Brb gonna go bootstrap a global logistics operation and hire 1.5M people to take on Amazon on the off chance this hypothesis is true and Amazon doesn’t just change their mind while I build a generational company to defeat them.
What if scaling the context windows of frontier LLMs is much easier than it sounds?
We’re excited to share our work on Recursive Language Models (RLMs). A new inference strategy where LLMs can decompose and recursively interact with input prompts of seemingly unbounded length, as a REPL environment.
On the OOLONG benchmark, RLMs with GPT-5-mini outperforms GPT-5 by over 110% gains (more than double!) on 132k-token sequences and is cheaper to query on average.
On the BrowseComp-Plus benchmark, RLMs with GPT-5 can take in 10M+ tokens as their “prompt” and answer highly compositional queries without degradation and even better than explicit indexing/retrieval.
We link our blogpost, (still very early!) experiments, and discussion below.
This new DeepMind research shows just how broken vector search is.
Turns out some docs in your index are theoretically incapable of being retrieved by vector search, given a certain dimension count of the embedding.
Plain old BM25 from 1994 outperforms it on recall.
1/4
The beatings (free books) will continue until everyone looks at their data:
1. LLM Evals FAQ:
https://t.co/BzEHwvobz5
2. Beyond Naive RAG: Practical
Advanced Methods
https://t.co/x2870kdHoZ
My bar for AGI is doing my taxes. Including gathering the materials, requesting necessary info from IRS, e-filing, checking and notifying me when return deposited.
I would measure how much wealth Musk created that never existed before.
Then I'd calculate what percentage of that wealth he kept.
Then I'd figure out what percentage of it he spent on his own lifestyle versus paid in taxes.
Then I'd calculate the incomes of the employees of his companies and their impact on the economy.
Then I'd look at the societal benefits of his companies because it's not all about cash.
Then I'd compare to the benefits Bernie Sanders brought to America. (Mostly complaining.)
Target relevant queries in Perplexity with “agentic” intent to prompt users to download Comet to automate their task. (Versus an always-on, context-unaware prompt on perplexity to download that users will learn to ignore). E.g., grocery related query -> “download Comet for your personal grocery shopping assistant”.
@AravSrinivas Latency feels super high for questions like ‘movie X streaming app’, for like simpler informational queries. May be a lightweight classifier to fast-track simpler queries and reduce latency. Don’t mind higher latency for more complex queries.
i've been trying to understand the transformer architecture for ages but it never clicked. so recently i started trying to learn its history and that has helped a LOT. to try to cement what i learned imma write up a essay.
Undefined Behavior presents: The Evolution of Attention
ChatGPT advanced voice mode does exactly this. Productivity hack for deep thinking + brainstorming alone is to go on a long drive in Tesla FSD to keep away from Twitter/doom scrolling, then think out loud to ChatGPT voice mode, which is a fully hands-free, great brainstorming buddy.
@sama Advanced voice mode with 4o is 🔥🔥. Hitting rate limits everyday with plus account. Take my money ($50/month tier?) for unlimited voice mode access. Don’t need pro level deep research.