Andrew Yeolo (@AndrewYeolo) profile picture

VC Investor @OIFVC | DMs are open

Australia
Joined March 2021
Photo by Andrew Yeolo (@AndrewYeolo)Photo by Andrew Yeolo (@AndrewYeolo)Photo by Andrew Yeolo (@AndrewYeolo)Photo by Andrew Yeolo (@AndrewYeolo)
Strong margin expansion in tech and SaaS imminent due to 5x+ productivity unlock for dev teams from GPT-3+, so R&D as % of revenue will decline. Also startups can build MvP and product with way less devs. Booolish
1
9
905
Excited for the launch of @rocksolidHQ and its first official rETH vault with @Rocket_Pool! Thrilled to support @wardy_ben and @paladin_eth on their mission to democratise access to secure DeFi strategies as stablecoins proliferate and finance increasingly moves onchain, and to invest alongside @CastleIslandVC, @Rocket_Pool, @GSR_io, @kindredventures, @theBBFund, @ideoVC, and @StanfordSBA!
1/ Excited to share that RockSolid has raised $2.8M in pre-seed funding led by @CastleIslandVC w/ backing from @Rocket_Pool, @GSR_io, @kindredventures, @theBBFund, @ideoVC, @StanfordSBA, and others. Today, we're also launching the first official rETH vault with @Rocket_Pool.
5
4
883
Check out the @Rocket_Pool community call where they’ll chat about the @rocksolidHQ partnership and launch
Our next community call is starting soon in the Rocket Pool Discord: Thu 25 Sep @ 11:30pm UTC We'll chat with @rocksolidHQ about our partnership to supercharge $rETH 🤝 Plus, download the @poapxyz Home App & grab a free POAP when you join the call Link below 👇
3
119
Marc Andreessen explains how to identify fake founders “There are definitely people that come in [to pitch us] and present themselves to be something they’re not. They’ve read all the books. They will have listened to this interview. They study everything and they construct a facade…. And the amount of this is exactly correlated with the NASDAQ.” As Marc explains, when stock prices are high and tech is hot, there are a lot people who decide being a tech founder is a fast track to high status: “They’re fundamentally oriented for social status — they’re trying to get the social status without the substance. And there are always other places to go to get social status. So after 2000, the joke was B2B meant back to banking and B2C meant back to consulting — which is, the people who showed up to be in tech were like, yeah, screw it. This is over. I’m going to go back to Goldman Sachs or McKinsey where I can be high status. So you get this flushing kind of effect that happens in a downturn. But in a big upswing, you get a lot of people showing up with, let’s say, public persona without the substance to back it up.” How does Marc identify these people? He uses the same technique that homicide detectives use to find out if you’re innocent — keep asking increasingly detailed questions: “You ask increasingly detailed questions and people have trouble making things up and things just fuzz into obvious BS, and fake founders basically have the same problem. They’re able to relay a conceptual theory of what they’re doing… But as they get into the details, it just fuzzes out. Whereas the true people that you want to back can do it. What you find is they’ve spent 5 or 10 or 20 years obsessing over the details of whatever it is they’re about to do. And they’re so deep in the details and they know so much more about it than you ever will.” Video source: @hubermanlab (2023)
50
213
63
2,424
346,009
Andrew Yeolo retweeted
Henlo — we’re live on Berachain! 🐻 Jump into Reservoir Market (link in bio) to: - Discover trending NFT collections - Easily mint your favorite Beras - Trade NFTs on Secondary And it’s all powered by Reservoir NFT.
2
3
1
48
4,746
Great take @AndrewYNg. Seems that foundation models are commoditising and value will accrue to the app/software layer as you’ve been saying. There should still be a lot of $ for the foundation layer similar to cloud storage and compute. It’s incredible what innovation occurs under constraints which @deepseek_ai showed I also think that @Google , @Meta and @x will win the foundation model war (if they can justify the RoI). The best models advance via compute, huge amounts of data (G suite, YouTube, FB, insta, Twitter), and algorithmic breakthroughs (as per deepseek). @Microsoft has swarms of enterprise data but lacks real time consumer data, which could explain their @tiktok_us interest (together with a potential bargain acquisition price) Agree that demand for intelligence will have no ceiling and more efficiency will see more consumption ie. jevons paradox teapawt.net/andrewyng/status/18850…
The buzz over DeepSeek this week crystallized, for many people, a few important trends that have been happening in plain sight: (i) China is catching up to the U.S. in generative AI, with implications for the AI supply chain. (ii) Open weight models are commoditizing the foundation-model layer, which creates opportunities for application builders. (iii) Scaling up isn’t the only path to AI progress. Despite the massive focus on and hype around processing power, algorithmic innovations are rapidly pushing down training costs. About a week ago, DeepSeek, a company based in China, released DeepSeek-R1, a remarkable model whose performance on benchmarks is comparable to OpenAI’s o1. Further, it was released as an open weight model with a permissive MIT license. At Davos last week, I got a lot of questions about it from non-technical business leaders. And on Monday, the stock market saw a “DeepSeek selloff”: The share prices of Nvidia and a number of other U.S. tech companies plunged. (As of the time of writing, some have recovered somewhat.) Here’s what I think DeepSeek has caused many people to realize: China is catching up to the U.S. in generative AI. When ChatGPT was launched in November 2022, the U.S. was significantly ahead of China in generative AI. Impressions change slowly, and so even recently I heard friends in both the U.S. and China say they thought China was behind. But in reality, this gap has rapidly eroded over the past two years. With models from China such as Qwen (which my teams have used for months), Kimi, InternVL, and DeepSeek, China had clearly been closing the gap, and in areas such as video generation there were already moments where China seemed to be in the lead. I’m thrilled that DeepSeek-R1 was released as an open weight model, with a technical report that shares many details. In contrast, a number of U.S. companies have pushed for regulation to stifle open source by hyping up hypothetical AI dangers such as human extinction. It is now clear that open source/open weight models are a key part of the AI supply chain: Many companies will use them. If the U.S. continues to stymie open source, China will come to dominate this part of the supply chain and many businesses will end up using models that reflect China’s values much more than America’s. Open weight models are commoditizing the foundation-model layer. As I wrote previously, LLM token prices have been falling rapidly, and open weights have contributed to this trend and given developers more choice. OpenAI’s o1 costs $60 per million output tokens; DeepSeek R1 costs $2.19. This nearly 30x difference brought the trend of falling prices to the attention of many people. The business of training foundation models and selling API access is tough. Many companies in this area are still looking for a path to recouping the massive cost of model training. Sequoia’s article “AI’s $600B Question” lays out the challenge well (but, to be clear, I think the foundation model companies are doing great work, and I hope they succeed). In contrast, building applications on top of foundation models presents many great business opportunities. Now that others have spent billions training such models, you can access these models for mere dollars to build customer service chatbots, email summarizers, AI doctors, legal document assistants, and much more. Scaling up isn’t the only path to AI progress. There’s been a lot of hype around scaling up models as a way to drive progress. To be fair, I was an early proponent of scaling up models. A number of companies raised billions of dollars by generating buzz around the narrative that, with more capital, they could (i) scale up and (ii) predictably drive improvements. Consequently, there has been a huge focus on scaling up, as opposed to a more nuanced view that gives due attention to the many different ways we can make progress. Driven in part by the U.S. AI chip embargo, the DeepSeek team had to innovate on many optimizations to run on less-capable H800 GPUs rather than H100s, leading ultimately to a model trained (omitting research costs) for under $6M of compute. It remains to be seen if this will actually reduce demand for compute. Sometimes making each unit of a good cheaper can result in more dollars in total going to buy that good. I think the demand for intelligence and compute has practically no ceiling over the long term, so I remain bullish that humanity will use more intelligence even as it gets cheaper. I saw many different interpretations of DeepSeek’s progress here in X, as if it was a Rorschach test that allowed many people to project their own meaning onto it. I think DeepSeek-R1 has geopolitical implications that are yet to be worked out. And it’s also great for AI application builders. My team has already been brainstorming ideas that are newly possible only because we have easy access to an open advanced reasoning model. This continues to be a great time to build! [Original text: deeplearning.ai/the-batch/is… ]
153
Andrew Yeolo retweeted
Important advice from Brother Kimbal here: be very careful with venture debt, it’s a ticking time bomb for startups. Having a small amount for financing hardware or as a safety net in a predictable business is just fine — but it’s not to be used to extend your runway.
Never use debt to extend your runway. Debt is a great tool for working capital. But if you use it to extend your runway, you are betting on both your execution and the timing of the markets. Instead, focus on raising the capital you need and take the lower valuation with good terms for your company. If your company succeeds, you will forget the lower valuation very quickly.
36
17
2
306
140,375
Andrew Yeolo retweeted
Wise words from a recent interview with @geoffreyhinton, one of the smartest people in the world regarding AI
6,036
24,541
2,526
100,548
37,797,129
Glad to see capitalism win and the compensation deal that shareholders agreed to many years ago honored. Many pundits ridiculed the performance targets tied to the deal as unachievable and too ambitious, but @elonmusk and @Tesla crushed the targets
Sending this cake to Delaware as a parting gift 😘
7
246
Heard from VC LPs this week: Some 👍🏻 feedback from my talk at @Jason ‘s @liquiditypod conference. Sharing a slide here that resonated. GPs, if hard to raise, this basic formula explains it. Complexity so often underestimated by GPs, but LPs see right through it…
20
34
5
273
79,779
Andrew Yeolo retweeted
“Tesla is far ahead in self-driving cars” – Jensen Huang
5,018
11,052
1,519
97,991
42,054,798
Andrew Yeolo retweeted
One of my biggest takeaways from the Llama3 announcement - data gravity is king. And access to high quality, annotated, LLM ready data will drive which companies truly utilize AI vs those who don't As the number of accessible, highly performant models continues to grow (both proprietary and open source), it becomes less important to ask "which model should I use or tie myself to" and more important to ask "how do I deploy / serve / train / fine tune / monitor a model on MY OWN data?" I don't think we'll see companies move data to the models. I think they'll move models to the data. It's easier - The gravity is with the data. It's not just about how you train / fine tune a model initially. It's really about how do you continuously re-train / fine tune the models you have in production as new data comes in, and monitor quality drift. It's much easier to do this if the models are sitting alongside the data. Data isn't stagnant, and models shouldn't be either. As many have shouted from the rooftops before, you don't have an AI strategy without a data strategy. And I think all of what I've just said is a big reason why. Data platforms have a lot of potential here. A big beneficiary here are they hyperscalers. Databricks has also done a killer job. Snowflake playing catchup, but probably done more than people give credit for So I think when looking looking to find big AI winners a key question is "who has the data?"
13
31
11
255
83,595
Peter Thiel: “Buzzwords are a tell, like in poker, that the company is bluffing and undifferentiated” “I think all trends are overrated… if you hear the words ‘big data’ and ‘cloud computing’, you need to run away as fast as you possibly can. Just think ‘fraud’ and run away… All these buzzwords are a tell—like in poker—that the company is bluffing and undifferentiated… I think the things that are underrated are the ones where there are no buzzwords and it doesn’t actually fit into any pre-existing categories” Video Source: @twistartups
20
208
43
1,300
324,917
Andrew Yeolo retweeted
Zooming out on hyperscaler revenue additions it's wild to see the ZIRP benefit to cloud providers so clearly on a chart: We're back to the pre-ZIRP trendline but consensus estimates have us growing well above on AI tailwinds. Looking specifically at $AMZN AWS you can see the even bigger benefit from digital natives, the cost optimizations, and a slow ramp of consensus back to the pre-ZIRP trendline by YE24 Contrast that to $MSFT Azure which saw little benefit from ZIRP but is seeing an even greater tailwind from AI: $GOOGL GCP is in between the two but looks more similar to Azure than AWS. If you take the same look at Infrastructure Software on a cumulative basis we're back at the pre-ZIRP trendline and have been for a couple quarters: $SNOW $CFLT $MDB $DDOG $ESTC Caveating the lol consensus estimates and games companies and sell-side play with guidance in software, there's a vastly different set of expectations for the hyperscalers vs infra software names.
3
7
4
71
32,499
Q* Leaked info: Source: An unspecified PasteBin.(L-I-N-K in next post) Can't confirm the authenticity as it's from unknown source but you can have a look. Q* is a dialog system conceptualized by OpenAI, designed to enhance the traditional dialog generation approach through the implementation of an energy-based model (EBM). Distinct from the prevalent autoregressive token prediction methods, Q* aims to mimic a form of internal deliberation akin to human thought processes during complex problem-solving, such as chess playing, where a deeper analysis of potential moves leads to better decision-making compared to rapid, less considered responses. This model shifts focus towards the inference of latent variables, reminiscent of constructs in probabilistic models and graphical models, fundamentally altering how dialog systems operate. Energy-Based Model for Dialog Generation At the core of Q* is the EBM, which operates by assessing the compatibility of an answer to a given prompt through a scalar output. This output signifies the "energy" of the response, where a lower value indicates a high compatibility (a better answer) and a higher value suggests low compatibility (a poor answer). This mechanism allows Q* to evaluate potential responses holistically, moving beyond the sequential prediction of tokens to understand the underlying relevance and appropriateness of an answer to the prompt. Optimization in Abstract Representation Space The innovation in Q* lies in its optimization process, conducted not within the space of possible text strings but in an abstract representation space. Here, thoughts or ideas are represented in a form that allows for the computational minimization of the EBM's scalar output, akin to finding the path of least resistance in a landscape. This process involves gradient descent, a method for finding the minimum of a function, applied to iteratively refine these abstract representations towards those that yield the lowest energy in relation to the prompt. From Abstract Thought to Textual Response Once an optimal abstract representation — one that minimizes the EBM's output — is identified, Q* employs an autoregressive decoder to transform this abstract thought into a coherent textual response. This step bridges the gap between the non-linguistic, conceptual understanding of the dialog system and the linguistic output required for human interaction. Training the System The EBM within Q* is trained using pairs of prompts and responses, adjusting the system's parameters to minimize the energy for compatible pairs while ensuring that incompatible pairs result in higher energy levels. This training process can incorporate contrastive methods, where the system learns to differentiate between compatible and incompatible pairs, and non-contrastive methods, which involve regularization techniques to control the distribution of low-energy responses across the space of all possible answers. Implications for Dialog Systems Q*'s approach, leveraging EBMs for dialog generation, represents a significant departure from traditional language modeling techniques. By optimizing over an abstract representation space and utilizing gradient-based inference, Q* introduces a more efficient, reasoned, and potentially more powerful method for generating dialog responses. This system not only promises improvements in the quality of generated text but also offers a blueprint for future advancements in AI's ability to engage in human-like reasoning and conversational interactions. Technical Considerations Q*'s effectiveness hinges on the intricacies of its EBM, the optimization landscape it navigates, and the accuracy of its abstract representations. The model's capacity to simulate deep reasoning, akin to human deliberation, sets a new benchmark for dialog systems. Furthermore, the method of training Q*—balancing the need for specificity in correct responses while avoiding the collapse of energy levels across diverse inputs—poses unique challenges and opportunities for AI research.
108
176
83
955
708,686
Andrew Yeolo retweeted
Starship will take humanity to Mars
23,260
31,690
4,363
449,541
76,451,713
Andrew Yeolo retweeted
Today we're excited to introduce Devin, the first AI software engineer. Devin is the new state-of-the-art on the SWE-Bench coding benchmark, has successfully passed practical engineering interviews from leading AI companies, and has even completed real jobs on Upwork. Devin is an autonomous agent that solves engineering tasks through the use of its own shell, code editor, and web browser. When evaluated on the SWE-Bench benchmark, which asks an AI to resolve GitHub issues found in real-world open-source projects, Devin correctly resolves 13.86% of the issues unassisted, far exceeding the previous state-of-the-art model performance of 1.96% unassisted and 4.80% assisted. Check out what Devin can do in the thread below.
4,053
9,493
12,433
42,108
31,493,037
Andrew Yeolo retweeted
Let's compare the gas to bridge to Blast: - via Blast Bridge = $62.27 - via Reservoir Relay = $5.84 More than 10x cheaper!! relay.link/bridge/blast/
10
18
10
120
21,813
Andrew Yeolo retweeted
The median growth adjusted software multiple (NTM rev multiple / NTM rev growth rate), for a basket of ~100 software companies, is currently at an all time high (excluding '20-'21) The market appears to be expecting growth to reaccelerate - will we see this in Q4 earnings?
3
5
1
61
17,204