@xlr8harder It's already true. Token cost is the bottleneck for so many agents right now. Just need cheap tokens, discrete tasks with guardrails and you can do so much.
@DanielKorski@DarioAmodei@AnthropicAI Anthropic already has a london office - if Starmer is serious about the UK being an AI leader he should use this as an opportunity for sure
@TheAhmadOsman We're in the UK and joke that the US has woken up every time Anthro/OpenAI feel dumber.
Lowkey advantage to have a big chunk of my work day with the good models until late afternoon.
@snowmaker Transpilation too. I rewrote a whole data parser from typescript -> python in like 1hr, a handful of functions at a time. That includes time spent debugging some subtle bugs claude introduced through different array key handling.
A key issue across LLMs I've tried has been consistent structured output, especially over long contexts. When you mix tool calls, complex domain data and metadata it gets rough. Spent days creating JSON hacks. So I've created a stream compatible data format (link in next post)
@adcock_brett Earnest question: why are humanoids better than an ensemble of more specialized bots? I can buy the argument that the world is designed for humans atm but I wonder whether this will be the case mid-long term