Latest / In the Loop AI / THIS WEEK IN AI: Qwen 3.6 is Scary Good, OpenAI's Record-Shattering Fundraise, Oracle Layoffs

Transcript

THIS WEEK IN AI: Qwen 3.6 is Scary Good, OpenAI's Record-Shattering Fundraise, Oracle Layoffs · In the Loop AI

0:00
Ejaaz: China has officially leapfrogged the US in AI models.
0:04
Ejaaz: Qwen released two models, Qwen 3.6 Plus and 3.5 Omni, which have absolutely crushed
0:10
Ejaaz: Claude Opus 4.6 and Google's Gemini.
0:13
Ejaaz: Qwen 3.6 can basically do everything Claude Opus 4.6 can do, but two to three
0:18
Ejaaz: times faster. It's also dramatically cheaper.
0:20
Ejaaz: It's free. And it also is closed source, marking the first time that a Chinese
0:25
Ejaaz: AI model is not open source. This is a signal that China may have finally beaten the U.S.
0:30
Ejaaz: And we should be concerned about that. We're going to dig into that.
0:33
Ejaaz: In other news, Open Air has also raised the largest private round ever,
0:36
Ejaaz: $122 billion. We're going to get into all of this and more. But let's start with Qwen.
0:41
Josh: I think I need to reel it in a little bit. I'm not sure Qwen is crushing the
0:44
Josh: United States just yet. We still have frontier models.
0:47
Josh: So, like, do not panic yet. But there are signs pointing towards a future in
0:53
Josh: which China continues to gain pretty quickly on the U.S. So one of the most
0:57
Josh: noteworthy things she mentioned is the fact that this model is closed source.
1:01
Ejaaz: This is new.
1:02
Josh: This is different. That we always said when we would discuss Chinese models
1:05
Josh: that the biggest threat we'd know when they're getting close,
1:09
Josh: when they start closed sourcing these models.
1:10
Josh: And clearly they're getting close. They've closed sourced the model.
1:13
Josh: Also, this model is available for free for a limited time, but totally available for free.
1:18
Josh: And it's Opus 4.6 or close to 4.6 level intelligent, right? We're looking at these benchmarks here.
1:25
Josh: It looks like it beats it in some things. It's a little less proficient in others.
1:30
Josh: And again, these are the Chinese benchmarks, so we're not entirely sure how accurate this is.
1:34
Josh: I suspect it is inaccurate. I suspect it's actually largely a distillation from
1:38
Josh: Opus 4.6 and other frontier models.
1:41
Josh: But yet the results do speak for itself and they are very impressive, right?
1:45
Josh: Like particularly the speed in which they're able to generate high quality tokens.
1:48
Josh: It's really, it's a pretty powerful model.
1:50
Ejaaz: Yeah, so it's about three times faster than Opus and much faster versus Gemini.
1:55
Ejaaz: It also has a 1 million context window, which now matches the American frontier models.
1:59
Ejaaz: That's usually typically harder to pull off because it helps give your AI model
2:03
Ejaaz: understanding and memory of what you're talking to it about.
2:05
Ejaaz: And the reason why I would say the benchmarks are probably more accurate than
2:10
Ejaaz: previous benchmark releases is they've been kind of honest about it.
2:14
Ejaaz: So if you look on the left here, Terminal Bench 2.0, which is a regarded benchmark
2:18
Ejaaz: for coding in particular, it surpasses and beats Opus 4.6.
2:23
Ejaaz: But if you look at software engineering as a whole, they admit that Claude Opus
2:28
Ejaaz: 4.6 is still better than them.
2:30
Ejaaz: So I don't know. I think this is a point.
2:33
Ejaaz: This coupled with the fact that
2:34
Ejaaz: they of closed source indicates that China might have the edge over here.
2:37
Ejaaz: Now, for those of you who are confused why open source and closed source is
2:41
Ejaaz: a signal that China is getting better,
2:42
Ejaaz: The idea with Chinese AI models was they build an amazing model and they open
2:47
Ejaaz: source it so that it decreases the value, price, and stock equity share of Frontier
2:53
Ejaaz: American Labs, who are typically private and don't open source their models.
2:56
Ejaaz: It also gives an opportunity for anyone and everyone to get access to Frontier
2:59
Ejaaz: intelligence and build an amazing AI product.
3:02
Ejaaz: The fact that they are closed sourcing it signals to me two
3:05
Ejaaz: things one they've figured out a way to leapfrog
3:08
Ejaaz: america maybe not on in this particular model but in
3:11
Ejaaz: the next version model that they release and they don't want that proprietary
3:14
Ejaaz: information going out and two i don't think they need to rely on american models
3:18
Ejaaz: much in terms of copying the types of designs or research that they do so i
3:23
Ejaaz: think they finally caught up it might be a hail mary it might be incorrect but
3:27
Ejaaz: i think this is another deep seek moment
3:29
Josh: Yeah, they're doing really well. And this is only one of two models that were
3:31
Josh: released this week that were very impressive.
3:33
Josh: The second one being Quen 3.5 Omni model, which does video audio vibe coating.
3:40
Josh: So this is multimodal vibe. Can you explain to me what's happening here?
3:43
Ejaaz: Okay, instead of explaining, let me just explain what's happening on the screen.
3:47
Ejaaz: For those of you who are listening, I am staring at probably a drawing that
3:51
Ejaaz: looks like a two-year-old's drawn it.
3:53
Ejaaz: It's of a website, but it's just a bunch of boxes that someone has sketched out.
3:58
Ejaaz: And it's a video of him explaining what each of the boxes should be.
4:02
Ejaaz: He's saying, hey, there should be a video of someone explaining my product.
4:07
Ejaaz: And then on the right here, this particular box should have some text explaining how my product works.
4:11
Ejaaz: And then the box in the bottom left should be the pricing and allow people to
4:15
Ejaaz: kind of purchase my product.
4:17
Ejaaz: But literally, what we're looking
4:18
Ejaaz: at right now is a marker pen with boxes around it. This does not exist.
4:22
Ejaaz: This is on paper. This is a physical thing and now what you
4:25
Ejaaz: see is him feeding this into this new model called quen 3.5 omni and it understands
4:31
Ejaaz: everything he's saying it understands the vision that he has through his sketch
4:35
Ejaaz: and it codes up a fresh website from scratch in seconds it's called audio visual
4:41
Ejaaz: vibe coding and this is the first instance of this ever happening
4:44
Josh: I will say, perhaps the models aren't always better, but the demos are always pretty strong.
4:50
Josh: And I think the demos really carry a lot of weight when it comes to evaluating
4:53
Josh: models, because if you are given a clear use case that a model is particularly
4:57
Josh: good at like this, I mean, that's a really fun selling point and a really compelling selling point.
5:01
Josh: There's one other demo from this model that I thought was pretty impressive, right?
5:05
Ejaaz: Yeah. So it's an omni model, which means it's not just visual, but it's also audio.
5:11
Ejaaz: It contains other types of mediums and in this particular demo an
5:14
Ejaaz: issue that people have is they like to speak with their
5:17
Ejaaz: ai model but sometimes when they're agreeing with the ai model or
5:20
Ejaaz: they say something the model interrupts or thinks it's getting interrupted and
5:24
Ejaaz: stops giving you the information that you want it becomes really annoying they've
5:28
Ejaaz: developed this type of model where you can interrupt it and it knows when you're
5:32
Ejaaz: actually trying to interrupt it and when it can just continue so i'll show you
5:35
Ejaaz: a little clip over here could
5:37
Josh: You give me a brief overview of this file please.
5:41
Ejaaz: This document primarily discusses the Quen 3 series of models.
5:45
Ejaaz: Its most prominent feature is the ability to autonomously decide whether to
5:49
Ejaaz: respond quickly or engage in deep reasoning.
5:51
Ejaaz: So there's a few things I want to point out here. Number one,
5:55
Ejaaz: he's... You can't explain why this is important because that
5:57
Josh: Seems so lame. It's like, obviously, it's not answering.
5:59
Ejaaz: Okay, this is actually really cool. Okay, so there's a few things going on here.
6:02
Ejaaz: Number one, he is scrolling through a research paper on his laptop and he's
6:07
Ejaaz: holding his phone, which contains this model. and he's got the camera feed on.
6:11
Ejaaz: So he's showing the pages that he's scrolling through.
6:15
Ejaaz: This model is reading this entire paper as he scrolls through,
6:19
Ejaaz: so much faster than a human can.
6:21
Ejaaz: It's digesting and understanding what it's reading and then it's summarizing
6:25
Ejaaz: all the key points to him.
6:27
Ejaaz: But as he's listening to her speak about what it's seeing on the screen,
6:31
Ejaaz: he's saying, oh, I see. Oh, that's interesting.
6:35
Ejaaz: Now, typically when you have a human conversation with someone and you're making
6:38
Ejaaz: those same comments, the human doesn't
6:40
Ejaaz: stop talking and telling you the information that you want it to hear.
6:43
Ejaaz: But this model is smart. It understands that he's not trying to interrupt it
6:46
Ejaaz: and it just continues flowing and speaking smoothly, which I think is a breakthrough. It's super cool.
6:52
Josh: Yeah, very impressive model. And this isn't the only news out of China.
6:56
Josh: Also, there's another team.
6:58
Josh: Okay, this team, I'm trying to figure out how to pronounce this.
7:02
Ejaaz: Say the name, Josh.
7:02
Josh: So it says Zipu, but I'm aware that it's not actually pronounced Zipu.
7:07
Ejaaz: It is cheapo
7:10
Josh: Jipu so jipu okay so yeah so china is now good enough at building models that
7:16
Josh: we're learning their dialect because we have to reference it that frequently
7:19
Josh: so that's probably a high signal if we're having chinese lessons on the show
7:22
Josh: from ijaz then china's getting pretty good what makes um this new jipu released
7:27
Josh: model uh so impressive here so.
7:29
Ejaaz: Glm which is their flagship model has kind of been similar to quen in terms
7:35
Ejaaz: of capabilities reasoning coding and when we looked at the Quen models just
7:40
Ejaaz: now, we saw it go from a simple sketch,
7:42
Ejaaz: a handwritten sketch, to a fully functioning web app, coded in real time in a couple of seconds.
7:49
Ejaaz: And you might think, ah, that's just one AI lab. They just got lucky.
7:52
Ejaaz: Who cares? We can crush them in the future.
7:54
Ejaaz: What you're seeing on your screen right now is another entirely different AI
7:57
Ejaaz: lab, which released their own model this week, and it does the exact same thing, sketch to code.
8:03
Ejaaz: So the point is, whether these Chinese labs are working together or not,
8:06
Ejaaz: whether they're training on the same hardware or they're not,
8:09
Ejaaz: they're able to achieve the same frontier research and breakthroughs around
8:12
Ejaaz: the same time, which is the reason why I think this isn't just a one-off.
8:16
Ejaaz: I think China has collectively caught up with the US and are posing a significant threat.
8:21
Josh: Yeah, it's very impressive what
8:23
Josh: they're doing. And this comes off the back of a lot of drama, I guess.
8:27
Josh: I was hearing rumors that some of the developers from Manus,
8:31
Josh: the company that was acquired by Facebook, were not allowed to leave the country.
8:35
Josh: And this is happening with a few other AI labs, I think Quen may be one of them,
8:40
Josh: or DeepSeek, where the developers are kind of, they've become government assets,
8:44
Josh: if you will, in a way that is a little unnerving, but progress is getting good.
8:49
Josh: So it's at the point now where we're still ahead.
8:52
Josh: I'm telling you right now, China does not have Capybara.
8:55
Ejaaz: Because if they did,
8:56
Josh: They'd probably be doing some serious damage. But they are making some serious progress.
9:00
Josh: And also, another company making serious progress is our good pals at Google,
9:04
Josh: who just as we started recording this release, Gemma 4.0.
9:08
Josh: This is a purpose-built model for advanced reasoning and agentic workflows.
9:12
Josh: On the hardware that you own, the thing that is optimized into Gemma is the
9:17
Josh: model performance per bit.
9:19
Josh: It's like, how much intelligence can you get per bit of information?
9:22
Josh: And according to this chart it's very very high.
9:24
Ejaaz: Yeah so um if we look at
9:27
Ejaaz: this chart it's showing that um gemma 4 which
9:30
Ejaaz: is the recent model is only a 30 it comes in two
9:33
Ejaaz: versions a 31 billion parameter model and a 26 billion parameter model so that
9:38
Ejaaz: is significantly small when you consider that capybara the the next anthropic
9:42
Ejaaz: model that you just mentioned is probably in order of magnitude higher it's
9:45
Ejaaz: five to ten trillion parameters so it's a much smaller model but it keeps up
9:50
Ejaaz: with some of the bigger models,
9:51
Ejaaz: specifically the Chinese AI models that we were just talking about.
9:54
Ejaaz: So if you look on the right over here, we've got GLM-5, Kimi-K 2.5,
9:58
Ejaaz: and Quen 3.5, which are each around 400 billion to a trillion parameters large.
10:05
Ejaaz: So the amount of intelligence per square density is not the right term,
10:10
Ejaaz: is really high with Gemma 4.
10:12
Ejaaz: And it's cool to see American Labs releasing open source models.
10:15
Josh: Yeah. So shout out to Google. We have another shout out this week for another
10:18
Josh: American company that goes by the name of OpenAI, you may have heard of them, they just closed.
10:24
Josh: The largest funding round in history. No one's ever raised more money than OpenAI
10:29
Josh: just did at $122 billion in committed capital.
10:34
Josh: It's important to note that this capital has not been inserted into a bank account.
10:38
Josh: This is committed capital at an $852 billion post-money valuation.
10:44
Josh: Now, there's some highlights here. Amazon is putting in $50 billion,
10:48
Josh: which is ironic because OpenAI a open ai
10:51
Josh: signed a 100 billion dollar aws deal nvidia is
10:55
Josh: putting in 30 billion dollars which is ironic because open ai
10:58
Josh: runs on nvidia gpus softbank is putting
11:00
Josh: in 30 billion dollars which is ironic because they're building stargate together
11:03
Josh: so there's a lot of this circular economic element to this um i mean 35 billion
11:12
Josh: of amazon's investment is just contingent on open ai achieving AGI or going public.
11:19
Josh: So this leads us to the natural conclusion. Well, this is this is their final
11:23
Josh: hurrah before they go public.
11:24
Josh: This is the last private round that we will ever see. This is the largest of
11:28
Josh: all time before their blockbuster IPO sometime possibly later this year.
11:32
Ejaaz: Yeah, exactly. And I don't know if the public is buying it.
11:37
Ejaaz: Bloomberg released a report this week, which claims that OpenAI shares have
11:42
Ejaaz: fallen out of favor on the secondary market.
11:44
Ejaaz: And can you guess which company, Josh, it's fallen in favor of?
11:48
Josh: Surely it's Anthropic.
11:50
Ejaaz: Ding, ding, ding, ding, ding.
11:51
Josh: The pretty cool new kid on the block.
11:53
Ejaaz: Yeah, although Anthropic and the Claw team have had a nightmare in security
11:57
Ejaaz: this week and done two very important leaks, the public doesn't care.
12:01
Ejaaz: They still want all of that Anthropic shares and equity, and they're selling
12:05
Ejaaz: or willing to sell OpenAI secondary shares in favor of this.
12:09
Ejaaz: But there is still some cool news from the raise, Josh. Did you know
12:14
Ejaaz: $3 billion of the $122 billion raised was available for individual investors,
12:20
Ejaaz: not retail investors, but individual investors to invest in.
12:24
Ejaaz: They were able to get exposure through their banks.
12:27
Ejaaz: Now, I don't know which banks. I don't know what the vetting process was like.
12:31
Ejaaz: But the fact is, OpenAI did not know who these individual investors were,
12:35
Ejaaz: but they offered up $3 billion to allow retail to get access to.
12:38
Ejaaz: But I think you needed to be an accredited investor. So it kind of sucked in that respect.
12:42
Ejaaz: But they kind of threw the retail audience who didn't get access to the $3 billion
12:46
Ejaaz: a bone by claiming in this latest fundraising round that ARK,
12:52
Ejaaz: invest which has several etfs which give you exposure to private ai companies
12:56
Ejaaz: they're going to add a bunch of open ai equity stock to that which i actually
13:00
Ejaaz: don't think is a good bit of news given the vcx drama from last week
13:05
Josh: Can we talk about that private fundraising around wait so who is available because
13:09
Josh: i never heard of this yeah um i had no idea that it was even possible for money
13:14
Josh: to be raised by individuals who are now on the cap table that's pretty cool
13:17
Josh: i would have love to have participated in that.
13:19
Josh: I think the ARK news here is interesting because the day that word came out
13:24
Josh: that secondary shares are having a little bit of trouble being liquid,
13:28
Josh: ARK decided to add those secondary shares into a public venture fund.
13:31
Josh: Like, oh, we've had enough of this now. We're done with this toy.
13:34
Josh: Now the public retail market can get access to it. And that's kind of what happened.
13:38
Josh: Perhaps the timing is ironic, but it's certainly a bit coincidental that three
13:43
Josh: of the new ARK Invest ETFs included these OpenAI shares on the back of the most
13:47
Josh: recent fundraising round and the rumor that it is becoming increasingly difficult
13:50
Josh: to offload these shares.
13:51
Josh: So the market, I mean, I don't want to say it's getting frothy.
13:54
Josh: It's just growing remarkably fast.
13:57
Josh: And it is showing some signs of trouble, I guess, in paradise for to some extent.
14:04
Josh: I mean, again, all of this is pretty warranted and justified,
14:06
Josh: but it is important to note that these are our commitments.
14:10
Josh: Again, like they did not close this money. These are commitments.
14:13
Josh: And we've seen what happens when commitments happen.
14:15
Josh: OpenAI has committed to buying a whole bunch of RAM that they're no longer buying.
14:18
Josh: And the price of memory the last couple of weeks has not collapsed,
14:23
Josh: but it has returned to...
14:24
Josh: Very high levels instead of unfathomably high levels.
14:28
Josh: So there is a lot of work that's being held in that commitment word.
14:33
Josh: But nonetheless, congratulations to OpenAI, $122 billion. It's more than anyone's
14:37
Josh: ever raised in history. And it's a really big deal.
14:40
Ejaaz: A lot of the capital was raised from the likes of NVIDIA, Amazon,
14:44
Ejaaz: and presumably all that money is going to go back to them in the form of compute.
14:48
Ejaaz: OpenAI president Greg Brockman went on an interview, I think like yesterday
14:53
Ejaaz: and basically said, we're focusing all our resources on compute.
14:57
Ejaaz: Compute and scaling laws are definitely intact. And he says,
15:01
Ejaaz: quote, if I could get my hands on all of the compute in the world and purchase
15:04
Ejaaz: it, I would do it because it's become a positive revenue cost center.
15:08
Ejaaz: So OpenAI thoroughly believes that more compute that they have equals to more revenue.
15:12
Ejaaz: And that trend isn't going to break anytime soon. So we'll see if that pays
15:16
Ejaaz: off. This is a pretty big bet.
15:17
Josh: Foot on the gas, up only, man. I'm still bullish. I am still very bullish.
15:21
Josh: I'm also bullish on Claude, which had a new update this week in terms of computer use, right?
15:26
Josh: Now it's like pretty much capable of doing anything on your computer for you.
15:29
Josh: This is OpenClaw, Claude edition, and it seems to be working pretty well.
15:33
Josh: I was actually playing around with it.
15:34
Ejaaz: It's really cool. It's good. It's good. Okay.
15:37
Ejaaz: So it seems to be every single week, heck, every single day,
15:42
Ejaaz: Anthropic has released a new feature and they've done that for the last,
15:45
Ejaaz: I believe, 18 days, which is just an insane feat for any frontier startup or company alone.
15:50
Ejaaz: But the point is, These features collectively together have automated a very
15:55
Ejaaz: important thing, which is the entire software engineering role.
15:59
Ejaaz: And I'm not dramatizing this.
16:01
Ejaaz: Previously, we had Claude Code, the most popular Claude product,
16:04
Ejaaz: arguably, earning billions of dollars for them every single month. But...
16:09
Ejaaz: It just generated the code. You would then need to launch the app that it had
16:13
Ejaaz: built, test the app out yourself.
16:15
Ejaaz: You would need to run a bunch of tests, give feedback to Claude,
16:18
Ejaaz: and it would still need or require a human in the loop.
16:22
Ejaaz: This update, computer use, completely removes it from you.
16:25
Ejaaz: And I actually summarized it in a tweet here where it basically says,
16:29
Ejaaz: Claude writes the code for you.
16:30
Ejaaz: Then it opens up the app that it coded. Then it clicks through the entire app
16:34
Ejaaz: and finds the bugs that it itself created. does the security testing and everything,
16:39
Ejaaz: then it fixes the bug and improves the app.
16:42
Ejaaz: Now, if you consider the fact that Anthropic also released a remote control
16:46
Ejaaz: feature where you can basically text code from your phone and you don't even
16:50
Ejaaz: need to be anywhere near your terminal,
16:52
Ejaaz: you start to see how you can just message it and come back a few hours later
16:57
Ejaaz: to a fully fledged app that has been reiterated,
17:00
Ejaaz: tested a bunch of times with real human feedback or agentic feedback.
17:03
Ejaaz: And you just have a award-winning product in a couple of hours or maybe even
17:07
Ejaaz: a couple of weeks. It's just insane.
17:09
Josh: It's cool. The qualification of everything has been a prominent theme of everything
17:13
Josh: that has been released over the last few weeks.
17:15
Josh: And I mean, we have new news from Manus as well, where they just rolled out,
17:19
Josh: remote phone control for your desktop.
17:21
Josh: So if you are running an instance of Manus, you can just text your computer
17:25
Josh: from the go on your smartphone and it will, again, take over your computer,
17:29
Josh: do all the computer use things that you expect from a product like OpenClaw,
17:34
Josh: except they're rolling it out themselves.
17:35
Josh: So there's a lot of this qualification. I think it's great.
17:38
Josh: This is like the transition into the operating system, AI first OS type progress.
17:45
Josh: And it's really exciting to see.
17:46
Josh: Now, on this agenda here, we have something that says Anthropic League 2. Is there another...
17:50
Josh: Anthropic leak. But what else has happened? We've been talking about Anthropic leaks all week.
17:55
Ejaaz: No, I just wanted to point out that like, it all looks really rosy with Anthropic,
17:59
Ejaaz: but they've just come off like a nightmare of a week.
18:02
Ejaaz: Two leaks in a week, the second one being the largest one that I'm showing on
18:06
Ejaaz: screen, the entire source code of CloudCode is now or was openly available.
18:11
Ejaaz: And I think it had by the end of the day, 90,000 forks, which means that it's
18:15
Ejaaz: living on at least 90,000 devices, which is just insane to think about.
18:19
Ejaaz: And if you want to understand how crazy this is and also the secret features
18:23
Ejaaz: and products that they're about to release that were revealed in the code.
18:26
Ejaaz: We created an entire episode about this. Definitely go check that out and watch
18:29
Ejaaz: it. We reveal all the details there.
18:31
Ejaaz: But the point being is Anthropic is kind of succeeding and winning in a lot
18:36
Ejaaz: of different ways, but they might be running too fast for themselves.
18:40
Ejaaz: They're building everything. I think 100% of the features that they build now is using Claude code.
18:44
Ejaaz: So it might also leave them up to exposure and weaknesses such as this.
18:48
Josh: I love it. move fast and break things man just get stuff
18:51
Josh: out the door if it doesn't work you'll fix it in hindsight i mean
18:54
Josh: sure this was a big l but then they automate this approval process
18:57
Josh: i saw boris the person who created a cloud code he actually addressed
19:00
Josh: this publicly on x and he said um it was a single person it
19:03
Josh: was an honest mistake they've built in implementations for that to no longer
19:07
Josh: happen again and they're just going to keep moving on and i really respect the
19:09
Josh: velocity and the transparency that they've been doing as they just get all the
19:13
Josh: stuff out the door velocity is such a big thing there's a company who has absolutely
19:16
Josh: zero velocity perhaps even negative velocity that we're going to talk about
19:19
Josh: next, which is Microsoft,
19:21
Josh: who has launched a product using Claude.
19:24
Josh: And when I first read this headline,
19:25
Josh: I was a little confused because if my memory serves me correctly,
19:28
Josh: Microsoft owns what, they're the largest single shareholder in OpenAI by an
19:33
Josh: incredibly large margin, and 31%, and that gives them access exclusively to all of the OpenAI IP.
19:41
Josh: They have internal access to all of the models. So please explain to me why
19:45
Josh: Microsoft, aside from them not making anything valuable in the world of AI,
19:50
Josh: why are they choosing to use Claude as their provider of choice instead of ChatGPT?
19:55
Ejaaz: I have a simple answer for this. It may sound biased, probably because I am a little biased.
20:01
Ejaaz: I think Claude is just better for certain use cases. I'm not going to say all
20:05
Ejaaz: use cases, but for certain things.
20:07
Ejaaz: This isn't the first product that they've launched that is powered by Claude.
20:10
Ejaaz: There is one product that they released two weeks ago called Microsoft Cowork.
20:15
Ejaaz: Does that name sound familiar to you, Josh?
20:18
Josh: Yeah, I've used Cowork and Claude for the last couple of months, actually.
20:21
Ejaaz: There you go. So they released an entire blog post talking about how they have
20:25
Ejaaz: managed to automate a bunch of desktop tasks.
20:28
Ejaaz: So basically, Microsoft's feature takes over your desktop and can do a bunch
20:32
Ejaaz: of things like operating your browser, opening up files, a bunch of cool stuff like that.
20:36
Ejaaz: And right at the end of the blog post, in tiny little print,
20:39
Ejaaz: it said, oh, by the way, this is powered by Anthropics Cloud Cowork Model.
20:43
Ejaaz: And now today or this week, they announced two new products.
20:47
Ejaaz: It's called Microsoft Critique and Microsoft Council. It is a deep research
20:52
Ejaaz: feature, which basically is now state-of-the-art in terms of research.
20:55
Ejaaz: You can ask Microsoft's feature, critique or counsel, a question,
20:59
Ejaaz: and it does a ton of research, really high quality, and the answers and outputs
21:02
Ejaaz: are very, very impressive. You can get access to it via co-pilot.
21:06
Ejaaz: But what's interesting is it's powered by two models, Claude and ChatGPT.
21:12
Ejaaz: And the way that it works is arguably the most interesting part.
21:15
Ejaaz: They spin up a bunch of instances of Claude and a bunch of instances of ChatGPT
21:19
Ejaaz: and get them to talk to each other.
21:21
Ejaaz: One produces the research, one reviews the research. By the end of the back
21:25
Ejaaz: and forth process, you end up with an amazingly perfectly polished research article or paper.
21:30
Ejaaz: And I actually think this is cool because it shows that running multiple instances
21:34
Ejaaz: of AI models and agents are actually a better way to have a better product.
21:39
Josh: You know, my favorite Microsoft news of the week, it just happened on the spaceship, the Artemis mission.
21:45
Josh: They were live streaming the live comms and they're using Microsoft Outlook
21:49
Josh: and all the instances crashed and they couldn't figure out how to get them to work or to open.
21:53
Josh: And they were trying to remote into them and the software is falling apart.
21:56
Josh: And I think that's a testament to kind of where Microsoft is right now in the world of AI.
21:59
Josh: It's just clunky, very slow moving and very corporate, I guess would be a good way of describing them.
22:05
Josh: Now, the next topic we're going to talk about, my BS meter is going crazy.
22:09
Josh: I'm like, there's no way this is right.
22:10
Josh: This seems like they're using a lot of big words. The two words that this company
22:14
Josh: has shrunk together is electromagnetic super intelligence.
22:18
Josh: Generally, when people speak like this,
22:20
Josh: It's veiled in a lot of complexity that might not be true.
22:24
Josh: Can you explain what on earth is going on with this electromagnetism story?
22:29
Ejaaz: Yes. Okay. So this company, or startup rather, has come out of stealth claiming
22:34
Ejaaz: to have built a new type of AI model. It is called an electromagnetic AI model.
22:41
Ejaaz: And what it does is you can describe the type of electromagnetic behavior that
22:46
Ejaaz: you want, and it will generate the physical shapes and structures,
22:50
Ejaaz: a high-fidelity design that will produce those electromagnetic signals.
22:55
Ejaaz: Now, you might be asking, who the hell cares? Why would I care about this particular model?
23:00
Ejaaz: Well, if you are building antennas, if you are scaling 5G and 6G,
23:05
Ejaaz: but most importantly, if you're designing AI chips in particular,
23:08
Ejaaz: which requires microscopic attention to detail shapes that allow these transistors
23:14
Ejaaz: to talk to each other, you need help in terms of an AI model describing that.
23:19
Ejaaz: Now, previous models, they tested this with Claude Opus 4.6 and GPT 5.4,
23:23
Ejaaz: can't get to the precise design because it doesn't actually understand how the
23:27
Ejaaz: physics of these AI chips and transistors works.
23:30
Ejaaz: Now, this company is claiming that their model does exactly that.
23:33
Ejaaz: And they had a demo chip, which they portrayed in their video or opening announcement demo.
23:40
Ejaaz: I don't know whether this chip works, but if it does, it might be able to automate
23:45
Ejaaz: the most important and valuable company in the world right now,
23:48
Ejaaz: NVIDIA, who spends all their money, NTSMC,
23:51
Ejaaz: designing all these next-gen GPUs and chips and gives them a valuation of $4.5 trillion.
23:56
Josh: Yeah, okay. So I guess it's one of those things where we'll see.
23:59
Josh: Like I imagine all the Frontier Labs know that this exists.
24:02
Josh: Reading the comments section, it appears as if this is not new research.
24:05
Josh: They've just actually acted on it and are starting to test it.
24:07
Josh: So one of the things that we will keep on the watch list of things to follow
24:11
Josh: along and see how they progress.
24:13
Josh: The next on our agenda list, and stick with us, we have three more topics to
24:16
Josh: go, is Oracle apparently fired 30,000 employees. What's going on here?
24:22
Ejaaz: It's exactly what it says. 30,000 employees, between 20 to 30,000 employees
24:26
Ejaaz: received an email two days ago in the morning, basically saying,
24:30
Ejaaz: sorry, you have been laid off.
24:33
Ejaaz: Now, there's a bunch of speculation as to why this is happening.
24:36
Ejaaz: Maybe something to do with the $300 billion partnership deal that they signed
24:40
Ejaaz: with OpenAI and a bunch of other semiconductor companies.
24:43
Ejaaz: Maybe it's something to do with that. Maybe they overstepped.
24:45
Ejaaz: Maybe it's something to do with Stargate's Abilene data center getting shut
24:50
Ejaaz: down because they weren't able to finance it.
24:52
Ejaaz: Oracle and OpenAI were the main partners in that, along with SoftBank.
24:55
Ejaaz: I don't know. How does this look to you, Josh? Is this a warning signal? Is the bubble popping?
24:59
Josh: No. Every company is grossly overinflated in terms of how many employees it has.
25:06
Josh: During COVID in particular, how many companies were hiring like 10,000,
25:10
Josh: like tens of thousands of people per year? So many.
25:12
Josh: And for no reason. And I think during that time when it became too okay to do
25:19
Josh: remote work and to switch away from in-person stuff, it was just very easy to
25:22
Josh: hire all these new people. Interest rates were zero. It was free money.
25:26
Josh: You can just, why wouldn't we double our workforce and get more productivity?
25:30
Josh: But now that the reality is setting in that they're really just,
25:32
Josh: and this isn't even as a result of AI. It's just there's not a need for that
25:36
Josh: amount of workforce in order to do these things.
25:39
Josh: People are just getting laid off. And that's kind of what's happening here is
25:42
Josh: there's just absolutely no need to keep hundreds of thousands of dollars on
25:47
Josh: the payroll that aren't being super productive.
25:49
Josh: We saw that with what company laid everyone off? Jack Dorsey Square.
25:52
Josh: Yeah, Jack Dorsey Square.
25:54
Josh: They laid off people. The stock went up 30%. 40% in one day.
25:58
Ejaaz: Wait, wait, sorry. They laid off 40% of people, right? And then their stock went up 30%?
26:03
Josh: I think the stock went up 40% too in the day. Like it was a crazy game.
26:07
Josh: It was this unbelievable thing. And I think the market is starting to reward
26:10
Josh: efficiency, right? It's like, it's obvious you don't need this bloated headcount.
26:13
Josh: A lot of people are not working. They're just getting by on their laptop.
26:16
Josh: A lot of them work remotely.
26:17
Josh: This is just a testament of this recalibration that's happening.
26:21
Josh: And I don't think you should really credit AI with too much of this because
26:25
Josh: a lot of these effects started happening before AI was even good at writing code.
26:28
Josh: Like AI just got good at writing code like three months ago.
26:31
Josh: It hasn't been that long. And companies haven't even really implemented this
26:34
Josh: in a meaningful way at scale. So I wouldn't worry too much about this.
26:39
Ejaaz: Josh, can I pull up the Oracle chart for you? Just to prove your point.
26:44
Ejaaz: I mean, it's nothing major, but it's up. It's up on the week.
26:46
Ejaaz: And this is after like a month of downtrend. So the market's obviously rating that as positive.
26:52
Josh: Nice good well that's a w um another possible w i don't know if this is a w
26:56
Josh: or not is the meta glasses news you know i'm the biggest hater of these classes
27:00
Josh: i'm actually i'm becoming the biggest meta hater ever okay because like meta
27:03
Josh: why am i calling you meta when you just cancel your meta division um but we
27:07
Josh: have new glasses news what's going on with these glasses okay.
27:10
Ejaaz: Okay listen let's not hype this up too much it is uh an optically friendly pair
27:16
Ejaaz: of their ai glasses So typically, if you had eyesight issues and you required
27:21
Ejaaz: prescription glasses, you couldn't use this. Well, guess what, guys?
27:24
Ejaaz: Now you can. But what excites me the most is some of the software updates which
27:28
Ejaaz: are included in this, which involves nutrition tracking, Josh.
27:32
Ejaaz: Now, you know me. I go to the gym quite a bit and I'm trying to figure out my
27:36
Ejaaz: calories so I can lose a few pounds before the summer, okay?
27:39
Ejaaz: Like, I'm just, I'm being honest.
27:40
Ejaaz: These glasses supposedly can see the things that you're drinking,
27:44
Ejaaz: that you're eating, that you're cooking and measure the amount of calories that
27:49
Ejaaz: is in the portion of your bowl or that you're making and can tell you when you've
27:53
Ejaaz: been eating too much or when you need to stop.
27:55
Ejaaz: Now, I don't know if the tech is actually accurate. I remember that there was
27:59
Ejaaz: a company which just got sold to, what's that fitness app, MyFitnessPal?
28:04
Ejaaz: It got sold to them. It was a vibe-coded app that basically you can show a camera,
28:07
Ejaaz: show your food in a camera, and it can basically estimate the calories.
28:10
Ejaaz: It got sold for upwards of $10 billion, I believe, or something crazy like that.
28:14
Ejaaz: It was acquired for like between two to ten billion dollars um this is a proving
28:19
Ejaaz: point that maybe something like this could be cool in the future i don't know
28:21
Ejaaz: if i want to wear meta glasses though to get this feature i would rather just use my phone
28:25
Josh: I don't believe them i don't believe a word that they say i think the last time
28:29
Josh: when they came out and they revealed these meta right band glasses they went
28:31
Josh: on stage and they actually had to demo it in real time in front of everyone
28:34
Josh: every single demo failed it was a nightmare and i just can't trust a company
28:38
Josh: that continues to fail over and over at all these promises that they share.
28:44
Josh: It's like our meta promise, our glasses promise, all these promises are just,
28:48
Josh: unfulfilled. And where's AI? They've spent billions of dollars.
28:51
Josh: There is none. So I'm remaining the biggest meta hater.
28:54
Josh: I am a Zuckerberg lover. I hope he can figure it out. He's an incredible CEO and an operator.
28:59
Josh: He's running a company that he founded and I'm so bullish on it.
29:02
Josh: But man, I just, I will not be participating in these glasses.
29:05
Ejaaz: Something is working. They said on their last quarterly earnings,
29:08
Ejaaz: they're on track to sell 20 to 30 million units of these this year.
29:12
Ejaaz: I don't know who the hell is buying that, but they have some kind of a market.
29:16
Ejaaz: And I do believe that glasses will be one of, not the, but one of the main form
29:20
Ejaaz: factors for AI. So we will see.
29:23
Ejaaz: But on our last news story, Valor Atomics raised a crazy round. What's happening here?
29:30
Josh: Yes, this is huge. If you made it this far in the episode, chances are you are
29:33
Josh: a real one. You are an OG. You have been listening to the show.
29:36
Josh: You love this show. And if you've been around for long enough,
29:38
Josh: you were here when we interviewed the CEO of Valor Atomics, Isaiah Taylor.
29:43
Josh: Since then, the valuation of Valor Atomics has gone up, I think, 27x.
29:49
Josh: And it has come to a point in which they have just concluded their $450 million
29:55
Josh: in fundraising at a $2 billion valuation.
29:58
Josh: Valor Atomics, for those who are not familiar, create modular nuclear reactors.
30:02
Josh: What you're seeing on screen is a little image of what this reactor looks like.
30:05
Josh: And it's just this tiny little thing you could put next to a data center and
30:08
Josh: power it using clean nuclear.
30:10
Josh: Reusable nuclear energy in this very beautiful package
30:13
Josh: they were the first private company ever to have a
30:16
Josh: successful version of nuclear energy they actually used
30:19
Josh: a radioactive isotope they generated energy from
30:22
Josh: it and things seem to be going very well so for those
30:24
Josh: of you who haven't seen that episode it is still timely it is still relevant
30:27
Josh: with isaiah taylor you can find it linked in the description below but just
30:30
Josh: a congratulations to the team at valor who are building some bad ass technology
30:34
Josh: and getting rewarded for it and i think it's just a nice way to wrap up the
30:38
Josh: week here is um with the big win for the people building the frontier of technology
30:43
Josh: and what's possible would.
30:44
Ejaaz: It have not been nice josh if we could have invested or even helped our audience
30:48
Ejaaz: and our listeners invest back then 27x
30:50
Josh: That's yeah it's pretty good yeah according to
30:53
Josh: claude at least the uh the limitless portfolio of when guests
30:56
Josh: have come on versus where they're at now has been pretty good we've
30:59
Josh: we've hit every single one all of the companies have done incredibly well
31:03
Josh: and um yeah it's just it's really exciting it's isaiah
31:06
Josh: is a great dude really ambitious really smart really talented along
31:09
Josh: with the entire rest of the team so a nice win to round up
31:11
Josh: our week with and i think that concludes everything for the week we've covered
31:14
Josh: a lot of ground this week there is so much stuff that happened
31:17
Josh: as it always does right we are on the frontiers of
31:20
Josh: the hottest industry in the world but if you've listened to all of the episodes
31:23
Josh: if you made it this far through the end of our final episode you're caught up
31:27
Josh: you can go touch grass this weekend you're done you could just like wipe your
31:30
Josh: hands clean you know everything that's going on go share it with your friends
31:33
Josh: share this episode with your friends if you enjoyed it and uh yeah thanks so
31:36
Josh: much for joining us for another amazing week.
31:38
Ejaaz: And I will just say, if you are listening to this and you are a startup founder
31:42
Ejaaz: that is roughly valued between 10 to maybe $100 million and you're going on
31:45
Ejaaz: to your next round, hey, maybe reach out to us. Maybe we'll platform you.
31:48
Ejaaz: You see the effect that we have.
31:50
Ejaaz: We're also looking for additional sponsors as well. So if this sounds like your
31:53
Ejaaz: shtick, please reach out. We would love to hear from you.
31:56
Josh: Awesome, yeah, well, on that note, thank you guys so much for watching and we'll
31:58
Josh: catch you guys next week.
31:59
Ejaaz: See you guys.