why aren't all 64 resolved? a statement only gets an assessment when the public
record can support or contradict it. opinions and what-ifs never can, and 3 checkable
ones are still open, waiting for their date. predictions held up or didn't;
assertions are supported or contradicted. on every card:
▮▮▮▮▮ certainty ·
▮▮▮▮▮ debate potential. speakers are clickable
Insight
AI API Costs Introduce Significant COGS to SaaS for the First Time
“I think this is probably the first time for many SaaS companies to have significant cogs. Usually infrastructure has become very cheap with AWS and other cloud providers. So usually it's not something that you have to think about a lot. But now if you're using…”
Assertion Partly supported
Lemkin: Replit deploys an OpenAI-powered Architect sub-agent for big features
“We actually get an extra bonus model because when Claude on Opus talks to Replit on Sonnet, if it's a big feature, Replit brings in a sub-agent called the Architect. And the Architect, it turns out, runs on Codex and OpenAI.”
Disclosure
Lemkin: 95% of SaaStr AI agent calls run on OpenAI Mini
“About 95% of the OpenAI calls we're using are with Mini, which costs less than a penny per call.”
Disclosure
OpenAI runs its go-to-market operations primarily on Clay
“At OpenAI, we pretty much run go-to-market on Clay. It is infused into everything that we do.”
Assertion Not checkable as stated
40 Harvard and Stanford MBAs who rejected OpenAI later begged to return
“I've had no less than 40 of those people reach back out to me over the years wanting to come back to OpenAI.”
Disclosure
Hott: Left a 50-person team at Webflow to become an IC at OpenAI
“When I went to OpenAI, so many people in my network said, why would you go to this small company? It was, go-to-market was only seven people, and I was going from leading an org of 50 people at Webflow as a senior director to being an individual contributor at…”
Disclosure
Persson: Snowflake internal agents run on Cortex with Anthropic and OpenAI
“These are proprietary models based, you know, using Snowflake. Cortex, and then we're using various large language models to build these agents. That could be, you know, it could be anthropic, you know, it could be, you know, OpenAI, you know, and other models…”
Assertion Contradicted
OpenAI Surpassed BloombergGPT Within Three Months of Its Launch
“Bloomberg trained a model Bloomberg GPT that outperformed the latest OpenAI model of the time. And then three months later, OpenAI shipped the new model, and it was way better.”
Assertion Supported
Hott: OpenAI scaled from ~200 to ~1,000 employees post-ChatGPT launch
“We were about 202 150 people, and we had reached a pivotal moment. We had to scale quickly. We had to end the year at somewhere close to a thousand people.”
Assertion Not checkable as stated
Hott: Slack, Webflow, and OpenAI paid early sales reps full salary
“We did this at Slack, Webflow, OpenAI, and I know many other greats have done it like Calendly, New Relic, and Airtable.”
Assertion Supported
OpenAI has kept the ChatGPT Plus subscription unchanged for over a year
“ChatGPT Plus has been our ChatGPT subscription for over a year now, and so that's an example of something we haven't changed that I think I'm proud of and we've been thoughtful about where and when We'd change that, but it has stayed the same in the market, wh…”
Assertion Not checkable as stated
Grill: Received 900 LinkedIn messages in four hours after joining OpenAI
“I think I had 900 in-mails within four hours of posting that.”
Assertion Contradicted
Tunguz: Microsoft OpenAI customer count tripled in recent quarter
“You look at the total number of customers that are starting to use Microsoft's OpenAI technology has three X, I think, in the last quarter in the reporting.”
Assertion Not publicly verifiable
AstraZeneca used Notion AI to save time on drug discovery
“Notion launched Notion AI, where you could translate, edit, improve your notes, which AstraZeneca used to save time and cost on drug discovery.”
Disclosure
OpenAI built two separate sales teams for its two main products
“A year and a half later, it's the opposite of chill and we've got now two different products, two whole different sales teams, we're hiring like crazy and we're on a rocket ship.”
Assertion Supported
Rosenthal: English language averages roughly 1.3 tokens per word
“A token is almost a word. It turns out there's about 1.3 tokens per word in the English language, and you consume tokens when you put in a prompt, and then you get an output or a completion.”