Darío Gil: The future of AI is open

In his keynote at Think 2024, IBM SVP and Director of Research Darío Gil outlines the latest technology breakthroughs from IBM that will allow enterprises to scale AI. Watch as he explains in detail what makes for a winning AI business strategy, and why the open-source community is so important to the future of AI innovation.
#ai #opensource #think2024

[@zefer_frey6446]

This is a HUGE win for the opensource community and I am glad I was here in the beginning

[@morespinach9832]

How “HUGE” is it? :)

[@joaquimfernandopiresdecamp6815]

Amazing presentation. The future of AI is OPEN! #instructlab

[@ver1.2mystory5]

As an ordinary person, I'm incredibly excited about the dawn of the quantum computing era and the singularity after reading a paper on these topics. With the rapid acceleration of technological advancements, I often reflect on the fundamental role of humanity in this evolving landscape. I'm eagerly looking forward to exploring what can be achieved when quantum applications and quantum computers are integrated with AI.

[@morespinach9832]

Yeah, this was meant for the ordinary person.

[@joshpipe7755]

Do you mind sharing which paper you read? I'm now excited to read it too!

[@balakrishnaprabhunallendra999]

Excellent presentation 🎉❤

[@morespinach9832]

For whom? :)

[@balakrishnaprabhunallendra999]

@@morespinach9832 Obviously to Dario Gill...

[@morespinach9832]

@@balakrishnaprabhunallendra999 I mean for whom was it an excellent presentation, for what audience. This info is so basic that it’s largely for those who have been sleeping for the last two years and haven’t watched any of the thousands of excellent KZbin introductions on the same subject. Then it was all instruclab advertisement which follows the same venv setup that many similar things already offer.

[@balakrishnaprabhunallendra999]

@@morespinach9832 Who presents matter!.

[@morespinach9832]

@@balakrishnaprabhunallendra999 sure.. I understood all this 2 years ago with a simple KZbin video.

[@LuisSanchez-gp4rq]

Finally someone explaned modern day neural networks to an audience rhatbdoesnt have to know much computer sciene or math at all in a simple way.

[@morespinach9832]

Modern day neural networks? :)

[@alkwayria..26]

Important information and data for the future of artificial intelligence

[@amritbro]

Working pattern of neural network is just amazing 😊AI surely is one of the greatest happenings of our time which is going to shape the tomorrow’s world.

[@aditya7069]

can i work for IBM

[@GabrielRymberg]

How is InstrucLab different than fine tuning? Specifically, how do you ensure that all the generality of the base model remains intact?

[@abdullahalsowaygh6791]

What instructlabs does is adds knowledge to the existing model (by using a teacher model to help generate much more examples that is called synthetic data). It will then train the existing model and output a new model for you that is trained on the data you have added. Let's say you have a new use case, you can do the process again to add more and more data. So this will give you the generality of using the same model which is trained with all the data that you need. Instead of having separate models that gets tuned only for specific tasks.

[@sombh1971]

Great innovation, I thought so too that this was something possible.

[@SHANEALAM-p5x]

🎉🎉🎉🎉🎉🎉🎉🎉🎉❤IBM GREAT

[@omermindivanli2981]

Great talk!

[@md.shoyongaming114]

Now we are in future

[@thesimplicitylifestyle]

The future is symbiotic 😎🤖

[@skydweller2049]

Transhumanism

[@vp5335]

I was worried why happend to IBM Watson which was way head of anyone 😢

[@ryan77anderson]

🚀

[@mentalisme]

You guys are opening a market for us to write simple non ai programs that work against AI. e.g. something that makes sure a writer uses no ai at all.

[@canonest]

literally no one will open their data.

[@fjaresj]

O granito não é transparente

[@simingma]

dr strangelove

[@RishabhSingh-ro6pt]

samjh gya m samjh gaya aaj se YT band ludoo khelna suru

[@bigusuromaciosa4494]

No, It must be private

[@angelcamiloguillenguzman6529]

He is talking about putting you data inside an open source model so the model knows what it needs to know about your data, but the resulting finetunned model with your data does not need to be public. Confidentiality it is very needed charasterisitic of information and your LLM is information and an engine to access it packaged in a single artifact so it must be confidential if it contains confidential information indide because we don't really have a way to enforce authorization in the LLM. That is why Rag it is native to AI workloads and it is not going any where. Actually RAG it is more similar to how we as human work, we do not know everything, but we have formats and supports to encode information and we have generall abilities and tools to access those information supports and decode and understanding the information in there and use that to solve problems.

[@LuisSanchez-gp4rq]

didnt he say its open source?