within a year, a new kind of AI will revolutionize the scientific process.
Follow me or
@AbzuAI
to keep up with the development Scientific AI
This is a high-potential new methodology in disease understanding and target identification.
"Using LLM Models and Explainable ML to Analyse Biomarkers at Single Cell Level for Improved Understanding of Diseases"
The authors,
@Salvatore3Marco
and Jonas Elsborg combine scBERT
@DrJimFan
Indeed, but do you have any idea how difficult it is to remove the GIL, and how many brilliant computer scientists have tried during the last 20 years?
Fingers crossed that this PEP succeeds where so many others have failed.
Europe🇪🇺 is still competing in the AI race.
@AbzuAI
is one of the few companies in Europe that trains and operates large language models similar to ChatGPT as part of our Scientific AI platform.
We get many of our best ideas on days such as these. Three of us hiking in Vall de
Today's AI agents are run on your behalf by centralized AI companies. This is why events like these can happen.
When
@OpenAI
operates your AI agents on your behalf, they can choose whether to let the AI agent access other services and information for you. I don't blame them. I'm
We've experienced an era where access to data was a game changer. Centralized information storage and retrieval exponentially increased business value, leading to the creation of today's tech giants.
Enter the era of AI: An LLM is essentially a compression of everything that
Symbolic AI in insurance - six use cases
I have dedicated my career to building symbolic AI. I believe that this approach is the only way to address the long-term risks associated with AI.
But symbolic AI is not just a technical term. It is very often the right way to do AI
@m7mdkurd
Seriously? Thinking back to when Denmark was occupied by Germany during the Second World War, and the Danish freedom fighters. Had they behaved like this, their struggle would have earned nothing but scorn forever, regardless of the cause.
ChatGPT generates text. Abzu AI generates new scientific knowledge.
The core AI is developed by the most talented AI team in
#Barcelona
.
The code is from Barcelona, the attention is global. Thank you for the recognition
@NordicAwards
We really need to slow down AI regulation until we have seen the problems we want to solve.
It is impossible to regulate well for problems that have not yet materialized.
I am not against regulation. I just want to make sure we solve problems, not make problems.
🧵
@ImNotShadyy
@AbzuAI
Yes, it is the flag of the autonomous community of Catalunya. It's called the Senyera. There are other Catalan flags as well with different symbolisms.
@DrJimFan
The LLama 2 license prohibits the use of its output to enhance other LLMs and imposes a 700M MAU cap. Both stray far from the spirit of
#OpenSource
.
Painting stripes on a horse🐴 doesn't make it zebra.🦓
#NotOpenSource
"Technology does not change the world, humans change the world with technology"
Thank you,
@EU_Startups
and
@Thomas_Ohr
for letting me speak my mind on this podcast. We got around some topics that are really important to me.
@yanisvaroufakis
Now please - run through your list and see if you can come up with a single, unilateral thing that each entity could do that would end the war immediately.
Please share your results 🙏
🔍 Generative AI: Relays known information.
🔮 Predictive AI: Forecasts without comprehension.
🔬 Scientific AI: Discovers previously unknown insights.
There is a need for all three, but
@AbzuAI
focuses on the latter. Follow us to learn about
#ScientificAI
"How could anybody ever have been led to do something so silly as to deny the existence of conscious experience, the only general thing we know for certain to exist?"
- Galen Strawson
I'm re-reading
@gstrawson
and re-loving it😍
Last year, when people said AI, they almost always meant machine learning, this year they almost always mean ChatGPT
AI means the study of building machines that can do what humans can do. It is a pretty broad definition