Tony Tong (@GoProAI) header
Tony Tong (@GoProAI) profile picture

CTO and Chief AI | Unfazed Tesla Investor

California, USA
Joined November 2009
The Economics of Context: Cost-Optimal Prompt Management for Long-Running LLM Agents synthfoundry.substack.com/p/…
1
177
The problem is the temperature the radiator has to operate at.
4
5
4,598
In my case, per task Terra (6.1) costs more. Tokens were attributed to each mission from the brief, and turned into weekly-quota points with weights fitted over 45 hours of quota readings (R² 0.82): 3.4 points per million output tokens, 0.17 per million new input tokens, 0.015 per million cached input tokens. The fit, made mostly on Sol hours, predicted Terra’s two Terra-only hours within half a point, so Terra is not priced higher per token.
🚨 GPT 6.1 Terra Is The Cheapest Frontier Model Per Task - cheaper than other frontier models - comparable to Argon and Sonnet in total score - very good at browser use A really good fit for personal agents and as a chatbot
1
135
How it started... and how it ended after 1.5 hrs of self play. This is why you should be a long term investor 🤣
83
Calabi-Yau manifold (n=6) 🤟
1
87
What a shift!
94
Need a Mad Max plan
171
I think it is a shift of the constraints. For terrestrial data centers, if we are good with natural gas power plants, or even solar+battery, we can get as much power as we want - quickly and cheaply.
Or maybe do some back of the envelope calc?
2
196