- AI Sidequest: How-To Tips and News
- Posts
- I can't emphasize enough how weird these models are
I can't emphasize enough how weird these models are
Plus, the """triple quote""" method.
Issue #15.
On today’s quest:
— AI isn’t sentient, but it’s not like a computer either.
— Try using triple quotes (“““).
— Axel Springer does a big AI content deal.
Tip: Prompting is not like working with a computer
Open AI released tips on how to write better prompts, and I’m struck again by the absolute weirdness of these models. Forget most of what you know about working with computers. Consider this tip:
Give the model time to "think"
If asked to multiply 17 by 28, you might not know it instantly, but can still work it out with time. Similarly, models make more reasoning errors when trying to answer right away, rather than taking time to work out an answer. Asking for a "chain of thought" before an answer can help the model reason its way toward correct answers more reliably.
Tactics:
* Instruct the model to work out its own solution before rushing to a conclusion
* Use inner monologue or a sequence of queries to hide the model's reasoning process
“Reasoning errors”? “Ask it if it missed anything on previous passes”? These are not phrases I associate with literal, logic-based machines I’ve worked with my whole life.
I’ve also seen multiple people say they get better responses by using encouragement — such as offering to give ChatGPT a $100 tip, telling it the results are really important, telling it you know it can do it, and so on.
AI tools like ChatGPT can behave as if they have desires or emotions. For example, one study found that the prompt “Take a deep breath and work on this problem step by step” resulted in higher accuracy than just “let’s think step by step.”
So although it’s important to write clear prompts, give AI examples to work with, and so on — if you aren’t getting the results you want, you can also try treating chatbots more like needy humans.
Tip: Use triple quotes
The full page of prompt-writing advice from OpenAI is worth reading, but the other practical tip that jumped out at me was that all the examples used triple quotes to designate text you want to give the system. For example:
Summarize the text delimited by triple quotes in about 50 words.
Use the provided articles delimited by triple quotes to answer questions. If the answer cannot be found in the articles, write "I could not find an answer."
News
Axel Springer strikes landmark deal with OpenAI over access to news titles
The publisher of “Bild,” “Politico,” “Business Insider,” and more will receive a large upfront payment to let OpenAI use its content to train AI models. Axel Springer will also receive ongoing annual payments (rumored to be “eight digits”) that will allow Open AI to create near real-time summaries of articles to use in responses. The summaries will include links to the original material. — Financial Times
Investing.com uses AI to copy competitors — wholesale
Taking the opposite approach from OpenAI, Investing.com seems to be just taking competitors’ articles. The company has been using AI to churn out a massive number of stories that are clearly based on other articles, doing it very soon after the original articles appeared, and posting them without attribution or block quotes. — Semafor
A few more quick hits:
Anthropic will help users if they get sued for copyright infringement.
H&R Block is introducing AI tools to help people with tax preparation, and the IRS is (separately) using AI to catch tax cheats.
What is AI sidequest?
Using AI isn’t my main job, and it probably isn’t yours either. I’m Mignon Fogarty, and Grammar Girl is my main gig, but I haven’t seen a technology this transformative since the development of the internet, and I want to learn about it. I bet you do too.
So here we are! Sidequesting together.
If you like the newsletter, please share it with a friend.
Written by a human.