Daniel Campos

The most interesting part about google's earning is despite click volume for organic content dropping like a sick beat,there hasn't realy been a decline in ad revenue. Either price per click went up a lot (nope it didn't) or ads are still seing similair engagement rate despite AI

Itrying to send and hype the dopeness of @dwarkesh_sp podcasts but iPhone keeps auto correcting his name to fearless. What a dope Freudian slip. I wish apple thought I was fearless. Fuck them spiders.

New trait of business metric recruiters use to tell you how hot a company is. Annualized token production/processing rate. I had 3 separate instances of hearing “annualized token rate is 1T a year”

Holy AirBall to those whos said we were running out of data. "At I/O in May, we announced that we processed 480 trillion monthly tokens across our surfaces. Since then we have doubled that number, now processing over 980 trillion monthly tokens, a remarkable increase."

Bonkers than substacks ARR is less than the salary for an AI engineer. Ironically that same AI engineer got famous by posting on substack.

Odds on Llama 5 being open source is getting pretty close to 0 techcrunch.com/2025/07/16/met…

Wake up bae. Rip vector db aws.amazon.com/blogs/aws/intr…

Its that time of year again, surveys for @airstreet's @stateofaireport. Make the data real by filling out their survey on GenAI usage. Moar data points == moar realz airstreet.typeform.com/survey

Claude vs Grok. Who did it better

This is a reminder to all the people calling RAG dead. RAG was the hip term describing embedding to LLM. Before the term got VCefied people were talking about 'in-context learning' and 'in-context reasoning' seems we are back. Death to bad acronyms

Here is your reminder not to blindly trust deep research reports. Had Opus kick of one on what kind of things in research Two Sigma the Hedge Fund had been doing. "RAG Systems: Development of retrieval-augmented generation systems for financial applications" coming from doc below

Note to self. Dont give LLM cat facts. It makes them have dog shit answers arxiv.org/pdf/2503.01781

Claude has maintenance on features, yet there is nowhere to figure out what 'this' refers to. All you can do is click X and hope they fix it.

My favorite part of vibe coding is reading the documentation models create. Everything is 'revolutionary' or 'groundbreaking'. Chill out, bro; it's just a TypeScript web app.

Reading through a blog really summarizes my feelings with some tool use recently. Models really do better with extracted content than tools, because when they use tools, they focus more on using the tool than using the context.

Who the heck is Wu-Wallace? Seems Like a pretty interesting way to grand Dylan the majority of voting rights without giving him the entire Class B shares.

I am trying out this Thought-Boi Thing. Give it a read. The Hidden Cost of Augmentation: Every Tool You Use Changes You. open.substack.com/pub/spacemanid…

You didn't have to be a genius to see it, but the data is unambiguous. People stop clicking on websites with AI overviews. Websites that have no monetization plan besides 'ads' will shrivel up and die. Long websites with strong brands and paid content like @theinformation

$TSLA deliveries down 14%. 2nd Year declines. Stock still goes up. Elon magic.

Wild that anyone could say that OpenAI folks are missionaries. Sure, compared to the Meta folks, but that's like a 6'10 dude telling people he is short. The price of perspective