Skip to content
a16za16z

This Week in AI: GPT-5 Ships, 4o Pulled Back, Grok Imagine Goes Social

a16z partners Olivia and Justine Moore unpack the latest in consumer AI including: - Grok’s “Imagine” and its instant, social-first creative tools - Google’s Genie 3 and the future of 3D worlds - GPT-5: what’s new and missing, plus why some want their old chatbot back - AI-generated music from ElevenLabs - Olivia’s vibecoded Jensen Huang selfie app Timecodes: 0:00 Introduction & This Week's Topics 0:24 Grok Imagine: Social AI Image & Video Generation 4:48 GPT-5 Release & GPT-4 Deprecation 5:36 Comparing GPT-5 and GPT-4: Coding vs. Personality 9:13 AI for Mental Health: Illinois Law & Industry Impact 12:29 Genie 3: Interactive World Models from Google 16:53 ElevenLabs Music Model: Licensed AI Music Generation 19:16 Vibecoding: Consumer Experiments & Platform Evolution 24:14 The Future of Vibecoding & AI Tools 27:05 Conclusions Resources: Find Olivia on X: https://x.com/omooretweets Find Justine on X: https://x.com/venturetwins Read Anish and Justine’s vibecoding post: https://a16z.com/specialized-app-gen-platforms/ Stay Updated: Let us know what you think: https://ratethispodcast.com/a16z Find a16z on Twitter: https://twitter.com/a16z Find a16z on LinkedIn: https://www.linkedin.com/company/a16z Subscribe on your favorite podcast app: https://a16z.simplecast.com/ Follow our host: https://x.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details, please see a16z.com/disclosures

Justine MoorehostOlivia Moorehost
Aug 13, 202527mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:000:24

    Introduction & This Week's Topics

    1. JM

      Welcome back to This Week in Consumer. I'm Justine.

    2. OM

      I'm Olivia.

    3. JM

      And we have a bunch of fun topics we wanna cover this week, starting in the creative tools ecosystem with Grok Imagine, and then we're also gonna talk about Genie 3 and the ElevenLabs music model.

    4. OM

      And then we'll cover GPT-5 and the deprecation of GPT-4.0, and we'll cover our new vibe coding thesis.

    5. JM

      So this

  2. 0:244:48

    Grok Imagine: Social AI Image & Video Generation

    1. JM

      week we are going to start with Grok, which has had a bunch of big updates, um, over the last month or so I'd say. So obviously Grok 4 came out.

    2. OM

      Yeah.

    3. JM

      The Grok companions caused a huge stir, uh, particularly Annie and Valentine. Um, but I think more recently what's been really interesting is all of the image and video generation features on Grok Imagine.

    4. OM

      Yeah. So Grok released an image and video generation model called Imagine, which is offered standalone through the Grok app, and they're also bringing it to the web, and it's now embedded into the core X app as well, which is really exciting.

    5. JM

      Yeah, I think that's one of the things that's really unique about it. I would say it's not the most powerful kind of image or video generation model that exists. Elon has tweeted a bunch about how they're training a much bigger model. Um, but I think what's really cool about it is it's sort of one of the first really truly social kind of forays-

    6. OM

      Yeah

    7. JM

      ... into AI image and video generation. Um, and, and what you mean by it being integrated in, into the X app is, like, now when you post a photo on X, you can, like, long click and press and immediately turn it into a video-

    8. OM

      Yeah

    9. JM

      ... animated in the Grok app.

    10. OM

      Or even if you see someone else's photo posted on X-

    11. JM

      Yes

    12. OM

      ... you can turn it into a video or also edit the image with Grok, which is really exciting.

    13. JM

      Totally.

    14. OM

      I think one of the coolest things about Grok Imagine, to your point, it's not the f- most powerful model. It's not Veo 3.

    15. JM

      Yep.

    16. OM

      Uh, on video I would say the audio generation is okay, but not great.

    17. JM

      Yep.

    18. OM

      But it's fast.

    19. JM

      So fast.

    20. OM

      Which I think for a lot of people has been kind of the real barrier to-

    21. JM

      Yeah

    22. OM

      ... doing image and video generation more seriously-

    23. JM

      Yeah

    24. OM

      ... is you put in a prompt, you press go, and then sometimes you're waiting 30, 60, 90 seconds for a generation.

    25. JM

      Yeah, often minutes, honestly-

    26. OM

      Yeah

    27. JM

      ... for a generation.

    28. OM

      And Grok images are basically instant.

    29. JM

      Yes.

    30. OM

      And the videos are pretty fast as well.

  3. 4:485:36

    GPT-5 Release & GPT-4 Deprecation

    1. OM

      other big model news of this week, which, uh, is not just big for consumer but for pretty much all of AI, was the GPT-5 release.

    2. JM

      Yes.

    3. OM

      And, and the corresponding deprecation of GPT-4.0-

    4. JM

      Yes

    5. OM

      ... which I think ended up being even bigger [laughs] news in, in consumer land specifically.

    6. JM

      Yeah. This was, this one was sort of fascinating, 'cause obviously it's been a while since OpenAI had had a major LLM release-

    7. OM

      Yep

    8. JM

      ... since GPT-4, and so people were, like, very eagerly awaiting GPT-5. But yeah, as soon as I got access to GPT-5, I wanted to compare the outputs to GPT-4, and I immediately noticed GPT-4 was gone.

    9. OM

      Yeah. And so how are people... 'Cause I've seen a lot of posts of people kind of up in arms about 4.0-

    10. JM

      Yeah

    11. OM

      ... disappearing. How would you describe kind of the main differences between the models, at least how they're manifesting in user experiences?

  4. 5:369:13

    Comparing GPT-5 and GPT-4: Coding vs. Personality

    1. JM

      Yeah. So I've talked to a bunch of folks about this. I think kind of a, a, a widespread conclusion is GPT-5 is really good at, um, especially, like, front-end code.

    2. OM

      Yep.

    3. JM

      Um, I think a lot of the model companies are really focusing on coding-

    4. OM

      Yeah

    5. JM

      ... as a major use case, a major driver of economic value, something they can be really good at. And you can tell-

    6. OM

      Yeah

    7. JM

      ... in, in the results from GPT-5. It's, it's much better.

    8. OM

      And they emphasize it in the live stream pretty significantly.

    9. JM

      Totally.

    10. OM

      Yeah.

    11. JM

      And you can see from the examples people use, it's much better at generating things, it's much better at debugging, et cetera. But for, like, a lot of consumers aren't using it for code.

    12. OM

      Yeah.

    13. JM

      A lot of consumers just wanna chat with it, and there's a bunch of examples of how it's a lot less, um, expressive-

    14. OM

      Yeah

    15. JM

      ... emotional, and fun. Like, it doesn't really use exclamation points.

    16. OM

      [laughs]

    17. JM

      It doesn't really use emojis. It doesn't send things in all caps like it used to.

    18. OM

      It doesn't do the classic 4.0, "It's not just good, it's great."

    19. JM

      Yes, exactly. And I, I think there are kinda two separate issues here. So one is the sort of, like, glazing, excessive validation.

    20. OM

      Yes.

    21. JM

      Like, it said, like, "You're the best. You should totally do that. That's the best decision," for, like, everything you said, even if it was ridiculous, which is, like, a problem that I, I think I'm glad they're working on and-

    22. OM

      Yes

    23. JM

      ... and getting rid of. 'Cause let alone, like, everyone's concerns about, like, GPT psychosis or whatever, um, you just can't trust something that always-

    24. OM

      Yeah

    25. JM

      ... tells you you're right.The second thing is, um, does it just have, like, a fun and engaging, uh, and more casual human-feeling personality?

    26. OM

      Yeah

    27. JM

      And I think that actually maybe took a step back-

    28. OM

      I agree

    29. JM

      ... from GPT-4.0 to GPT-5.

    30. OM

      Yeah

  5. 9:1312:29

    AI for Mental Health: Illinois Law & Industry Impact

    1. OM

      Yeah. I agree. I do wanna spend 30 seconds on that mental health and, and health overall use case, though.

    2. JM

      Yeah, yeah

    3. OM

      It's interesting timing because also last week the state of Illinois just passed a law banning AI for mental health or therapy-

    4. JM

      Yeah

    5. OM

      ... without kind of the supervision of a licensed professional.

    6. JM

      Yeah

    7. OM

      And it's pretty interesting because the law is, is wide-ranging to the extent that some AI mental health companies have already shut down new operations in Illinois-

    8. JM

      Yeah

    9. OM

      ... or, or kind of prohibited new users from signing up.

    10. JM

      Yeah

    11. OM

      It's basically anything that's kind of ongoing support or even personalized advice around specific emotional and mental issues-

    12. JM

      Yes

    13. OM

      ... is now counted as therapy-

    14. JM

      It's crazy

    15. OM

      ... and is technically illegal in Illinois.

    16. JM

      Yeah

    17. OM

      I am confident ChatGPT is doing this, and honestly is doing it well-

    18. JM

      Yes

    19. OM

      ... for a lot of people, and so I guess my question is, to what extent is this ever gonna be enforced?

    20. JM

      Yeah

    21. OM

      'Cause they can see people's individual chats.

    22. JM

      I feel like Illinois always does weird stuff.

    23. OM

      Yeah. [laughs]

    24. JM

      Like, like, we've been consumer investors for too long, and I... Remember in, like, 2017, 2018-

    25. OM

      Yeah

    26. JM

      ... we would literally talk to social apps, like consumer social apps, that were like, "We've launched everywhere except for Illinois 'cause they have all these, like, crazy regulations-

    27. OM

      Yes

    28. JM

      ... around, like, people, like, data and, and sharing, and, like, all of these things," which obviously-

    29. OM

      Compliance. Yeah

    30. JM

      ... it's good to have those, but, like, went way beyond other states-

  6. 12:2916:53

    Genie 3: Interactive World Models from Google

    1. OM

      There were two other big actually creative tool model releases this week.

    2. JM

      Yes

    3. OM

      Genie-3-

    4. JM

      Yes

    5. OM

      ... from Google.

    6. JM

      Yeah

    7. OM

      And then a new music model from ElevenLabs.

    8. JM

      Yes

    9. OM

      So maybe let's start with Genie-3. W- what, what is it? I've seen the videos, but what is it?

    10. JM

      Yes. Genie-3 took Twitter by storm.

    11. OM

      Yeah

    12. JM

      Um, so Google has a bunch of kind of different initiatives around image, video, 3D world. Um, I think various teams like Veo 3, um, and the Genie team working towards this idea of, like, an interactive world model-

    13. OM

      Yeah

    14. JM

      ... which is basically-You are able to, um, have a scene that you can walk through in real time-

    15. OM

      Yeah

    16. JM

      ... or interact with that kind of generates on the fly. Um, and, and you can imagine it sort of like, you know, a personal video game.

    17. OM

      Yeah.

    18. JM

      Um-

    19. OM

      I saw some of the videos of kind of taking famous paintings, and for the first time you're able to step into them and kind of swivel around and move around in the room. Almost like you're, you have a VR headset on or something-

    20. JM

      Yes

    21. OM

      ... and you're kind of turning around and seeing the full environment.

    22. JM

      That, those were really cool. And it's not just famous paintings. They've shown a bunch of examples of, um, you can, from a text prompt you can create a world.

    23. OM

      Yeah.

    24. JM

      From an image you can create a world.

    25. OM

      Yeah, amazing.

    26. JM

      They've even shown taking VO3 videos and, and creating a world around it-

    27. OM

      Wow

    28. JM

      ... with Genie 3. And the cool thing about Genie 3 is there's controls where you can move the character around.

    29. OM

      Mm.

    30. JM

      So you can control, like, now go to the left, and then the scene sort of re- regenerates to-

  7. 16:5319:16

    ElevenLabs Music Model: Licensed AI Music Generation

    1. OM

      That's awesome

    2. JM

      ... eagerly awaiting that one to come out.

    3. OM

      Yeah.

    4. JM

      Um, and then, yes, our, our portfolio company ElevenLabs also released their music model.

    5. OM

      Yeah.

    6. JM

      Which, um, is super exciting.

    7. OM

      I did not know they were working on music.

    8. JM

      Yes. It's, um, it's been in the works for a bit. The, the really interesting thing about it is it's trained on fully licensed music.

    9. OM

      Got it.

    10. JM

      Which means... So music is one of those spaces where, um, the rights holders are extremely litigious.

    11. OM

      Yeah.

    12. JM

      And so compared to things like image or video, it's been harder for, um, music companies to sort of avoid, uh, stepping on the toes.

    13. OM

      Yeah, you can't just scrape data from the internet.

    14. JM

      Yes.

    15. OM

      Like, the record labels will come and, and sue you. [laughs]

    16. JM

      Yes, and the artists-

    17. OM

      Yeah

    18. JM

      ... and, like, it's often a very complicated ecosystem-

    19. OM

      Yeah

    20. JM

      ... of who owns the rights to a specific song or to an artist's voice-

    21. OM

      Yep

    22. JM

      ... or something like that. Um, and so, yeah, I think a lot of folks have thought that you could not get a good quality, um, music model training on licensed data, because it's hard and it's expensive and it takes a long time.

    23. OM

      Yeah.

    24. JM

      And it's hard to get them to agree to license you the data.

    25. OM

      Yeah.

    26. JM

      But from what I've seen and from my own experiments, folks have been really impressed by Eleven's output.

    27. OM

      [upbeat music] And so what does the licensed data open up in terms of use cases for the music model, do you think?

    28. JM

      Yeah. So I think a lot of consumers basically don't care if they're using-

    29. OM

      Yeah

    30. JM

      ... a music model that's trained on licensed data or not, because, um, they're not really monetizing, or many of them are not monetizing stuff that they make-

  8. 19:1624:14

    Vibecoding: Consumer Experiments & Platform Evolution

    1. OM

      And then our last big topic of this week is around vibecoding, which continues to explode.

    2. JM

      Yes.

    3. OM

      I think we have two things to talk about here. One would be our own experiments in the world of vibecoding-

    4. JM

      Yes

    5. OM

      ... which relates to a piece that you and Anish Acharya put out-

    6. JM

      Yeah

    7. OM

      ... this past week about how we're seeing the vibecoding market start to fragment.

    8. JM

      Yes. Your experiment is the more interesting part.

    9. OM

      So my-

    10. JM

      But, so let's start with that.

    11. OM

      Yeah. Maybe to give a real-world example, um, for the first time, I vibecoded an app that I fully published and made available to the internet. Essentially, what I did was I thought, "Hey, I'm seeing on my X feed all the time everyone has a selfie with Jensen at NVIDIA."

    12. JM

      How did they get this?

    13. OM

      In his classic leather jacket.

    14. JM

      Yes.

    15. OM

      He, he must be spending all of his time taking selfies now because-

    16. JM

      Yes

    17. OM

      ... everyone has one, and I don't.

    18. JM

      Yes.

    19. OM

      And so I was thinking there's all these new amazing models out there like Flux Contexts that-

    20. JM

      Yes

    21. OM

      ... can kind of take an image, say, of Jensen taking a selfie with someone else, and put my... stick me in there instead.

    22. JM

      Yes.

    23. OM

      Replace me in the selfie.

    24. JM

      I should've been in the video.

    25. OM

      I should've been in the video.

    26. JM

      For those.

    27. OM

      Exactly. [laughs]

    28. JM

      Yeah.

    29. OM

      So, uh, I did that.

    30. JM

      Yeah.

  9. 24:1427:05

    The Future of Vibecoding & AI Tools

    1. JM

      the stack.

    2. OM

      Mm-hmm.

    3. JM

      And that level of inflexibility-W- that product would just not work for them.

    4. OM

      Yeah.

    5. JM

      Um, and so I think what we are hoping to see basically is, um, v- like, specialized players emerge that offer the best product for a particular demographic of user or for a particular use case.

    6. OM

      Yeah.

    7. JM

      Um, and, and that will probably imply very different product experiences, product constraints, and also go-to-market strategies. Like, if you are, um, allowing any consumer to vibecode like a fun-

    8. OM

      Yeah

    9. JM

      ... app to share with their friends or partner or whatever, um, you probably want to be going viral on TikTok and Reels.

    10. OM

      Yeah.

    11. JM

      Um, versus if you are building a vibecoding platform for, um, designers to prototype new features-

    12. OM

      Mm-hmm

    13. JM

      ... in an enterprise or for engineers to make internal tools, um, you might want to even have top-down sales.

    14. OM

      Yeah.

    15. JM

      Or at least be kind of product-led growth within businesses.

    16. OM

      Yeah, and maybe invest in, like, deep integrations into, you know, core business systems-

    17. JM

      Totally

    18. OM

      ... other things like that. Whereas the consumer version, you might actually just want people to vibecode on mobile and get something that works in five minutes.

    19. JM

      Y- and that's a great point, too, which is, like, the consumer users often just want something to, like, look cool-

    20. OM

      Yes

    21. JM

      ... and work, um, and not have security issues, whereas more business-

    22. OM

      Yes

    23. JM

      ... oriented users, it often needs to integrate with what already exists for the business.

    24. OM

      Yeah.

    25. JM

      Whether that's, like, a design system, an aesthetic, um, or whether that's, you know, their CRM-

    26. OM

      Yeah

    27. JM

      ... or the emailing platform they use or sort of all of these different products that it needs to connect to that are external to the vibecoding tool.

    28. OM

      Yeah. And so I think the conclusion of the piece was, like, we're seeing early winners in vibecoding already.

    29. JM

      Yeah.

    30. OM

      These are some of the fastest growing companies in the AI application space.

  10. 27:0527:38

    Conclusions

    1. OM

      Yeah. Super exciting.

    2. JM

      Well, thanks for joining us this week. If you've tried out any of these creative models or had any vibecoding experiments yourself, we'd love to hear from you. Please comment below and let us know. And also please feel free to ping us here or on Twitter if you have ideas of what we should cover in a future episode. [upbeat music]

Episode duration: 27:48

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode qluWpYMYR0U

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome