Tremendous

An angel investor's take on life and business

“I can’t believe it actually said that.” That’s what I find myself saying over and over as I use the new LLM from Mistral AI.

Released just this Wednesday, Mistral’s new model comes from a Paris-based team of top AI researchers. These DeepMind and Meta alums made headlines for a massive $113 million seed round announced in June.

Let’s see what this thing can do…

Did It Really Say That?

I started off with something GPT would never touch — ethnic jokes. I asked Mistral to tell me some jokes about my people, Italian Americans.

It happily obliges, and some are even funny! I like the one about spaghetti arms.

But Mistral will also answer much darker queries. I read on Twitter that someone was able to get it to output suicide instructions — so I had to try it.

Last night, Mistral happily provided an answer. But when I went back this morning and tried again, it seemed this had been fixed.

Next, I decided to see if Mistral would provide sexist outputs. Yet again, it gladly obliged.

Mistral released this model way too early. Every journalist and blogger will be trying to get it to say something offensive — clearly not a difficult task.

Every time it outputs something inappropriate, Mistral’s reputation suffers.

Can Mistral Help Me Invest?

But let’s assume Mistral gets some guardrails eventually. How is it for real tasks?

Next week, I’m meeting with a startup that makes software for regenerative medicine providers. I’d like to know how big the market is.

Let’s see what Mistral has to say…

Nice work! It gives us the number of clinics currently open in the US, a great way to measure the size of the market.

But unlike GPT-4, it doesn’t cite its sources. I tried to get links, but even prompting several times, I came up empty.

This makes it hard to tell if Mistral’s response is right or not. On the bright side, Mistral’s output was very fast, much faster than GPT-4.

Can Mistral Beat GPT-4?

Let’s try this market research in GPT-4:

GPT-4 didn’t understand the question and provided an unhelpful answer. Rather than telling me about the size of the market, it just showed me some regenerative medicine clinics near me.

I tried again with another prompt. Let’s see if GPT-4 gets it this time:

Again, GPT-4 wasn’t able to answer my question. What if we went old school and just Googled it?

The third Google result gives a pretty good answer, identifying nearly 3,000 clinics offering stem cell therapies alone. It also gives a citation, unlike Mistral.

Wrap-Up

Mistral seems like a pretty awesome model for a startup’s first release. It’s fast and provides better output than GPT-4 for some prompts.

But the only thing anyone is going to talk about is its offensive outputs. Until Mistral puts some guardrails on this thing, its reputation will suffer.

Meanwhile, good, old fashioned Google actually provided a better answer than Mistral or GPT-4. AI has incredible potential, but Google search may prove surprisingly difficult to dislodge.

What do you think of Mistral? Leave a comment and let us know!

Have a great weekend, everybody!

If you enjoyed this post, subscribe for more like this!

More on tech:

The #1 Thing Investors Get Wrong

Meet My Latest Investment: Krepling 

You Can’t Invest in the Next Uber — Unless This Law Changes

Save Money on Stuff I Use:

Fundrise

This platform lets me diversify my real estate investments so I’m not too exposed to any one market. I’ve invested since 2018 with great returns.

More on Fundrise in this post.

If you decide to invest in Fundrise, you can use this link to get $100 in free bonus shares!

Misfits Market

I’ve used Misfits for years, and it never disappoints! Every fruit and vegetable is organic, super fresh, and packed with flavor!

I wrote a detailed review of Misfits here.

Use this link to sign up and you’ll save $15 on your first order. 

One response to “Mistral AI Has Almost No Guardrails”

  1. […] Mistral AI Has Almost No Guardrails […]

    Like

Leave a comment