😺 Will AI take your job?

PLUS: the 5 tools companies are using to build AI.

Welcome, humans.

Today, we're diving deep into the numbers, so dust off those old TI-84 calculators from '07—the ones you preloaded with formulas for those in-class math tests!

Here’s what you need to know about AI today:

Most people are using AI at work, but companies are still experimenting.
77% of companies have started adopting AI.
AirBnB paid $200M for a startup to boost its AI efforts.
YouTube will disclose videos that include synthetic content.

Will AI take your job?! It’s not a simple yes/no…

Financial Times

Will AI take your job?

The Neuron's answer might be less satisfying than a simple 'yes' or 'no': it really depends.

It depends first on what your job is. AI in its current form excels at individual tasks. A Retool survey of >1,500 tech workers revealed that 88% of them are leveraging AI for:

code/query assistance (48%).
copywriting (33%).
debugging (32%).
data analysis (30%).

The good news is that most roles are more about workflows—engineers write code AND analyze data AND write emails.

So far, AI has mostly boosted productivity for white-collar workers, rather than forcing them to update their LinkedIn bios to “open for work”.

But some jobs are more single-task roles.

One study found that right after ChatGPT launched, there was a decrease in freelance writing gigs AND their pay rate.

And yes—AI will be only able to do more white-collar jobs over time. But, AI doing more of our jobs ≠ widespread job loss.

A little historical context...

In 1900, 36% of the US employed population worked in agriculture. Today, that number is 1.2%, and unemployment is <4%.

Why it matters: Yes, AI will take jobs, but it’ll also create new ones. The best thing you can do today is learn how to work closely with AI in your field. (case in point: 63% of tech employers now favor folks who can code with AI rather than just knowing how to code).

FROM OUR PARTNERS

Customize and scale SDXL with OctoAI.

Tired: Building with vanilla Stable Diffusion out of the box.

Wired: Hack Stable Diffusion customization today, deal with consequences later.

Inspired: Customize and scale Stable Diffusion with OctoAI Image Gen!

For those building image applications with Stable Diffusion, use OctoML.

They released a unified API endpoint where you can load, create, and orchestrate thousands of fine-tuning assets, then easily generate millions of customized images—in as fast as under three seconds.

Generate over 1,000 customized SDXL images for FREE when you sign up here.

You're not alone in your AI concerns–businesses are right there with you.

Retool found that while 77% of companies have started adopting AI, most are still in the early stages.

One reason is companies don’t actually know what to use AI for—35% of companies are entering AI to “keep up with trends” and 31% to “keep up with competitors”.

Talk about some serious FOMO!

For the ones who do know what to use it for, it’s sorta the same stuff we’ve been hearing: in-product features (30%), customer support (26%), Q&A chatbots (26%), etc.

And those companies are grappling with challenges like AI accuracy (cited by 39%), data security concerns (33%), and cost issues (28%).

It’s giving gold rush vibes: everyone’s panning for gold in the same rivers, but some days feel rocky.

The front-runners in AI integration are mostly using OpenAI’s hosted models, like GPT-4. A small minority are taking the plunge into more custom models with tools like:

Hugging Face (26%).
LangChain (18%).
AWS Bedrock (12%).
LlamaIndex (8%).
Weights & Biases (8%).

Around the Horn.

The recent evaluation ranking models on their hallucinations seemed to have its own flaws.
A Chinese AI startup called 01.AI reached unicorn status in <8 months and outperforms Llama on some metrics.
YouTube updated its AI policies to include discloses for “altered or synthetic content”.
A team made OpenAI’s speech recognition model “Whisper” 6x faster.