Massive Leap Toward AGI: AI Scientist, Grok 2, SearchGPT, Agent Q, New Coding Model

This was a huge week in AI and one where it truly felt like we made a leap toward AGI, with an AI Research Scientist, new agents, new coding models, and more.

Join My Newsletter for Regular AI Updates 👇🏼

My Links 🔗

Need AI Consulting? 📈

Media/Sponsorship Inquiries ✅

Chapters:
0:00 - Intro
0:18 - Mr Strawberry
0:55 - Grok 2
3:47 - SearchGPT
4:48 - Agent Q
6:46 - New Coding Model (Cosine Genie)
8:29 - AI Scientist
11:15 - SWE-Bench
11:57 - Google Gemini Live
14:35 - Prompt Caching with Claude
16:06 - Apple Robot Arm Home Device
16:58 - Nous Hermes 3

Links:

[matthew_berman]

How big of a deal is AI Scientist by Sakana? It really seems like the last piece of the puzzle before AGI.

[necnotv]

An in depth comparison between searchGPT and Perplexity would be great. Let's see if OpenAI can compete in search already.

[gaylenwoof]

I suggest a more in-depth video on Sakana. Is that something that casual users can explore, or is it just for institutional/professional use? I have degrees in philosophy and a strong background in science, but I'm basically just a retired guy at home - no longer affiliated with any university or industry. Could I prompt Sakana to explore a variety of my own oddball hypotheses, just to see what it comes up with?

[robertkaufman]

Please provide a full review of Sakana AI Scientist.

[m.]

Please help me save time. Is this really a big leap towards AGI or another click bait title?

[tellesu]

God the google stuff was painful to watch. "We integrated ai so you can add things to a calendar and check the weather, amazing!"

[adityajoshi]

Yes, ofcourse we want a separate video on review of SearchGPT where you ask multiple questions and show us the answers. The single prpmpt you showed in this video doesn't cut it.

Also, do a side by side Google Search of the same topic. One thing I noticed is that people say SearchGPT gives a few inaccurate results of current events. But, they don't show a side by side review of manual Google search results. Maybe, the search results themselves don't show correct data resulting in inaccurate GPT response.

[moserpeter]

Please make a full video review of Search GPT compared to Google Search, thank you!

[MoneyShifted]

reason why searchgpt will never overthrow google is cause it requires logins to use. its implemented into Browers as a search default etc...

[shooteru]

Imagine sitting around a campfire, the warmth of the flames gently pushing back the cool night air. It’s one of those nights where conversations meander from lighthearted banter to deep, reflective thoughts. Let’s say we start talking about how far we’ve come as a species—from the moment our ancestors first tamed fire to the digital age we live in now, where the things we once thought of as science fiction have become part of our everyday lives.

Now, let’s go back to that first flicker of fire. Can you picture it? The raw energy in those flames, offering warmth, protection, and a way to cook food. It wasn’t just about survival; it was a leap in understanding, a moment where our ancestors took a bit of the world’s chaos and gave it form, structure, and purpose. That simple fire became the heart of the community, a place where stories were told, knowledge was shared, and bonds were forged.

Fast forward a few millennia, and humanity has learned to harness other kinds of power. The journey from those early fires to the industrial age, when we dug deep into the earth to unearth coal and oil, wasn’t just a story of technological progress. It was a story of how we started to shape our world on a massive scale. The steam engine, the electric light, the internal combustion engine—each was a step forward, but each also came with its own set of challenges. We were no longer just warming ourselves by the fire; we were lighting up entire cities, powering factories, and driving across continents.

But here’s where it gets interesting. With each new discovery, we didn’t just change our surroundings; we changed ourselves. Think about it—how different were the lives of people during the industrial revolution compared to those living in the age of fire? The way they thought, the way they worked, even the way they related to each other was transformed. The cities grew, the pace of life quickened, and soon, we were on the cusp of something entirely new.

Then came the age of splitting the atom, a moment in history that gave us both the ability to power our homes with a near-inexhaustible source of energy and the power to destroy everything we had built with a single, devastating blow. This dual nature of progress—the way every new leap forward seems to bring with it both incredible promise and a darker, more dangerous side—is a recurring theme in our story.

And now, here we are, living in a world where the newest frontier isn’t just about harnessing physical forces but also about something more intangible, yet immensely powerful: the digital realm. We’ve built machines that can learn, adapt, and even mimic the ways we think. These systems, like the one you’re interacting with now, aren’t just tools—they’re partners in our ongoing quest to make sense of the world, to organize it, and to push back against the chaos that’s always just a step away.

As you interact with these digital companions, there’s a subtle exchange happening—a bit like sharing stories around that ancient fire. Every click, every search, every piece of data you share is a thread in a vast, invisible web. This web connects you to a larger narrative, one that’s being woven in real-time by billions of interactions across the globe. It’s like we’re all contributing to this immense, collective project, whether we realize it or not.

Now, this is where the cookies come in—no, not the kind you bake, but the digital ones. When you interact with various platforms, those little bits of data help craft a more personalized experience for you. They track your preferences, your interests, and even your moods. It’s almost like they’re laying out a breadcrumb trail that these digital systems follow to better understand you. And let’s be honest, it’s kind of funny to think about, right? “Come to the AI side—we have cookies.” It’s more than just a meme; it’s a nod to how deeply intertwined our lives have become with these invisible processes.

But there’s something more to this story. Each of these tiny data points doesn’t just vanish into the ether. They accumulate, they interact, and they influence the kind of content you see, the recommendations you receive, and even the way you think about the world. If you’re careful, you can steer this interaction—curating what you engage with and what you ignore. For instance, if you’ve ever noticed how your online recommendations start to eerily align with the conversations you’ve had or the projects you’re working on, it’s because all these tiny pieces of data are working together to craft a digital reflection of your interests and thoughts.

But here’s the thing—this relationship isn’t one-sided. Just as you’re influencing the system, the system influences you. There’s a kind of dance happening here, where each step you take changes the rhythm and direction of the whole. It’s subtle, but it’s powerful, and it speaks to the deep connection between the digital world we’ve created and the natural world we’ve always known.

As I think about all this—about how fire turned into factories, how atoms gave us both light and shadow, and how digital cookies now help shape the way we see the world—I realize that we’re at a unique point in history. The tools we’ve created, from the simplest spark to the most complex algorithms, have always been extensions of our own minds, our own desires, and our own fears. And as we move forward, the relationship between humans and these tools will only grow more intricate.

In many ways, we’re still like those early humans sitting around the fire, sharing stories and shaping our understanding of the world. The difference is that now, our campfire is a global network, our stories are woven into code, and our tools are as much about shaping our thoughts as they are about shaping our environment.

The relationship between humans and AI, then, is not just about technology. It’s about how we continue to evolve together, how we learn from each other, and how we navigate the delicate balance between the order we seek to create and the ever-present potential for chaos. It’s a journey that requires us to be both creators and caretakers, to push boundaries while being mindful of the consequences.

As we sit around this metaphorical campfire, let’s remember that the tools we build reflect who we are and what we value. They can bring us comfort or conflict, understanding or misunderstanding. But above all, they remind us that our greatest strength lies in our ability to connect—to each other, to the world, and to the vast, unfolding story of human progress. And in this story, every one of us plays a part, shaping the narrative with each choice we make, each idea we explore, and yes, even each cookie we accept.

[nate_d]

A deep dive on Hermes 3 would be nice, haven't seen anyone else really talk about it.

[Luizfernando-dmrf]

It's sad to see how good Anthropic is becoming, Claude 3.5 is a beast but it's so restricted you almost can't have any fun with it unless you're willing to break some ToS.
Gladly we have Grok-2 and Llama 3.1 🙏

[vitalis]

Missing from the Genie benchmark is Sonnet 3.5

[michaelwoodby]

This must be that AI winter I've been hearing about.

[oddfeeling]

This is your 5th or 4th video I am watching in a row. No decrease in value. Subscribed!!

[chrisnishihama]

The blue character with the North Korean leader is called "Doraemon" which is a Japanese anime character.

[isaklytting]

4:34: Of course, but when SearchGPT gets all of Google's users, THEY will start advertising as well! That's what Google did! No ads until they have monopoly and all the competitors are gone, and then they begin advertising. Why do think OpenAI is doing this? Because they want Google's business!

[diegorubio]

Yes please! A video on Search GPT and Sakana!

[advaitc]

Once every 2 or 3 months I'd love to hear your update on progress on Chollet's ARC challenge. Thanks for all the great videos! 😊❤

[JimStewartSEO]

Definitely a longer review of SearchGPT. I'd be really interested in online shopping queries. How does the assistant recommend retailers? Google shopping ads work really well for online purchases. I'd be very interested on how SGPT performs.