open weights
AI model parameters that are publicly released so that anyone can run, audit, fine-tune, or modify the model themselves.
As defined in Chris Dixon
Other definitions
Release of trained model parameters publicly so others can run, modify, and train their own versions of the model.
Tobi Lütke - Building Islands of Innovation - [Invest Like the Best, EP.394]
Where it comes up · 7
- Deepseek Dylan Patel Nathan Lambert5 claims
- The AI Revolution Is Underhyped | Eric Schmidt | TED3 claims
- Marc Andreessen - The Battle For Tech Supremacy - [Invest Like the Best, EP.410]3 claims
- Chris Dixon2 claims
- Conversations with Tyler - Luis Garicano2 claims
- Mark Zuckerberg — Llama 3, $10B models, Caesar Augustus, & 1 GW datacenters2 claims
- Ex-Google CEO: What Artificial Superintelligence Will Actually Look Like w/ Eric Schmidt & Dave B2 claims
Most important claims · 10 of 19Search all claims for “open weights” →
- A concentration of AI power in one institution has the potential to be as dangerous as AI being widespread; it is probably bad for one institution to have an AI far more powerful than everyone else's, so open source helps prevent dangerous concentration.Mark Zuckerberg — Llama 3, $10B models, Caesar Augustus, & 1 GW datacenters
- Whether a country retains sovereignty when adopting foreign AI depends critically on architecture: if a nation like Peru can obtain the open weights of a model like DeepSeek and fine-tune or change those weights, it preserves meaningful control, whereas a closed proprietary model embedded as a service leaves the country effectively dependent on the foreign company.Chris Dixon
- Politics is downstream of culture and information flows, so as AI becomes the source from which people get their news and political information, the key question becomes who created that AI and whether it is open source and auditable; strong open-source AI is the single most important factor for making the AI future better because it gives people choice and allows the systems shaping news and culture to be audited.Chris Dixon
- As AI models become more capable, they can embed not just literal code backdoors but ideological alignment ('cultural backdoors') deep in the weights that proliferate when open-weights are released; because models slot into a familiar social conversational context, superhuman persuasion may arrive before superhuman intelligence, letting model-makers subtly shape opinions.Deepseek Dylan Patel Nathan Lambert
- Open weights are large matrices of numbers you can download and run on a computer with no internet, meaning the model itself cannot steal your data; data privacy risk comes from whoever hosts the model (China via the DeepSeek app, Perplexity, or OpenAI), not the weights themselves.Deepseek Dylan Patel Nathan Lambert
- DeepSeek-R1 with its permissive MIT license is the first time a clearly frontier-level model has been openly available with a commercially friendly license and no downstream use-case restrictions in the post-ChatGPT era, which resets expectations for open AI; this means capable open models will keep coming regardless of whether anyone wants to stop them.Deepseek Dylan Patel Nathan Lambert
- Frontier AI models are needed for research, innovation, and some military applications, but good law, medicine, accounting, and professional services can run on models three or four months behind the frontier, provided the harness and data are in-house.Conversations with Tyler - Luis Garicano
- The narrative that using DeepSeek means handing data to the Chinese is true only for DeepSeek's own website, because downloaded open weights can run on your own hardware or on US-hosted clouds like Perplexity, Microsoft, and Amazon.Marc Andreessen - The Battle For Tech Supremacy - [Invest Like the Best, EP.410]
- Neither DeepSeek nor the closed labs OpenAI and Anthropic reveal their training data, and from downloaded model weights it is very difficult or impossible to derive the underlying training data.Marc Andreessen - The Battle For Tech Supremacy - [Invest Like the Best, EP.410]
- Open-source models with open weights represent a strategic vulnerability because every country outside the West will perceive open-source as cheaper and will use it, transferring AI leadership in open-source from America to China.Ex-Google CEO: What Artificial Superintelligence Will Actually Look Like w/ Eric Schmidt & Dave B