Join us for a continuous exchange of ideas about research in the era of general AI. *Register for the Microsoft Research Forum series:* aka.ms/ForumRegYT

I remember the 'sparks of AGI' paper struggling to be taken seriously due to lack of evidence... Not sure that was regarded as 'edgy'. Still, GPT-4 is still being explored with novel prompting strategies. We will have GPT-5 before GPT-4 has fully been exploited, such is the pace of progress... and the open source community may even leap frog that! Such exciting times.

@medoeldin

4 ай бұрын

Sebastian said that Sparks of AGI was not intended to be scientific. It was a early window into capabilities of GPT4. It should be taken for what it is.

@alertbri

4 ай бұрын

​@@medoeldinI think at the time it was regarded as hype. Scientific papers are churned out like blog posts these days I guess.

@medoeldin

4 ай бұрын

@@alertbri I can see that but if u watch the video at MIT he was clear about what it was.

This was one of the best presentations Ive seen in a long time. Can we get more videos from this guy?

Thank you Peter Lee for this important keynote

Thanks to youtube and Microsoft for this ❤❤❤❤

I've been blown away what I can get Chat GPT to do through the API, it's got a good enough representation of reality to be incredibly practical for logical reasoning, like he said, get the context right and you can both generate and condense meaningfully 😀

Thats beautifully done from Peter Lee, the history about the disruptions in biology in the beginning, even if that doenst directly relate to AI, the way he tells made it very interesting and engaging. Very different from other companies that organize AI seminaries online just to sell their products or advertise some "new AI feature" no one cares about, we need companies really compromised in creating a engaging comunity in scientific field, specially AI, and they have to be clear that this may not generate profit at short term, but is a fundamental step if you want the name of your company in the future of AI and science. Keep it up Microsoft

Thanks for this video. It is interesting to observe that the distribution of research papers in top tier conferences (e.g. NeurIPS) by institutions doesn't significantly change from year to year. This stagnant pattern raises questions about how truly accessible and inclusive the field of AI research is.

Research in AI is a noble pursuit, indeed. It will advance the knowledge and welfare of humanity in general as it can produce novel and useful artifacts and systems. The importance of it should not be underestimated.

📝 Summary of Key Points: 📌 The speaker discusses the historical context of scientific disruption and draws parallels between past disruptions in biology and the current disruption in AI. 🧐 The speaker highlights the collaboration between Microsoft Research and OpenAI, specifically the development of the GPT4 model, and the importance of responsible AI. 🚀 The speaker discusses various research advances in AI, including open-source models, model specialization, and the impact of AI in healthcare and medicine. 💡 Additional Insights and Observations: 💬 "This disruption affects all areas of computer science and presents both challenges and opportunities." 📊 No specific data or statistics were mentioned in the video. 🌐 The Sparks Paper publication showcased the cognitive capabilities of the GPT4 model and sparked collaborations with other researchers. 📣 Concluding Remarks: The speaker expresses optimism about the future of AI research and emphasizes the importance of collaboration, openness, and responsible AI practices in shaping a bright future. The video highlights the historical context of scientific disruption, the collaboration between Microsoft Research and OpenAI, and various research advances in AI, particularly in healthcare and medicine. Generated using TalkBud

The a-hls at my lab layed me off as soon as I started using AI to program and write papers.

Thank you sir

We enjoyed it, thanks❤

Astounded the views are under 5K, 2 hours after publishing. Wonder how much more it'll pop off...

I really like OpenAI's methodology of deploying specialized GPTs which "confer" among one another to arrive at a solution, versus an AGI. I have a feeling these models can be backdoored. As proof of this, I present you with the fact their GPT-4 model can speak Klingon, a totally fictional language only found in a Star Trek movie. Just as a child may mistake unsuitable behaviors witnessed in film as applicable knowledge in unrelated contexts, may easily misconstrue learned rules for witnessed rules of behavior.

tldr: The future is incredibly bright

😮

real.

Terrific talk,but Dr.Lee should learn more synonyms of tremendous.😅😅😅

I don't care who makes this happen before me but, I'm on a mission to build an algorithm, framework, architecture, whatever you want to call it to be able to test while we sleep, for cures in medicine. I've been taking smaller steps to get to the point, to do so. Anyone interested, let's chat.

hmmm

If big is better then small, why not let big analyze small?

Safety this safety that. Guys just announcing and primitive scaled AI had led to many thousands lives being disrupted destroyed by the massive job layoffs. So it makes me wonder if your rhetoric on safety is just couching the language of secrecy, so that you can get a competitive advantage. After all, you know what they say about the road of good intentions.

Chat GPT says it can take between 8 and 15 years to build up the skills needed to join MS Research, seems doable, wip.

I thought he was a bot

That smells like mea culpa, for when they look at everyone's heads in a bowl. And, Microsoft lost the biggest contribution made in the medical field, called alphafold, by deep learning.