- Writing Inc
- Posts
- 🤖 AI reels from Silicon Valley Bank's collapse
🤖 AI reels from Silicon Valley Bank's collapse
PLUS: How much Microsoft spends on running ChatGPT
Happy AI Tuesday.
Our hearts go out to any startup affected by the recent SVB news (covered later in this article.)
We’re glad your funds are safe.
🤖 Top Stories
🤯 How much Microsoft spends on powering ChatGPT

Source: Microsoft
9-figures. Hundreds of millions of dollars.
That’s how much it costs to build the core infrastructure for ChatGPT.
In case you were wondering.
Microsoft has an ungodly amount of AI infrastructure. Good thing, too. That's infrastructure to power ChatGPT, Bing AI, Github Copilot…you get the idea.
Want a visualization? They needed tens of thousands of localized GPUs to make it happen.
Trust us, add this one to your reading list. Their blog post outlines the massive technical challenge behind building for AI at scale.
What’s that quote about selling pickaxes during a gold rush?
😬 AI is already making healthcare decisions

Hi, I’m your new AI doctor, and, well, it’s not looking good.
An AI might be making your healthcare decisions without you even knowing it.
An investigative report from Statnews revealed just how much insurance companies use AI. Those decisions are behind the scenes, about coverage and when to cut it off from patients.
Medicare Advantage is a program in the US offering lower cost healthcare coverage for senior citizens. The tradeoff is there is much more leeway for the program to deny coverage to patients.
The report outlines how companies calculate coverage based on historical data, using AI. Let's say AI calculates that a 67-year-old woman with a certain set of conditions needs 13 days of recovery from a procedure. That’s what the program offers. Regardless of the actual recovery time. And appeals can take years.
Now, Data science is no new concept to insurance companies. Actuarial medicine is an old profession.
What's new is the role of AI. Healthcare is controversial enough. Applying AI-based decision making is a step in a new direction.
♥️ AI startups everywhere breathe sigh of relief

As a lot of you noticed, there wasn’t a lot of AI news over the weekend.
That’s because Silicon Valley bank collapsed.
It’s not the biggest bank in the world. But it banked with A TON of startups. So when it failed Friday, and the FDIC took it over, a lot of them panicked.
Why the panic?
See, the FDIC insures any bank account up to $250k. An account will receive that amount of money, from the FDIC, if a bank fails.
Most startups keep more than that in cash.
Startups went from “things are fine” to “I can’t pay my employees next week” in about 24 hours.
The community responded well. OpenAI CEO Sam Altman personally offered to send money to affected companies, too.
Now, thankfully, the Fed stepped in and guaranteed all deposits.
Every customer will have full access to their funds, no matter how large.
But many say startups - job creators - held their collective breaths over the weekend.
We’re glad that these companies are okay.
😎 AI tools, links, guides
GIVE US MIDJOURNEY V5
Midjourney v5 is releasing this week.
They're asking paying members to help rate images, I did my part and spent the morning on it.
The level of detail is wild.
— phil desforges ⟠ (@storybyphil)
4:00 PM • Mar 13, 2023
Some more additive prompting.
ok, I got ChatGPT working with Additive Prompting
Here's a 1 paragraph ChatGPT prompt you can use to generate infinite interior design/architecture photographs w/ 90%+ coherence to the prompt in Midjourney
Full prompt w/ examples in thread. Try reading the prompts as you go
🧵
— Nick St. Pierre (@nickfloats)
3:12 AM • Mar 13, 2023
AI is getting scary good.
It is crazy that, in a span of less than 3 months, AI beat most of our (admittedly flawed) major tests for creativity & sentience. It acts like it has Theory of Mind. Bing would pass the Turing & Lovelace Tests. It beats the alternative uses & remote association creativity tests. https://t.co
— Ethan Mollick (@emollick)
8:54 PM • Mar 13, 2023
Stanford takes on LLaMA.
LLaMA has been fine-tuned by stanford,
"We performed a blind pairwise comparison between text-davinci-003 and Alpaca 7B, and we found that these two models have very similar performance: Alpaca wins 90 versus 89 comparisons against text-davinci-003."
— anton (@abacaj)
7:02 PM • Mar 13, 2023
🤖 Got feedback for us?
Head over to our Twitter, @RobotRockDaily, and let us know what you’d like to see more or less of.