Latest / Today’s AI News / OpenAI Breaks Free From Microsoft, China Blocks Meta's Manus Deal, AlphaGo Creator's $1.1B Lab

Transcript

OpenAI Breaks Free From Microsoft, China Blocks Meta's Manus Deal, AlphaGo Creator's $1.1B Lab · Today’s AI News

• 0:00
  Welcome back to today's AI News, brought to you by 9X
• 0:02
  Productions. Imagine a high profile corporate
• 0:06
  marriage suddenly deciding to open things up.
• 0:09
  Like less control, but they're still splitting the bank
• 0:11
  account. That's exactly what just
• 0:13
  happened between two of the most powerful players in tech.
• 0:16
  I mean, it's an open relationship, but with billions
• 0:20
  on the line. It really is, and you know we
• 0:22
  are unpacking a massive stack of the latest developments today.
• 0:25
  We're going to run through the quick hits of the day and then
• 0:27
  cap it off with a really inspiring community workflow.
• 0:31
  Definitely, because the goal here is to give you, the
• 0:34
  listener, the ultimate shortcut to being well informed on
• 0:37
  today's AI landscape without all the info overload.
• 0:41
  OK, let's unpack this. So the reworked partnership
• 0:45
  between Open AI and Microsoft for years, I mean, this has been
• 0:47
  the defining alliance of the whole artificial intelligence
• 0:50
  boom. Oh.
• 0:51
  Absolutely the cornerstone of the industry, really.
• 0:53
  Right, but Microsoft no longer has exclusivity over Open AI's
• 0:57
  IP. And the part that really made me
• 1:00
  do a double take, the highly debated AGI clause, has been
• 1:03
  removed entirely. Yeah, what's fascinating here is
• 1:07
  the underlying tension. You know the relationship had
• 1:09
  clearly gone sour. Clearly.
• 1:12
  Removing that exclusivity, it allows open AI to meet their
• 1:15
  enterprise clients wherever they're actually housed.
• 1:18
  Because right now a lot of them are on AWS, right?
• 1:21
  Amazon Web Services. Yeah.
• 1:22
  And Amazon CEO Andy Jassy actually called this whole thing
• 1:26
  very interesting, which, you know, is a massive
• 1:29
  understatement in CEO. Exactly.
• 1:31
  And he said that right after a memo leaked from Open a Eyes
• 1:35
  Chief Revenue Officer Denise Dresser, where she was basically
• 1:38
  talking up the Amazon Bedrock platform.
• 1:41
  Right, because open AI can now use rival clouds like bedrock.
• 1:44
  Oh wait, let me push back on this for a second.
• 1:46
  It's like letting your partner keep the apartment but demand to
• 1:48
  see other people. Why on earth would Microsoft
• 1:51
  surrender the AGI clause? I mean, that was their ultimate
• 1:54
  trump card, right? If open AI achieves artificial
• 1:56
  general intelligence, that clause was everything.
• 1:59
  It was, yeah, but you have to look at the trade off.
• 2:02
  Microsoft is wisely trading these really ambiguous AGI
• 2:06
  definitions for a guaranteed hard coded 6 year revenue
• 2:09
  stream. OK, a guaranteed.
• 2:11
  Stream, yeah, they keep a revenue share running all the
• 2:13
  way through 2030, plus they retain Azure first launch access
• 2:17
  through 2032. Oh wow, so it's a long term
• 2:19
  financial lock? Exactly.
• 2:21
  And, you know, it completely settles Microsoft's reported
• 2:24
  lawsuit threat over that $50 billion Amazon Open AI deal.
• 2:28
  Right, right. Because they were supposedly
• 2:30
  threatening to sue over that. Yeah, so now the milestones are
• 2:33
  tied to calendar dates rather than the elusive announcement of
• 2:37
  AGI. Because honestly, who even
• 2:40
  defines when AGI is achieved? It's totally subjective.
• 2:44
  Right, it's a philosophical debate.
• 2:45
  Microsoft basically said let's skip the philosophy and just
• 2:48
  lock in the cash. That makes total sense.
• 2:50
  Yeah. So it's a shift from corporate
• 2:52
  power struggles to, like, guaranteed money.
• 2:55
  But if that's the corporate power struggle, there's a much
• 2:58
  bigger global power struggle happening right now.
• 3:00
  Oh, heavily. Yeah, it's all about the
• 3:02
  shifting control of AI assets by national governments.
• 3:05
  So Beijing just completely blocked Meta's $2 billion
• 3:09
  acquisition of Manus? This is a huge story, yeah.
• 3:14
  Manus is a Singapore based AI startup, but they have really
• 3:17
  deep Chinese roots and China's National Development and Reform
• 3:21
  Commission basically stepped in, barred the foreign investment
• 3:24
  and ordered the companies to undo the deal.
• 3:26
  Just completely unwind it, yeah. And Meta only announced this
• 3:30
  deal back in December. By January, Chinese officials
• 3:34
  had already launched A probe regarding export control and
• 3:37
  foreign investment rules. It was incredibly fast.
• 3:39
  It really was, and Meta claims the teams are already deeply
• 3:42
  integrated at their Singapore office.
• 3:44
  I mean the Manus website literally said now part of Meta.
• 3:48
  Right. They're already operating as one
• 3:49
  unit. Exactly and reportedly, Manus
• 3:53
  executives were barred from leaving China during this whole
• 3:55
  probe. Which is terrifying, honestly.
• 3:57
  Completely. And all of this is happening
• 3:59
  just weeks before Trump meets XI in Beijing.
• 4:02
  But logistically, I have to ask, how do you unwind A deeply
• 4:05
  integrated tech team? That is the big question.
• 4:08
  I mean, can you unmix the paint once it's in the bucket?
• 4:11
  Once the code is shared, it's shared, right?
• 4:13
  Yeah, you really can't unmix the paint, but Beijing has
• 4:17
  officially classified AI talent as a national security asset.
• 4:21
  Wait. The talent itself, the people.
• 4:23
  Yes, the human capital. They are applying the exact same
• 4:28
  export control logic to these human engineers that the US
• 4:31
  applies to physical semiconductor chips.
• 4:34
  Oh wow, that is a massive escalation.
• 4:37
  It is. The geopolitical stakes are just
• 4:39
  astronomical right now. It creates this incredibly
• 4:42
  precarious situation for any founders with Chinese roots who
• 4:46
  are trying to move their tech outside of Beijing's reach.
• 4:49
  Because even if your office is in Singapore, if your engineers
• 4:51
  are in China or you have ties there, you are under their
• 4:54
  jurisdiction. Exactly.
• 4:56
  You are essentially A controlled export at that point.
• 4:58
  That is wild. Well, if talent and computing
• 5:00
  data are being geographically bottlenecked by international
• 5:03
  governments like that, the ultimate workaround is removing
• 5:06
  human data from the equation entirely.
• 5:08
  Right. Precisely.
• 5:09
  Which brings us to David Silver. Right.
• 5:11
  So David Silver, the X Deep mind researcher who created Alphago
• 5:14
  and A0, he just launched this New London lab called Ineffable
• 5:19
  Intelligence. And the numbers on this are just
• 5:21
  staggering. Insane.
• 5:22
  They raised $1.1 billion at a $5.1 billion valuation.
• 5:27
  Yeah, which is Europe's largest seed round ever, by the way.
• 5:30
  A seed round over a billion dollars for a seed ring.
• 5:33
  No, no, it's unprecedented. So the whole goal of ineffable
• 5:36
  intelligence is to build an AI that learns purely from
• 5:39
  experience in simulations like completely skipping pre training
• 5:44
  and human data. No Internet scraping at all,
• 5:47
  right? Silver actually refers to human
• 5:49
  data as a fossil fuel, and he says his model is a renewable
• 5:53
  fuel that can quote, learn and learn forever.
• 5:56
  It's a bold claim. They are aiming for a scientific
• 5:59
  breakthrough of comparable magnitude to Darwin.
• 6:02
  But here's where it gets really interesting.
• 6:05
  I mean, isn't human data the entire foundation of AI right
• 6:08
  now? Traditionally, yes.
• 6:09
  So how does a machine learn to think without our fossil fuels,
• 6:14
  without her writing and our history?
• 6:16
  Well, this connects directly to Yan Likun's theory.
• 6:18
  He's the chief AI scientist at Meta, and he's been saying for a
• 6:22
  while that traditional LLM's, the ones trained on human texts,
• 6:25
  are essentially a dead end. Really.
• 6:27
  A dead end? Yeah, because they just predict
• 6:29
  the next word based on what humans have already said.
• 6:32
  They don't actually reason. So companies like AMI Labs,
• 6:35
  Recursive Superintelligence, and now Ineffable, they are betting
• 6:39
  massive, massive capital on this pivot.
• 6:42
  Over a billion dollars of capital.
• 6:44
  Exactly, By bypassing human constraints and just letting the
• 6:48
  AI figure things out in a simulated physics or logic
• 6:52
  environment, Silver hopes to make what he calls first contact
• 6:56
  with super intelligence. First contact.
• 6:58
  He's talking about it like it's an alien species.
• 7:00
  He really is. It's shifting the entire
• 7:03
  paradigm from data scraping to boundless infinite simulation.
• 7:07
  That is just mind blowing. But you know, while these
• 7:10
  multibillion dollar labs are trying to build alien super
• 7:12
  intelligence, the most immediate shift for you, the listener is
• 7:15
  probably building practical autonomy on your own desk today.
• 7:18
  Yes, let's bring it back down to earth.
• 7:20
  Right, so there is this new guide on setting up AI teammates
• 7:25
  using Chat GPT's Workspace Agents tool.
• 7:28
  Which is such a powerful feature.
• 7:30
  Yeah, it's currently in a research preview for business
• 7:32
  enterprise Edu and teachers plans, but the mechanics are
• 7:35
  super cool. You basically go to chatgpt.com.
• 7:38
  Agents Create an agent and give it ownership of a tool or a
• 7:42
  database. Right, like you can literally
• 7:44
  hand it the keys to a specific workflow.
• 7:47
  Exactly. The prompt structure they
• 7:49
  suggest is you ask it to manage a specific pipeline, say a
• 7:53
  Notion database or a Google Docs folder, and you give it three
• 7:57
  main tasks that run autonomously on schedules.
• 7:59
  And the magic here is that codecs figures out all the
• 8:02
  integrations. Wait, codecs actually writes the
• 8:04
  integration code? Yes, you don't have to be a
• 8:06
  developer. Codecs figures out how to
• 8:08
  connect the API, so you can just set the agent to run every
• 8:11
  single day in the background. That is incredible.
• 8:13
  And you know, even if you don't have a business plan, there's a
• 8:16
  massive pro tip here from the sources.
• 8:18
  Oh, for sure. The approach of prompting an AI
• 8:21
  to manage tasks autonomously. You can and apply that across
• 8:25
  other tools too. It's a total mindset shift from
• 8:28
  thinking of it as a chat bot to treating it as a teammate.
• 8:31
  Exactly. You stop asking it questions and
• 8:33
  start giving it responsibilities.
• 8:35
  Yes, giving it responsibilities and this agentic autonomous
• 8:40
  shift, it is causing ripple effects across literally every
• 8:44
  sector of the tech landscae. It's changing everything.
• 8:46
  Which brings us to our quick hits and everything else haening
• 8:49
  in AI today. Let's just wrap and fire through
• 8:51
  these. Ounds good.
• 8:52
  Let's start with hardware. Go for it.
• 8:55
  So highly respected tech analyst Ming Chi Crew says Open AI is
• 8:59
  actively working on its own smartphone.
• 9:01
  A physical. Phone.
• 9:02
  A physical phone. They are partnering with
• 9:04
  Mediatek and Qualcomm, and instead of just apps, it will
• 9:08
  feature native AI agents baked right into the OS.
• 9:11
  They are targeting a 2028 production.
• 9:13
  Oh man, Apple and Google have to be sweating.
• 9:15
  Oh. Absolutely.
• 9:16
  OK, on the legal front, the trial between Elon Musk and Open
• 9:20
  AI has officially kicked off with jury selection.
• 9:22
  That's. Going to be messy.
• 9:23
  Very. Both sides are already just
• 9:25
  trading intense barbs on X. It is a massive spectacle.
• 9:28
  Yeah, and Speaking of legal moves, deepfakes are getting
• 9:31
  serious pushback. Taylor Swift just filed 3
• 9:34
  federal trademarks for her likeness and voice.
• 9:36
  Wow. Three of them.
• 9:38
  Yeah, she's joining Matthew McConaughey in taking proactive
• 9:41
  legal action to prevent AI deepfakes from using their
• 9:44
  identities. Which makes total sense,
• 9:46
  honestly. Protect your brand over in
• 9:48
  Creative Tools. Adobe just opened public beta
• 9:50
  access for its Firefly AI assistant.
• 9:53
  This is a game changer for designers.
• 9:55
  Seriously, it allows creators to prompt multi app creative cloud
• 9:59
  workflows, so you can use it across Photoshop, Illustrator,
• 10:02
  all of it. And crucially, it keeps the
• 10:05
  outputs completely editable with layers.
• 10:07
  Right, it doesn't just flatten the image.
• 10:09
  Exactly. And finally, for video and
• 10:11
  mobile, Alibaba's new Happy Horse video model just took the
• 10:15
  top spot on artificial analysis as leaderboard.
• 10:18
  Happy horse, great name. Right.
• 10:20
  Also, Cling 3 Point O launched a new 4K mode for cinematic video
• 10:25
  which looks insane and lovable. The AI app building platform is
• 10:30
  now available natively via a mobile app.
• 10:32
  You know, this raises an important question.
• 10:34
  What's that? Look at how fast hardware and
• 10:36
  legal frameworks are scrambling right now.
• 10:38
  They're just desperately trying to catch up to these software
• 10:41
  advancements. It's like the software is moving
• 10:43
  at lightspeed and the physical world is stuck in traffic.
• 10:45
  Exactly. The laws and the phones just
• 10:48
  weren't built for autonomous agents.
• 10:50
  So what does this all mean? Like really?
• 10:52
  We have all these massive global tech updates, but we need to
• 10:56
  bring it back down to earth. Right.
• 10:58
  How does it affect regular people?
• 10:59
  Exactly how is a real person applying this in their daily
• 11:03
  life? And that brings us to today's
• 11:05
  community AI workflow. I love this.
• 11:08
  One Me too. Today's workflow comes from
• 11:10
  Lydia F in North Liberty, IA. Shout out to Iowa.
• 11:13
  Yeah, so Lydia used codecs to build a custom web-based
• 11:18
  application for her teenage daughter.
• 11:20
  And what was it for? To study for the Iowa driver's
• 11:23
  permit test. That is so.
• 11:25
  Right. Instead of just buying flash
• 11:27
  cards, she uploaded the full official Department of
• 11:30
  Transportation manual and a bunch of practice tests.
• 11:32
  She just fed the whole thing into the model.
• 11:34
  Yep, and she generated a Bank of 200 custom flash cards.
• 11:38
  But here is the crazy part. She had codex build in a
• 11:41
  weighted quiz feature. Oh wow, so it adapts.
• 11:44
  Exactly. It pulls 25 multiple choice or
• 11:47
  true false questions at a time, but it weights them based on
• 11:51
  what her daughter gets wrong so she gets tested more heavily on
• 11:54
  her weak spots. That is brilliant.
• 11:56
  And she deployed it, got a URL and shared the link with other
• 11:59
  local moms in her town. You know, if we connect this to
• 12:03
  the bigger picture, this right here is the true revelation.
• 12:06
  Yeah, it isn't just happening in billion dollar London labs with
• 12:09
  supercomputers. It's happening in Iowa.
• 12:12
  Yeah, empowering non-technical people to just build stuff.
• 12:15
  Exactly. Empowering a mom to build a
• 12:17
  highly specific, practical, adaptive tool for her own
• 12:21
  community. That is the promise of AI,
• 12:23
  right? There it really is.
• 12:24
  It's so cool to see it actually working in the wild.
• 12:26
  Absolutely. Though, as we wrap up today, I
• 12:29
  do want to leave you with something to ponder going back
• 12:31
  to that ineffable intelligence segment.
• 12:33
  The simulation guys. Yeah, think about this.
• 12:36
  If the AI of the future entirely abandons our human fossil fuel
• 12:41
  data, right, and it learns entirely in its own simulated
• 12:44
  vacuum, will it still understand human intuition?
• 12:48
  Or will it evolve a completely alien way of solving problems
• 12:53
  that we literally cannot even comprehend?
• 12:56
  That is. Wow, that's a heavy thought.
• 12:59
  If it doesn't read our books, it doesn't know how we think.
• 13:01
  Exactly. It might solve the problem, but
• 13:03
  in a way that makes zero sense to a human brain.
• 13:06
  That is both fascinating and slightly terrifying to think
• 13:09
  about. Just a little bit.
• 13:10
  Well, if you want to keep exploring these wild scenarios
• 13:12
  and stay on top of the news, make sure you hit Subscribe to
• 13:16
  stay updated. Yes, please do.
• 13:17
  And hey, we would love it if you took a second to rate the show
• 13:20
  with five stars. It really helps us out.
• 13:22
  Thank you so much for hanging out with us.
• 13:24
  And we will see you right back here for tomorrow's episode.