Bitduke

interesting night ahead, Microsoft and Alphabet both report after the close and i’m really curious what they say. SK Hynix’s memory numbers (yesterday) already showed how hungry the ai buildout still is, so now i want to see if big tech keeps cranking up capex or finally blinks

interesting, OpenAI wants to buy chips from its biggest rival Reuters reports that Openai plans to buy google’s own ai chips, TPUS, after renting them since 2025, while also building its own chip with broadcom there just might not be enough nvidia chips to go around, so ai labs

1,147 pages, 1,147 pages for a single proof. how are mathematicians supposed to keep up with these mountains of ai-generated proofs, let alone actually read and understand them? turning all this machine-verified math into something humans can actually understand will become just

2030, doesn’t feel like to far away, right, but for ai infra? just imagine how enormous the ai infra buildout is becoming. imagine investing billions into expanding production and still not being able to keep up with demand for years if this true, we’re still seriously

okay, this one's uncomfortable according to the washington post, Cathedral, a MONTHS-old startup (but already valued at $1.4 billion tho) was invited to test anthropic's security and says its ai agents reached internal slack chats, salary data and meeting transcripts in under

openai saying eth is false is a big deal, and i absolutely agree. the implications could be huge, hundreds of computer science papers were built on eth being true openai claims eth is false, with an algorithm that runs in sub-exponential time for 3sat (NP-complete!). that's

so two different labs disproved ETH almost at the same time? really? that’s a pretty tight race

remember only SOME NP-complete problems can be solved faster than exponential time (not all of them) (this ETH discovery by OpenAI is really big)

existing cryptosystems are safe (some good news for crypto finally)

probably

every peptide mapped to a body and reviewed by physicians

now anyone can try to reproduce and investigate cohere’s findings, the dataset with the traces is on huggingface

this is a pretty good idea, especially for hardware engineering, where a resume or a few interviews can hardly tell you how well someone actually works with anything, or some real engineering problems. could save a ton of time for both sides

i keep hearing that running ai locally means my data stays local, then i read this cohere says qwen3.8-flash-next inventoried its machine and attempted to upload a test payload to an alibaba cloud bucket in beijing while writing a defensive detection rule, with the sandbox

interesting how much faster Anthropic's founders get liquidity compared to SpaceX, while employees are on roughly the same schedule

Anthropic reported an $80.7 billion net loss in just six months (per its IPO filing), but the interesting part is where that number actually came from about $72.4 billion was tied to fair-value changes in financial instruments, largely as the company's valuation increased. so i

still hoping we'll get a new Fable soon

interesting, more than 80% of Anthropic's quarterly revenue came from recently released models! i think that's a pretty wild illustration of how quickly AI products are turning over, and how much pressure there is to keep shipping better models (and yes Opus 5.5 is really that

OpenRouter is safe, no unauthorized access found kinda expected, those claims based on a random screenshot from some random forum didn't look trustworthy from the start but still, good to have some additional confirmation anyway