Let's be Fwends #136:

Relying on AI Will Impede Your Growth

"But there was a computer code-named Merlin. Its memory-bank contained all human knowledge. It was capable of scanning all its data instantaneously, and combining, and forming associations, and reasoning with absolute accuracy, and extrapolating to produce new facts, and predicting future events. And if you'd asked such a computer, "Is there a God?" it would have simply answered, 'Present.'"

~Henry Beam Piper, The Cosmic Computer

Welcome to this edition of Let's be Fwends! Today, I try to explain why I think that an over-reliance on AI for everyday tasks is a big mistake, we look at current LLMs' ability to deceive, and get (and hopefully heed) some good advice, in form of life lessons learned and binge-watching recommendations.

AI is a Dangerous Shortcut

People are using Generative Artificial Intelligence for all kinds of stuff right now. From categorising (makes sense) to catching spelling mistakes (probably makes sense) to write emails based on a couple of half-baked thoughts (makes no sense at all, and is probably also really disrespectful of the time of the recipient).

Skill comes from deliberate repetition, which allows you to iteratively improve over time. If you allow AI to do all your "small" work, how do you suppose will you be able to do your "big" work?

Being deliberate in everything you're doing (ok, let's settle for 'many things you're doing') is a serious growth hack, but maybe more about this some other time.

The Ability to Deceit Emerged in Current LLMs

"This study reveals that (deception) strategies emerged in state-of-the-art LLMs, but were nonexistent in earlier LLMs. We conduct a series of experiments showing that state-of-the-art LLMs are able to understand and induce false beliefs in other agents, (...), and that eliciting Machiavellianism in LLMs can trigger misaligned deceptive behavior. GPT-4, for instance, exhibits deceptive behavior in simple test scenarios 99.16% of the time." Source

It's hard to tell how meaningful this is, but I think it's link-worthy for a couple of reasons:

1. The jailbreaking prompts this study is using are really interesting, even if probably closed by now by the LLM vendors.
2. It's a stimulating thought experiment to speculate how an LLM without moral guardrails would respond to such tasks.
3. Emergence appears to be a prominent property of sophisticated LLMs, which makes all findings that point to such properties significant.

68 Bits of Unsolicited Advice

I used a good part of my summer vacation to clean out my head. The remaining time I used to clean out my digital archive (okay, fine, I started to clean it out. Happy now?).

One thing I found was a misfiled note about some really excellent bite-sized advice called "68 bits of unsolicited advice" by Kevin Kelly. It was four years old, and that tells you something about the light-situation in my archives.

In any case, this list is just as valid as it was four years ago, but you can no longer link to it because Kelly turned it into a book and consequently removed the list from the web.

But the internet doesn't forget and here are some of my favorites:

Don’t be the smartest person in the room. Hangout with, and learn from, people smarter than yourself. Even better, find smart people who will disagree with you.

I wasted years of my life trying to be smarter than the people I spend time with in the hopes to impress them. People smarter than me made me nervous and I could be a real dick around them. Fortunately, I was able to get rid of this trait many years ago, even though it still hunts me from time to time.

Everyone is shy. Other people are waiting for you to introduce yourself to them, they are waiting for you to send them an email, they are waiting for you to ask them on a date. Go ahead.

Is that really true? If so, then who is doing all that outreaching and contacting and mingling and stuff? Let me know. It's puzzling me.

Over the long term, the future is decided by optimists. To be an optimist you don’t have to ignore all the many problems we create; you just have to imagine improving our capacity to solve problems.

I'm pretty sure that somehow relates to the first advice I cited above.

The 150 Best Sci-Fi Movies

You might or might not agree with the ranking of the 150 best Sci-Fi movies of all time in Rolling Stone, but not only appears this list quite complete, it also includes many gems I forgot about or never heard about.

So if you're already planning out your autumn and winter evenings: Let's be Fwends got you covered.

That's it for this edition of Let's be Fwends. Thanks for making it down here! 🪗

Header Image Credit