Summary of Key Points
Silicon Valley billionaire Databricks and Perplexity co-founder Conwinski has invested $100 million to establish the Laude Institute, urging top young AI researchers to stay in academia or engage in open-source research rather than joining large corporations. The underlying issue is the paralysis of the public AI research ecosystem: the computing power and talent required to produce AI knowledge (such as research papers) have been monopolized by large companies. Universities and national funds lack the resources to afford this expensive infrastructure, while large companies use high salaries and resources to attract researchers but lock in their findings within their own walls. Conwinski aims to help academia break through with “lightweight system innovations.” The Laude Institute is accompanied by a venture capital fund, creating a “charitable incubation + commercial exploitation” dual-model approach. However, whether this model can replicate the success of institutions like Bell Labs faces significant challenges, including the allure of high salaries offered by large companies.
1. Why did Conwinski invest $100 million to discourage researchers from joining large companies? Public AI research is on the verge of collapse
Cutting-edge AI knowledge should be a “public good”—for example, an algorithm formula written in a paper can be used by anyone without affecting others. However, the resources needed to generate this knowledge (computing power and talent) are now controlled by large corporations. Training large models requires tens of millions or even hundreds of millions of dollars in hardware and electricity costs, which universities cannot afford. The National Science Foundation (NSF) in the U.S. allocates so little funding that it’s not enough to even purchase a few high-end GPUs for each professor.
Large companies have taken advantage of this situation, offering professors million-dollar salaries and computing power in exchange for locking in their research findings. For instance, OpenAI has shifted from open-source to proprietary development, and DeepMind’s research is now integrated into Google products. Conwinski’s investment aims to free academia from the control of large companies and ensure that AI knowledge remains openly accessible.
2. Academia doesn’t need to compete with large companies for computing power! “Post-training” represents a breakthrough
The development of AI goes through three stages:
- Pre-training: Similar to building a foundation, which requires vast computing resources; academia cannot handle this.
- Post-training: Tasks like RLHF (Reinforcement Learning with Human Feedback) require extensive data labeling and engineering expertise, where large companies still have an advantage.
- Post-post-training: After the model is created, the next step is to make it more intelligent and efficient—this involves optimizing the instructions for the model and developing neutral evaluation platforms.
This third stage is where academia can play a key role. Projects like Stanford’s DSPy and Berkeley’s LMArena focus on optimizing model instructions without the need for expensive training; they serve as critical components of AI systems. The first batch of projects funded by Laude (the “Slingshot Initiative”) aim to leverage lightweight innovations in this area, representing an asymmetric battle between academia and large companies.
3. Laude’s “dual-model” approach: Charitable funding for incubation, venture capital for commercialization?
Laude consists not only of a non-profit research institute but also of a venture capital fund with the same name. The strategy is as follows:
- The institute initially uses charitable funds to support open-source projects and young researchers, identifying the most promising cutting-edge research.
- Once the projects mature and become commercially viable, the venture capital fund invests to generate profits.
This approach replicates Conwinski’s own success with Databricks, which originated from an open-source project at Berkeley and later grew into a company worth billions of dollars. The advantage is that it ensures the continuity of open-source research (without funding constraints). However, there are concerns: Could this model lead to commercialization similar to OpenAI’s, initially using non-profit rhetoric to gather support before turning the results into profit? Will the evaluation criteria for researchers shift from serving large company shareholders to satisfying this billionaire and his investment managers?
4. Can we replicate the glory of Bell Labs? The reality is too daunting
Many of history’s major breakthroughs (such as transistors, the UNIX operating system, and Ethernet) came from laboratories within monopolistic companies (like AT&T’s Bell Labs). Monopolistic firms had stable profits, allowing them to support talented researchers for decades without focusing on profit-making activities.
Conwinski aims to separate basic research from monopolies by using a charitable and venture capital model to create a “new Bell Lab.” However, the challenge is maintaining funding without the stability of monopoly profits. Projects must balance “selfless open-source efforts” with the pressure to generate commercial returns quickly. More realistically, will 25-year-old AI researchers give up stable high salaries to pursue a risky entrepreneurial opportunity? This highlights the dilemma for researchers: they understand the importance of open-source research, but the practical incentives are too strong.
In this context, Conwinski’s $100 million investment seems more like a “noble gesture” that highlights the issues in AI research. To truly change the landscape, additional stakeholders and time will be needed to prove whether this model is feasible.