Відео

Knowledgeable as always! Thanks for bringing this to us!

Awesome, thank you for sharing! I missed this one

This is an interesting discussion on how to continue to improve model quality without human input. Thanks for hosting Sophia!

Interesting Content!

Interesting discussion on a really important topic!

This was probably my favorite one so far! It’s packed with a lot of info. Thanks for bringing this to us!

The hits just keep on coming! How do you get such great speakers?

Thank you for sharing the recording!

Great contents!

Timestamps: 0:00 Introduction 1:19 AGI by 2027? 3:24 The definition of AGI 5:36 Current AI systems are already improving AI research, accelerating the path to AGI 7:54 How deep learning works and why LLMs are not just "next-token predictors" 9:21 How to solve the long-term memory problem 10:46 What is the current recipe to achieve AGI: more compute, more data, more talent? 12:24 Compute is the gold of the 21st century 13:42 The role of academia in the AI race 15:22 Bringing energy supply to enable AGI 16:38 How do we know when it’s AGI? 22:26 If we hit a plateau with compute, how would that affect AI system development? 24:20 AI safety concerns 26:23 The main risks of AI (the deceptive nature of AI) 32:44 Instrumental convergence 35:03 The paperclip maximizer theory and how AI could harm humans

Thank you for sharing the recording Sophia! Nice job on the intro :)

Insightful discussion on the effect of context size and document ordering on RAG accuracy. Thanks for hosting this excellent talk, Bowen and Sophia!

Great discussion! Thanks for organising!

Thank you for sharing the recording! I'm curious how these models can become cheaper to train.

Interesting!

$2,000 is still a lot...

How to create a Stable Diffusion model with a low budget! Very useful! Thanks for sharing, Sophia!

Interesting. I presume the cost to do so is going to come down too as times progress. Great talk as usual, thank you for uploading this!

I saw in the paper that you used a ~7b LLM for experiments. How sure it is that the RLHF components aren`t relevant for way bigger LLMs?

The models I use call themselves a tool , not oppressed individuals

It already is on the political agenda in Europe, see the recently adopted regulations of the EU AI Act, for example

"AI" is as of yet (like the LMMs) is nowhere near intellegent, nor actually artificial (seeing the biasing and data labeling problems). A lack of understanding by those pro-AI people is betrayed, in them thinking current (LMM) AI is anything more than a very fancy database interface in which language is stored in a novel way that (somewhat randomly) tries to guess the next sets of letters, based on some prompt. Let alone AI being near sapience or sentience. Debate on those two aspects shouldn't be held yet, but rather we should talk about current "AI" "achievements". What we have now is rather "Virtually Intellgent" a VI, if not just some older scripts that've been slapped with an "AI✨" sticker for marketing purposes. To present this as a binary issue is at best an accidental misinterpretation or at worst misleading polarisation. I for one, encourage the use of "AI" for (medicial) research, but not current attempts to rid artists of their creativity and work. The EU AI Act, with its framework for determining how impactful a piece of "AI" might be, and its prescribed methods of dealing

So people who are pro-AI are anti-human? Yeah, those aren't the camps...

Another binary non-issue, like the pro-TikTok and anti-TikTok camps. This should not become a political issue.

Thank you so much for bringing such informational talks straight to our devices. They’re super knowledgeable!

This is such exciting work.

Great talk on Swarm, a framework for rapid prototyping of agentic systems. Also, a good demo and introduction of Cursor. Thanks for hosting Sophia!

I suppose the government will not view this as an opportunity to let citizens do whatever they like. Instead they’ll find some place to employ your labor; bolster their military, re-build infrastructure across the nation…after all, you will still want to earn an income higher than those that are taking their sabbatical. The economy will reward those that continue to contribute to the nation through their labor, innovations, etc

As long as those large corporations who develop AI pay taxes properly... :)

AI helps to advance AI. Fascinating! Thanks for hosting Sophia!

One more step towards making humans obsolete. I'm in!

Interesting discussion

Lawrence seemed incredibly knowledgeable in his field. This was a solid talk, thank you!

Anyone else notice how the questions to ChatGPT vs DALL-E reveal different thinking patterns? Visual AI seems to unlock pure creativity, while text AI brings out their inner philosophers. 🤔 So much potential for supporting different learning styles! Thank you for the talk!

A fascinating discussion on children interacting with AI. Thanks for hosting Sophia!

I loved this talk. Thank you for sharing it with us!

Thank you for sharing the recording! I really enjoyed it!

Recent news of Australia banning social media for below 16 years (not sure if it’s true) is an interesting decision (I support it). It will be interesting to see how AI is regulated for children and will be good to see how things are balanced because on the positive side it can promote creativity but can have many negative consequences far worse than social media.

Loved seeing the prompts the kids were making for Dalle. They’re so creative,and it’s wild how quickly they’re learning to use AI, probably faster than most adults! They're mastering how to make prompts more fun and imaginative.

I’m sorry, your robots kept arming themselves and that didn’t raise quite a few alarms?

Insightful!

I want to to install ai scientist in cloud can help with this

That was a great talk, thank you for sharing!

Great to see the convergence of robotics and AI. Thanks for sharing, Sophia!

Another banger. Thanks for uploading the video!

Timestamps: 0:00 Introduction 0:23 High-level goal of robot learning 1:16 Impact of Deep Reinforcement Learning on robotics 3:46 What makes robotics special? 5:43 Challenge 1 in robotics: Data 10:38 How simulation can help advance robotics 16:07 Challenge 2: Usability of a robot 18:51 Importance of practice for robots 19:53 How large language models can help manage many robots 23:55 The viability threshold for robots 26:10 Why Alex has changed his research direction to AI safety 29:21 Q&A with the BuzzRobot community

Amazing to see scientific research accelerated by AI. What scientific discoveries will be made?

Along with these skills, I believe the most important skills to integrate in children even in the age of AI are values. Importance of money and time especially when information and tools are so easily accessible. We are observing the great impact of Social media (pun intended). We can expect the worse with AI, if people don't have their own values.

What a time to be alive!

Great video as usual!! Thank you for sharing this with us :)