![cover of episode 2024 in AI Startups [LS Live @ NeurIPS]](https://substackcdn.com/feed/podcast/1084089/post/153389370/2d1909e2fbbd5c9267a782756c04d8a3.jpg)
2024 in AI Startups [LS Live @ NeurIPS]
Latent Space: The AI Engineer Podcast — Practitioners talking LLMs, CodeGen, Agents, Multimodality, AI UX, GPU Infra and all things Software 3.0
Deep Dive
What are the top five themes in AI startups in 2024 according to Sarah Guo and Pranav Reddy?
The top five themes in AI startups in 2024 are: 1) A closer race in foundation models, with OpenAI no longer dominating as it did in 2023. 2) Open-source models becoming increasingly competitive, especially in areas like math and instruction following. 3) The price of intelligence dropping significantly, with OpenAI's API costs decreasing by 80-85%. 4) New modalities like biology, voice, and video beginning to work effectively. 5) The debate around the end of scaling, with new paradigms like test-time compute scaling emerging.
Why has the competitive landscape for foundation models shifted in 2024?
The competitive landscape for foundation models has shifted because OpenAI is no longer the clear leader. In 2023, OpenAI models were significantly better than others, but by 2024, proprietary and open-source models have become increasingly competitive. For example, Google's models are now outperforming OpenAI in some evaluations, and open-source models like LLAMA are among the top performers in specific areas like math and adversarial robustness.
How has the cost of AI intelligence changed in 2024?
The cost of AI intelligence has dropped dramatically in 2024, with OpenAI's flagship model API costs decreasing by 80-85% over the past year and a half. This trend is not limited to OpenAI; across the industry, the price per token for AI models has significantly decreased, making it more affordable for companies to leverage AI capabilities at scale.
What new modalities in AI are showing promise in 2024?
New modalities in AI showing promise in 2024 include biology, voice, and video. For example, Chai Discovery released an open-source biology model that outperforms AlphaFold3. Low-latency voice models are creating new interaction experiences, and video models like Sora and HeyGen are enabling advanced capabilities such as lip-syncing and dubbing for live speeches.
What is the current state of open-source AI models in 2024?
Open-source AI models have become increasingly competitive in 2024, particularly in areas like math, instruction following, and adversarial robustness. Models like LLAMA are now among the top performers in these domains. However, there are still areas, such as agenting and tool use, where proprietary models maintain an advantage due to more specialized training and data.
What are the key trends in AI startup funding in 2024?
In 2024, AI startup funding has seen a substantial recovery, with foundation model labs raising upwards of $30-40 billion. However, the broader funding environment remains rational, with most money going to companies demonstrating real traction and growth. The narrative of an AI bubble is largely debunked, as startups are raising funds based on actual outcomes rather than hype.
How are startups addressing the challenge of competing with incumbents in AI?
Startups are addressing the challenge of competing with incumbents by focusing on better products, innovative business models, and leveraging AI to create new user experiences. Incumbents may have distribution and data advantages, but startups are finding success by rethinking workflows, offering outcomes-based pricing, and building products that are more efficient and user-friendly than traditional solutions.
What are the emerging opportunities in AI for startups in 2024?
Emerging opportunities for AI startups in 2024 include first-wave service automation, better search and productivity tools, democratization of creative and technical skills, and enabling layers like compute and data. Startups are also exploring new markets traditionally considered difficult for venture capital, such as legal, healthcare, and education, by leveraging AI to make capabilities cheaper and more accessible.
How is the debate around scaling in AI models evolving in 2024?
The debate around scaling in AI models is evolving with the recognition that there are limits to the benefits of increasing scale. However, new paradigms like test-time compute scaling are emerging, where models can dynamically allocate compute resources based on the complexity of tasks. This approach is particularly effective in well-constrained domains like math and physics, but challenges remain in less-defined areas.
What is the role of open-source models in the AI ecosystem in 2024?
Open-source models play a significant role in the AI ecosystem in 2024 by providing competitive alternatives to proprietary models. They are particularly effective in specific domains like math and instruction following, and they contribute to lowering the cost of intelligence. Open-source models also enable startups to experiment and innovate without the high upfront costs associated with proprietary solutions.
- OpenAI's dominance decreased, with Google's models becoming increasingly competitive.
- Open-source models showed significant improvements in various areas.
- The cost of using flagship OpenAI models decreased by approximately 80-85%.
- New modalities like biology, voice, and video are showing promising results.
Shownotes Transcript