Ceren Ates

Been thinking about this a lot recently: the J-curve for gaming investments. Takes more capital and more time to launch a game compared to a consumer app or LLM-wrapper. This is more similar to biotech or hardtech investing but with less moats (no patents, no capex, Tencent/Netease might fast follow you). A few ideas to ease the curve: - Move faster with AI-enabled production (sfx, vfx, code, 2d assets, video) - Bootstrap to more before raising - Earlier liquidity through pre-sales (kickstarter or NFTs) - Go after high-leverage, multiplayer genres - Playtest earlier to get to retention earlier What else?

120

35 Comments

Have we reached the tipping point for the Spatial Web? If you follow me, you already know that I am Bullish on the Spatial Web (aka Metaverse). Since Linden Labs launched Second Like over 18 years ago, I have been a proponent of these open world experiences. When I look at Roblox, FORTNITE, Minecraft, and others, I see this open world, immersive medium as a viable gateway to the Spatial Web and Web 3.0. An entire generation of Digital Natives has grown up with games as their primary entertainment medium, and now this large group of over 3.3 Billion global gamers are carrying the next generation wallet share. Entire worlds needed to be built, and consumer adoption needed to scale. Are we now at that inflection point where the social and economic opportunities will come together for these next-gen platforms? Interested in the Spatial Web for your business or industry? Let’s discuss in the comments below ⬇️ [Read more here](https://lnkd.in/dCs88wHk?)

2

I'm listening to “Meta’s Earnings Call” right now and R&D investment into their Reality Labs division (VR & AR) reached $4.8bn in Q2. This means that despite the naysayers and “Meta’s year of efficiency” Zuckerberg’s investment into our industry has grown by 20% since 2023 and 49% since 2022. Mark Zuckerberg’s conviction is truly historic and it only “ups the ante” for companies like Apple, Sony, and Samsung to continue investing. The headsets that are being built right now are indeed the computing devices of the future. On the call, Mark said himself that the “Quest 3 sales are exceeding expectations” and that “in addition to gaming, people are increasingly taking advantage of the Quest’s capabilities as a general computing platform.“ While Meta pioneers the “platform,” here at Immerse we’ve made it our mission to prove the most compelling use case for VR: language learning. As of this month, our research partners published their 38th study, contributing to our growing body of evidence that learning in Immerse is empirically 3x-4x more effective than on your computer or mobile. With every passing month, I believe VR is the technological wave most poised for disruption. In order for the wave to crash though, it requires cutting-edge hardware to be paired with compelling software use cases that leverage VR’s affordances. When those two come together this wave will be unstoppable.

54

6 Comments

If you're reading this, that means you have almost certainly already seen the clip below. For those who have not yet seen it, at Ray Summit, OpenAI's Chief Product Officer Kevin Weil said the company's o1 model can write legal memoranda that previously were the domain of $1000/hour associates, leading Mr. Weil to ask: "what does it mean when you can suddenly do $8,000 of work in 5 minutes for $3 of API credits?" I have an answer to that question. And, from what I've seen thus far on social media, mine is apparently a bit of a contrarian take: 1) Your memo is no longer worth $8,000: Sorry to be the bearer of bad news, but this bit is fairly obvious, no? Two things are happening here. One, law firms no longer have the sort of information moats they once did - for example, I do not need to go to my local law library to search the municipal code; rather, it is one click away. And two, for a majority of junior associate level tasks (e.g., legislative summarization and analysis), the LLMs produce comparable output exponentially faster. 2) So you better adapt your skill set and practice to the times: Not familiar with these AI tools yet? It's time to familiarize yourself, or you're giving your competitors a huge advantage. Have a practice that is highly reliant on junior associate billing for quasi-administrative legal work? Likely time to rethink that business model. Lawyers that adapt to the changes that AI presents will continue to thrive. If I were the associate in Mr. Weil's hypothetical, I would focus on technological literacy as well as all of those "soft skills" that continue to be value add and not yet easily displaced by technology: ability to synthesize information, interpersonal communications, excellent judgment, etc. Curious to hear others' takes in the comments...

12

1 Comment

In specific creative circles, I understand the frustration many ppl have about AI in the age of LLMs and GenAI, and the one-dimensional practice of prompting. A lot of that frustration comes from the threat (and reality) of job loss in current industries, and just how lacking in originality a lot of the early AI art can be. I always looked at the progression of any technology spurring waves of xenophobia because the "Other" always struggles before we find equilibrium in its relationship to human value. Its funny to remember how I was made fun of in middle school for just making friends online! The internet was sooo taboo back then. Just look at where things are now. Prompting alone does not necessarily provoke a sense of "creativity" for those of us who have found the love of accessing and developing our skills/taste through a more multi-directional process. The UI/UX in this arena is still in its infancy and has not really accessed a human's creative potential and agency. As an artist, I understand that 100%. There is an actual joy and value in the creative process that isn't just about how the end things look, but how one gets to an end thing and what you do and learn along the way. With the right entry into AI and with the right tools and workflows, I'd argue it can be just as fulfilling and interesting. This is coming from years of multidisciplinary work and teaching creative technology to hundreds of students over the years. However, outside of art making, not many creatives consider how recent developments in AI have made a whole lot of improvements for non-English native speakers, or how AI has dramatically changed research, and how recent scientific discovery is made possible through sheer statistical analysis that previous computation couldn't do before. It's hard to break it to everyone, but AI has been here, and its advancements are here to stay. That's not to say we shouldn't have ethical considerations to the changes impacting our lives. No technology can ever really take our creativity away from us as humans. We just have to be open to imagining what ways we want new technology to be a part of the creative process. This may look different for different people. This collective imagining will usher a lot of changes, including experimentation, where and how humans craft agency and the new languages to describe these experiences. That's something im very hopeful for! As for the jobs, let's first imagine how we can better co-create with AI, and the rest will come! ✨️

27

3 Comments

The top trend for 2024 is tech companies overpromising in AI demos, so I'm withholding judgment on whether this will be as useful as it looks until after I have tested it. Still, the only way to run consumer-facing AI at scale is at the edge, and this architecture seems promising. I had hoped for a native federated learning framework for developers so we could train models using on-device customer data in a privacy-first way.  Perhaps next year.

17

1 Comment

After an amazing run at Tome, alongside some of the most talented and kindest people I know, I'm excited to share that I've cofounded Outerport (YC S24) with my friend Towaki Takikawa. https://lnkd.in/g5Fad4Zv I became fascinated with AI in 2015, both as a scientific question and for its practical utility. The playing field has transformed since then - before, models could only perform well-defined tasks with limited capabilities, but now the XXXB parameter models that ate up all of the Internet data can exhibit general intelligence. As these models move to production, we need a new set of tools. Outerport (YC S24) is a registry for managing model weights and orchestrating deployment onto GPU servers. The system sends the model weights to application servers just-in-time and keeps them in the right level of cache, which enables you to minimize cold start time, scale horizontally, hot swap models instantly, and maintain version control. I've spent most of my career so far in AI. I've conducted research on the mathematics behind machine learning, worked on computer vision for autonomous vehicles, and most recently applying LLMs to help people communicate important ideas. I'm excited to continue building systems to support deploying foundation models in production. We're eager to partner with early adopters who are pioneering deployment of generative AI applications - let me know at [email protected] if you are interested.

171

30 Comments

𝐎𝐩𝐞𝐧𝐀𝐈 𝐍𝐚𝐦𝐞𝐬 𝐂𝐨𝐢𝐧𝐛𝐚𝐬𝐞'𝐬 𝐊𝐚𝐭𝐞 𝐑𝐨𝐮𝐜𝐡 𝐚𝐬 𝐅𝐢𝐫𝐬𝐭 𝐂𝐌𝐎 OpenAI has hired its first chief marketing officer, nabbing Kate Rouch from crypto company Coinbase. Rouch joined Coinbase in August 2021 to lead global marketing and public relations, and oversaw the company’s Super Bowl ads. Before that, she spent more than 11 years at Meta, including a role as global head of brand and product marketing for Instagram, WhatsApp, Messenger and Facebook. The news of Rouch joining OpenAI comes days after the company’s ChatGPT artificial intelligence (AI) chatbot turned 2 years old. Since that debut, the market capitalization for the six largest tech companies has increased — in aggregate — by more than $8 trillion. OpenAI’s ChatGPT has added 250 million weekly active users since its launch two years ago and the company aims to reach 1 billion users over the next year. The company anticipates a new wave of growth driven by new products, new data centers and its partnership with Apple.

5

1 Comment

New preprint! "Pictures Of MIDI: Controlled Music Generation via Graphical Prompts for Image-Based Diffusion Inpainting" ArXiv: https://lnkd.in/eDyHNWxt Examples and Live Interactive Demo (draw your own ideas): https://lnkd.in/eM-nGbPs A solo paper exploring "graphical prompts" that allow generative music models to follow rough melodic contours specified by users. Inspired by songwriting workflow of hit producer/songwriter & Hyperstate AI co-founder Louis Bell! Even allows for accompaniment generation, autoregressive(-ish) continuation, and even "spelling out words" (though not as well as Jacob Collier). Objective and perceptual scores are comparable to a similar model from last year (Polyffusion) yet we can now go longer and support more detailed prompts. Fun fact: results sound qualitatively much better and offer better control than the Transformer model I put together last fall for a course at Belmont University (https://lnkd.in/eM2xDKVh). Trained on a mere 81 MB of MIT-licensed MIDI data. Thankful that both Belmont and Hyperstate value preserving the human artistic process while leveraging new technologies to help humans do their best work. :-) This work is similar to multiple prior works, yet manages to do so via a very simple implementation with an Hourglass Diffusion Transformer (HDiT), mostly by just swapping out the training images in the examples for Katherine Crowson's k-diffusion package (https://lnkd.in/erH2Z6vq) and adding some MIDI<-->image translation utilities, and adding inpainting to the sampler.

22

1 Comment

Llama 3, the latest GenAI model by Meta has arrived and it directly competes with Google’s Gemini 1.5 PRO and Anthropic Claude 3 Sonnet. More specs in the link 👇👇👇 #genai #ai #technology https://lnkd.in/dtVJS82U

16

1 Comment

"All AI can do is semi-automatic screensavers", said legendary director Benicio Del Toro. And indeed, when you watch this video I just generated with Kling, it seems clunky, dumb, superficial, almost hopeless. But Benicio is wrong. The progress in AI video is mind boggling. In a few years, these kind of early AI AI-generated videos will feel like the mute, black and white videos from the 19th century - they were clunky, but they were the first brick of what would become a foundational industry. It has been said time and again: when photography became popular, painters refused to called it art because it was "low-effort". Now, directors that have been through the nightmares of production (and I know what I'm talking about) consider AI "low-effort", and as such, not worthy of their attention. Ironic, given how many of them use CGI instead of practical effects. Months ago, we didn't believe AI could comprehend any physics. Now, it's getting pretty good - you can even help it understand what should move. We can foresee the next models getting even better at that, with consistent characters, places, vehicles, etc. Already, millions of new creators are painstakingly creating first video concepts with the very clunky tools at their disposal. And it's only going to get better. I'm extremely excited at the idea of being able to one day create a movie out of my science-fiction novel, without having to spend millions of dollars. I will do it with the passion of artistic craft that I used to create my movie - the medium will just be different. AI will create millions, if not billions of new artists and creators. Let's embrace the technology.

32

2 Comments

Yesterday, my team at Vareto announced the first-ever multiplayer modeling platform for Finance. 😍 Now that I can finally speak publicly about what we've been working on, the first thing I want to share is a screenshot from a live multiplayer session we did internally at the company. One of the interesting things about building enterprise software as a startup is that we're still too small to *need* our own product. So instead of doing what our customers do--expense budgeting, headcount planning, etc.--we use Vareto's multiplayer modeling to forecast fun things like total pet spend. Yep, that's right. We had 40 people go into a Vareto model and enter in bottom-up inputs for how much they expect to spend on their cats and dogs this year. 😂 While this isn't very useful--especially since many people entered in clearly exaggerated numbers--sessions like this helped prove that our new 2.0 platform can handle even "extreme" usage situations. (In reality, even if some enterprise customers have thousands of users in Vareto, only a handful are ever likely to be in the same model page at the same exact time given granular data permissions.) You can see more useful, realistic ways that finance and business teams can use Vareto multiplayer here: https://lnkd.in/gV9gqKwz

123

22 Comments

GPTZero has skyrocketed to profitability within 18 months, raking in millions. With 500% ARR growth in six months and user growth from 1 million to 4 million in a year. While many AI-detection companies struggle with accuracy, GPTZero excels, leveraging vast data and advanced LLM models using the latest open-source tools. Initially celebrated for helping teachers spot AI-generated student work, GPTZero's clientele now includes government agencies, grant writers, hiring managers, and AI training data labelers. One founder noted, "We have a big data advantage with millions of examples of human versus AI-generated text," proving the quality of data fed into AI models is crucial. https://lnkd.in/gREnQ2ps #ai #sustainabledevelopmenttechnology #emergingtechnology

5

Over the last year and a half, I've worked with a handful of client accounts and helped them decide whether to integrate closed-source or open-source models in their applications. More times than not, the ridiculous amount of computing power needed to run self-hosted models at scale made it an unrealistic option for many. On top of that, fine-tuning the models to meet specific needs without compromising proprietary data added another layer of difficulty. The introduction of Meta Llama 3 is really exciting for a few reasons. This model's design for flexible hosting across multiple cloud platforms like AWS and Azure, combined with its advanced fine-tuning capabilities, directly addresses these pain points. Businesses can now customize the AI to their particular environments and data sets more easily and securely. Additionally, with built-in robust safety tools such as Llama Guard and CybersecEval, Llama 3 enhances the security and reliability of AI deployments. This makes powerful AI technologies not only more accessible but also more practical, enabling our clients to innovate while controlling costs and protecting sensitive data​. While the advancements in AI like Meta Llama 3 are impressive, the decision to build (self-host and fine-tune) or buy still isn't always cut and dry. Factors such as speed to market, integration complexities, and ongoing maintenance costs must still be considered. This environment of technological evolution presents more options and flexibility than ever before, empowering businesses to select solutions that align closely with their strategic goals and operational needs. I must be a bit crazy because, amid all of this chaos, I'm more excited than ever to explore ALL OPTIONS and help businesses navigate this promising but nuanced landscape. More on Llama 3: https://lnkd.in/gmMYtJkP

14

Suno raised $150M. Udio raised $10M. Investors are eager to get in these generative AI music startups, but these rounds highlight some big questions: - What is the data that these products trained on? Was that data licensed? - Are startups in music incentivized to ask for permission (instead of forgiveness)? - Who is the audience for these products? Is this what that audience wants? - Is this music's "Instagram moment" that the industry talks about? - Will these products become real businesses? Or are they features? - How will this all play out? To break it all down, I was joined by Tatiana Cirisano. Listen to our episode on Trapital "The Generative AI Music Sweepstakes" here: https://lnkd.in/eN8xWdZJ What are your thoughts on these companies and the products they're building?

113

15 Comments