peterjliu

it seems not just the 'middle', but also the 'lower' classes are getting removed

I mentioned a potential OpenAI insider in https://x.com/peterjliu/status/2024901585806225723, that was from 5 minutes of investigation. There are probably more. And then there's a lot of other companies.

Post author here: To clarify, this is not a post from Polymarket.

This is talking about using Compound AI (product I'm working on) to query Polymarket data, including finding insiders, just as a fun example analysis you could do.

Often you need a well-calibrated probability of a future event to feed into some other analysis, and Polymarket is pretty great for that. An example is how much insurance (hedge) to buy for some disastrous event.

Some people have better data, like insiders.

Some have better models that predict with higher accuracy, given the same data.

emacs and vim are not niche, lol

seems like misinformation for AWS. CloudFlare probably depends on GCP.

interesting are LLMs a lot better at Go than Rust?

another advantage is people want the Google bot to crawl their pages, unlike most AI companies

From documentation: "TLDR; Agentic applications needs both A2A and MCP. We recommend MCP for tools and A2A for agents."

Agents can just be viewed as tools, and vice versa. Is this an attempt to save the launch after getting scooped by MCP?

I would start by making the examples yourself initially, assuming you have a good sense for what that real-world task is. If you can't articulate what a good task is and what a good output is, it is not ready for out-sourcing to crowd-workers.

And before going to crowd-workers (maybe you can skip them entirely) try LLMs.

We've (ex Google Deepmind researchers) been doing research in increasing the reliability of agents and realized it is pretty non-trivial, but there are a lot of techniques to improve it. The most important thing is doing rigorous evals that are representative of what your users do in your product. Often this is not the same as academic benchmarks. We made our own benchmarks to measure progress.

Plug: We just posted a demo of our agent doing sophisticated reasoning over a huge dataset ((JFK assassination files -- 80,000 PDF pages): https://x.com/peterjliu/status/1906711224261464320

Even on small amounts of files, I think there's quite a palpable difference in reliability/accuracy vs the big AI players.

The code can be used to train on other data. All you really need is a collection of news articles. I think there are some free ones available.

This dataset was only used to benchmark against other published results. It was first proposed in https://arxiv.org/abs/1509.00685.

We're not "selecting" sentences as an extractive summarizer might. The sentences are generated.

As for how does the model deal with co-reference? There's no special logic for that.

Author of post here. I'd say most of the examples generated from the best model were good. However we chose examples that were not too gruesome, as news can be :)

We encourage you to try the code and see for yourself.