The Only Important Technology Is The Internet — Kevin Lu
Although progress in AI is often attributed to landmark papers – such as transformers, RNNs, or diffusion – this ignores the fundamental bottleneck of artificial intelligence: the data. But what does it mean to have good data? If we truly want to advance AI, instead of studying deep learning optimization, we should be studying the internet. The internet is the technology that actually unlocked the scaling for our AI models. Inspired by the rapid progress made by architectural innovations (in 5 years, going from AlexNet to the Transformer), many researchers have sought better architecture priors. People bet on if we could devise a better architecture than the transformer. In truth, there have been better architectures developed since the transformer – but then why is it so hard to “feel” an improvement since GPT-4? Compute-bound. Once upon a time, methods scaled with compute, and we saw that more efficient methods were better. What mattered was packing our data into the models as effici
The Only Important Technology Is The Internet How can we continue to scale large language models? Published July 2025 Although progress in AI is often attributed to landmark papers – such as transformers , RNNs , or diffusion – this ignores the fundamental bottleneck of artificial intelligence: the data. But what does it mean to have good data? If we truly want to advance AI, instead of studying deep learning optimization, we should be studying the internet. The internet is the technology that actually unlocked the scaling for our AI models. Transformers are a distraction Inspired by the rapid
Explore this link on the map →saved by
- Winnie Xu
- Blake Elias
- Anson Yu
- Kaylee George
- Ratan Kaliani
- Linda
- Freeman Jiang
- Samuel Lo
- Hangyul Lyna Kim
- Aileen Luo
- Nhật Nguyễn Thế
- Sidney Nimako
related reading
- There Are No New Ideas in AI… Only New Datasetsblog.jxmo.io
- The Era of Experience Paper.pdfstorage.googleapis.com
- Andrej Karpathy — AGI is still a decade awaydwarkesh.com
- Questions about the Future of AI - by Dwarkesh Pateldwarkesh.com
- RLHF: Reinforcement Learning from Human Feedbackhuyenchip.com
- State of Data (Jan 2026)seancai.com
- Dario Amodei — "We are near the end of the exponential"dwarkesh.com
- How to scale RL to 10^26 FLOPs - by Jack Morrisblog.jxmo.io
- A Stargate for Data - by willdepue - Will DePuewilldepue.substack.com
- GenAI Handbookgenai-handbook.github.io
- Reinforcement learning, AI, and general intelligenceartfintel.com
- What I've Learned About AI in the Past Two Months.sheracaolity.ghost.io