Scaling Compute on Context — Jack Morris, Engram
Jack Morris from Engram explains why it is difficult to get AI models to understand private company data.
Models are trained on public internet data and perform very well there, but when you try to teach them your own data, they often become nearly useless when generating results. He describes three classical ways to improve this—more data, more computing power, and larger models—but says that when it comes to private data, only computing power is a viable option. He goes through various techniques such as KV-compression and synthetic training data, but shows that they all plateau once the model has learned everything you've given it, and there is no magical way to continue improving with more computing power.
Vibekollen prepared this summary with AI from the original publication. The content belongs to AI Engineer.
More from AI Engineer
Inside Arca's Long-Horizon Agents That Run the Business - Aman Singh, Arca
AI Engineer 23 Oct
Real AI Agent Attacks and What Would Have Stopped Them — Andrey Rybka, Bloomberg
AI Engineer 23 Oct
Outside of tech there has been no financial payoff to US companies for using AI, now viewed as a margin-hurting cost center, as VC-backed firms burn more tokens
Techmeme 4 h ago
Source: Replit recently moved most AI spend from Anthropic to OpenAI, which gave 30%+ off in a deal worth $100M+/year, after Anthropic discontinued its discount
Techmeme 6 h ago