Visual Prompting Livestream With Andrew Ng

[@WilliamDye-willdye]

Thank you, Dr. Ng, for making so much of your work freely available.

[@williamwei3003]

Awesome presentation, Thanks, Dr. Ng.

[@ramzirebai3661]

Thank you so much Mr.Andrew Ng for all your contributions to the ML community I have benefited from all your courses, Thank you so much

[@jacksonlim7607]

As usual, Dr Ng enables learning interesting, enjoyable and useful. Awesome.

[@MarioLopez-eu8tj]

Finally a direct channel to Andrew Ng.👍

[@kennette2121]

Thanks for a wonderful live stream and such a concise depiction of visual promptings value. I've already built a proof of concept that is an imitation of a model that took us a team of experts and months to achieve! So exciting! As a developer, I would love to see more explicit documentation about the API, some pointers for bitmap decoding, and some simple tips how to leverage the returned bitmaps...

[@kavinduteknika9315]

Thank you Dr.Andrew Ng for sharing your amazing work. This seems a great tool...

[@yilei1051]

I see the examples are more about labeling and pattern recognition. I think this type of tool would be really helpful for text documents too. People would frequently format one section of the doc and want the rest of the document be done the same way, but LLM today cannot do that automatically yet. It would be really helpful to have such a pattern recognition + reapplication tool for everything.

[@KG-iy5ll]

Andrew, where is your NgGPT? I think you are just using meta SAM

[@Damanpreetkaur94]

This is great work! Although I am a bit confused, how is this different from Interactive Image Segmentation?

[@gusmein5144]

Very useful idea Sir. Is it possible to "after labeling and then counting the amount of colonies in the image using visual prompt? Please make a video counting object after labeling images using visual prompt. Thanks Sir.

[@davedorn6764]

Thank you! This was an excellent explanation for a very useful tool.

[@fgfanta]

Andrew Ng saw the democratization of AI coming, along with data-driven AI, so if he now talks about text/visual prompting, I listen. (Pity the audio quality of this video is bad, though)

[@TheVetzzang]

Thanks a lot! Dr. GREAT Lecture

[@erniea5843]

So basically the ‘in context learning’ approach used for LLMs but for CV?

[@XiaoZhao-d4j]

awesome!!

[@gogakakabadze]

Thank you! It was impressive!

[@anilraghu8687]

Tell us what has been happening between when you released your course and chat GPT.

[@SuperYutubu]

This is amazing

[@taopaille-paille4992]

Dr Ng was more relaxed and articulated during the COursera courses

[@waqaskhalidobeidy2673]

Thanks Prof

[@FuzzerHash]

Thank you

[@حواءلكلالاسرةمنوعاتوطرائفمنكلم]

تكنولوجيا حديثة جدا 👍💕💕💕💕

[@Neuronalchannel]

Awesome!

[@animesh8206]

Thanks

[@neuralDEEPnet]

Super !

[@cyrusazamfar6220]

this technology is literly 7 years old! I remember Amazon released something identical with quite similar UI while back.. like way back...

[@fintech1378]

Whats the name

[@davidw8668]

But you couldn't set it up in a notebook on a rainy sunday afternoon on your own 😉 Without a lot of specific data, labelling, and complicated training, it would have been impossible and took you month even if you had the data and the know how - that's what he mentioned in the beginning.

[@PazLeBon]

@@davidw8668 sooo, just 6 years 11 months behind then

[@fintech1378]

​​@@PazLeBon doesnt even matter, some key features were not available in the past, making it hard to use..for example you can say neural nets were already available decades ago

[@lorenzoleongutierrez7927]

good application, but there is no technical documentation about Deep Learning models are used on it...sort of black box ...strange on Andrew Ng

[@maverick456-33]

I want to know why you never uses high quality videos even in his lecture courses

[@R1l2eiy589]

Hello from china

[@ernieobrien7733]

'PromoSM' 🎶

[@boscho1986]

Ok, let's assume improvements are going to go on for a while and it becomes so easy to build models, that virtually a 10 year old could do it. Then why should I be interessted in this field? Take a stylus mark some class-representatives and by magic things work. No need to understand backpropagation, regularization whatever. The magic is going to be developed by some whizz-kids under the guidance of people like Mr. Ng, Sutskever et al. - and that's probably a good thing. But nobody, not having the potentials of one such whizz-kid, needs to learn anything about AI, like nobody needs to learn about aerodynamics except for a few hundred engineers building air-planes. Time to leave the field?

[@geoffreyanderson4719]

A prepared model is actually a small but critical part of a useful application! When i look around in daily life, so many times i say to myself, why aren't they using ai here? These are our opportunities!

[@PazLeBon]

thats just logical. in fact, universal incomes now seem a probability within a decade. why do anything? :)

[@kennette2121]

This is only the natural progression of all technology, no? I studied computer science at school, learning the details of computer networking and basic app building. Then I worked as a Full Stack web/mobile developer for 5 years learning the ins and outs of you build and serve software to millions of users to, in the end, fulfil some desire of the user. The technology we build sovles end user's problems. Every year a tool comes out that abstracts some of my work as a full stack developer away. Weather is seamless build tools like Vercel, backend as a service providers like supa & firebase, auth providers, the list goes on, but ultimately, they enable me to move away from the technical complexity and closer to the user problem solution. As you say, I think it's better to let a small field of ultra focused professionals work on these tools which abstract technical complexity away. The masses should use these tools, the collective human technology, to augment their goals. For me, its to build better, more intuative, and practical software that quickly addresses user problems and evolves with the domain. I still do full stack work, but a lot of it is on the conceptual architectural side. Tools help implement these concepts many many times faster. There is still room in this space for you to take and run with this technology if you want to stay AI focuses. Maybe not in building the models, but in making genuine applications of them! Instead of building 1 or 2 models a month, you could build 1 a day! Imagine how many novel usecases & applications you could discover! Old ways of work are going to be eclipsed by this (as with all other tools), but the reflected light illuminates potential previously unknown! So no need to leave the field, explore the aspects hither to unknown!

[@SuperMyckie]

Yea.. i left language model already.. spent a good part of last year v learning transformers and attention algorithms..

[@Thefare1234]

At this point I’m just learning it because I like it. There is no hope of being employed in this field in a few years.