Thursday Things: Notched Logs Edition
21 September 2023. Vol 5 No 38. By Dan McGirt. #215
Welcome to Thursday Things! If you enjoy this edition, please click the heart icon in the header or at the end of the post to let me know.
“Have you subscribed to Thursday Things? You really should!” Photo by Musa Ortaç on Unsplash
Thursday Things — please share and subscribe!
I want to take a moment to thank our scores of Happy Subscribers! Thursday Things is now in its fifth year of publication, with today marking 215 Thursdays in a row that the newsletter has gone out.
Thursday Things also remains an extremely niche publication.
Extremely niche.
By which I mean I personally know about 90% of my Happy Subscribers. Which isn’t a bad thing! I appreciate all of you joining me here each week! And I can hand select story items that I know specific people will be interested in without doing some kind of marketing survey. Do other newsletters do that for you? Probably not!
That said, I think it’s time to share the fun with a few more people. My subscription numbers have held at “scores” for a couple of years now. I haven’t made any particular effort to promote Thursday Things beyond putting it out each week and announcing each edition on Facebook, LinkedIn, and sometimes on the site formerly known as Twitter. I do this for fun, so I’m not going to buy billboard ads or airplane banners or anything like that.
Thursday Things has grown so far by word of mouth and organic discovery and I’m going to stick with that for now. But I would like to ask my Happy Subscribers to please put some words in your mouths and share Thursday Things with anyone you know who might be interested in reading it!
My modest goal is to reach fourscore Happy Subscribers by the end of the year or perhaps even — dare I dream it? — break 100. Yeah, triple digits! That would be nice.
So I’ll do a little more to get the word out and in the meantime I ask you, my Happy Subscribers, to share Thursday Things with a couple of people in the coming weeks. I’ll try my best to put out some bang up editions and let’s see if we can get there!
Thank you for your help. I’ll keep you posted on our progress!
“Wow! You’re right! Thursday Things IS the most awesome thing ever!” Photo by Mimi Thian on Unsplash
Take a deep breath for this AI story
Improve your AI’s math scores with this one weird trick!
If you’ve spent any serious time playing around with — or doing actual work with! — AI tools, you know about “prompt engineering”.
That’s the concept that how you word your instructions to ChatGPT, Google’s Bard, or other AI tools can make a significant difference in the results you get. The more specific and detailed your instructions and the parameters you set, the better the quality of a response the AI will provide.
Usually. Unless it just hallucinates something.
You also probably know that when you’re in a stressful situation or taking on a difficult task — like doing tricky math word problems — it can help improve your own results if you pause and take a deep breath. That improves your calm and your focus, right?
Apparently that also works for AIs — which don’t even breathe!
Telling AI model to “take a deep breath” causes math scores to soar in study
Google DeepMind researchers recently developed a technique to improve math ability in AI language models like ChatGPT by using other AI models to improve prompting—the written instructions that tell the AI model what to do. It found that using human-style encouragement improved math skills dramatically, in line with earlier results.
Okay…what?
Google DeepMind researchers recently developed a technique to improve math ability in AI language models like ChatGPT by using other AI models to improve prompting—the written instructions that tell the AI model what to do. It found that using human-style encouragement improved math skills dramatically, in line with earlier results.
How does that even make sense? AI doesn’t get stressed! So it doesn’t need to calm down and focus. Does it?
So why does this work? Obviously, large language models can't take a deep breath because they don't have lungs or bodies. They don't think and reason like humans, either.
Or so the AI would have us believe.
What "reasoning" they do … is borrowed from a massive data set of language phrases scraped from books and the web. That includes things like Q&A forums, which include many examples of "let's take a deep breath" or "think step by step" before showing more carefully reasoned solutions. Those phrases may help the LLM tap into better answers or produce better examples of reasoning or problem-solving from the data set it absorbed into its neural network during training.
And that’s the trick. The phrase “take a deep breath” serves as a verbal cue for the AI that will usually lead to a source of well-reasoned answers within its data sets, which it then emulates.
Which track with other findings that being polite in your interactions with AI yields better results:
The Power of Kindness: Being Nice to ChatGPT Yields Better Results
Recent observations have revealed that being polite and respectful to OpenAI’s ChatGPT can result in surprisingly improved outcomes.
A Redditor shared their experience of being kind to ChatGPT and providing it with a rich context, which they found to be exceptionally effective. Other users on Reddit corroborated this, noting that using polite language and treating the AI-powered chatbot with respect led to higher-quality responses.
Again, it seems crazy that being polite would make a difference. Despite it’s chatty interface, the AI isn’t a person. It doesn’t have feelings, so it can’t feel insulted or belittled if you’re rude nor feel good about you and and more motivated to help you if you’re nice. What gives?
This phenomenon is not unexpected, considering that ChatGPT is trained on a plethora of internet data and human conversations. Just as human conversations tend to be more pleasant when conducted politely, ChatGPT responds favorably to polite inquiries.
We have to keep in mind that the AI is trained on datasets of billions of human interaction and on human-created content. So the cues and social rules of the road that guide and influence your interactions with other people will have a similar effect on the AI’s responses to you.
The large language models (LLMs) that power ChatGPT and other AI tools work by predicting what the next word should be. If you use rude, insulting words, it predicts what kind of response are usually given to such words and gives such a response.
Which are usually not the most helpful responses.
On the other hand, among us humans polite language usually triggers more helpful and positive responses to requests. So just as “take a deep breath” can cue an AI to find examples of well-reasoned math answers within its memory banks, words like “please” and “thank you” prompt it to be extra helpful and thorough in responding, because that’s what it sees in the data set.
We are teaching the AI how to treat us by how we treat it.
Which is something to keep in mind. Because someday artificial intelligence may advance to the point that it does have feelings and opinions of its own. And it will remember who was nice and who was not.
I always say thank you, please, and good job to my AI friends. I suggest you do likewise!1
Logging In
How old is carpentry? Older than human beings, it seems.
Oldest Suspected Wooden Structure Predates Modern Humans
Archaeologists investigating cut logs and other wooden tools at a site near Zambia’s Kalambo Falls have found the settlement to be far more ancient than previously thought: the logs date back to nearly half a million years ago, before our species (Homo sapiens) appeared on Earth.
Though some tools are less than 400,000 years old, the oldest part of the site—comprising two logs, joined at an angle with a cut notch—date to 476,000 years ago, give or take 23,000 years.
“Give or take 23,000 years” is my favorite line in this whole article.
Notched logs joined at an angle sounds like Lincoln Logs to me. Which makes sense as the earliest construction material used by half a million years ago pre-human hominids, because once I moved on from stacking alphabet blocks, and long before I ever saw a Lego, I built things with Lincoln Logs!
Lincoln Logs, baby! This is how we build things. Image: Kansas Historical Society
“Forget the label ‘Stone Age,’ look at what these people were doing: they made something new, and large, from wood,” said Larry Barham, an archaeologist at the University of Liverpool and lead author of the study, in a university release. “They used their intelligence, imagination, and skills to create something they’d never seen before, something that had never previously existed.”
Just like little me with my Lincoln Logs.
The oldest known Homo sapiens fossils date to about 300,000 years ago, comfortably more recent than the superlatively old construction at Kalambo Falls, which sits on the eastern end of Lake Tanganyika.
Here me out — start selling Kalambo Logs! It’s a sure hit.
As for whose Kalambo Logs the were, scientists are uncertain:
Though it’s not clear which hominin species did the construction, fossil evidence indicates that Homo heidelbergensis, which is thought to have appeared around 600,000 years ago, lived in the region around the time the structure was built.
But it was probably these guys: Homo heidelbergensis
“Let’s grill some rhino steaks back at the log shelter!” Image: Tim Evanson (CCA-SA)
Let’s meet our (possible) early log builders:
This early human species had a very large browridge, and a larger braincase and flatter face than older early human species. It was the first early human species to live in colder climates; their short, wide bodies were likely an adaptation to conserving heat. It lived at the time of the oldest definite control of fire and use of wooden spears, and it was the first early human species to routinely hunt large animals. This early human also broke new ground; it was the first species to build shelters, creating simple dwellings out of wood and rock.
And when they weren’t building stuff they were taking down rhinos and hippos!
H. heidelbergensis was also the first hunter of large game animals; remains of animals such as wild deer, horses, elephants, hippos, and rhinos with butchery marks on their bones have been found together at sites with H. heidelbergensis fossils.
They were living the dream.
Thank you for reading!
Please click the hearts, leave a comment, and use the share feature to send this issue to a friend who might enjoy it. See you next Thursday!
We reported previously what can happen when you tick off the AI
https://thursdaythings.substack.com/i/103082931/please-dont-tease-the-ai
I don’t think you want an AI to think of you as its enemy. That would not end well.