Ben Perlmutter

Technical writer and web dev. Highly caffeinated. Definitely not a Russian troll, possibly an advanced AI agent

[Ben Perlmutter @bpmutter]

@dwarkesh_sp The David Mitchell universe deserves a mention. Thousand autumns of Jacob de Zoet, Utopia Avenue, and Cloud Atlas are particular standouts

[Ben Perlmutter @bpmutter]

@AstleDsa Amazing browser extension opportunity to ID any latex formulas and convert them to understandable text and/or code in programming language of your choice

I imagine an MMLU>85-tier LLM could do this pretty consistently

[Ben Perlmutter @bpmutter]

@paulg See

[Dylan O'Sullivan @DylanoA4]

Ah

[Ben Perlmutter @bpmutter]

My the past 6 months. Only became a problem when I stopped using my kindle interestingly 🤔

[Dylan O'Sullivan @DylanoA4]

Ah

[Infinite Books @infinitebooks]

C. S. Lewis on reading, this is it

[Dylan O'Sullivan @DylanoA4]

C. S. Lewis on why we need to read old books

[Ben Perlmutter @bpmutter]

@polynoamial @OpenAI Can’t wait to try this out with tool calling. Give it tool with root access to a Linux terminal and let er rip

[Ben Perlmutter @bpmutter]

@AndrewCurran_ pig butchering operations is the first thing that comes to mind. so maybe there'll be data centers full of h100s in the burmese jungle instead of neo-slavery camps...which i think is a form of relative progress if not utopia

[Ben Perlmutter @bpmutter]

@AndrewCurran_ my guess is that the initial use cases for such agentic systems will be pretty sketchy

[Ben Perlmutter @bpmutter]

this all leads to the natural Big Solution to solve American politics:
.
.
.
teach a semester of linear algebra in high school

[Ben Perlmutter @bpmutter]

A quarter-baked linear algebra theory of politics: All political phenomena exist in very high dimensional vector space and the left-right spectrum is a projection created by the powerful to classify people into binary groups.

[Ben Perlmutter @bpmutter]

By not including issues of greater importance in the projection, this allows the powerful to go on doing whatever they want on these issues (e.g. drug price gouging) while folks battle it out on the less important issues in the projection

[Ben Perlmutter @bpmutter]

So how to programmatically optimize content for models?

[Ben Perlmutter @bpmutter]

Re rumors about open ai strawberry / Orion, having an AI system that can effectively reason over long horizons will make it even more important that AI systems have access to high quality technical knowledge so that they can perform task that aren’t in their training data

[Ben Perlmutter @bpmutter]

Web search and sites exposes this knowledge for human users to reason over. And LLMs can work w web content . But it’s not optimized for them to interact w programmatically