Headlines You Should Know

 

OpenAI Announces ‘Preparedness Framework’ to Track and Mitigate AI Risks

Forget the Friday at 5 news dump — this is the just-before-you-slam-the-laptop-shut-til-January news dump … right? Considering recent events at OpenAI, it sort of feels that way. 

The ChatGPT creator published a blog post yesterday with the ominous title “Preparedness.” Within, OpenAI unveiled its “processes to track, evaluate, forecast, and protect against catastrophic risks posed by increasingly powerful models.” But why? Is this the precursor to artificial general intelligence (AGI), which has been the company’s goal since Day 1? Some have theorized that. Is it just mop-up duty after safety concerns were voiced around the time of Sam Altman’s temporary firing? Perhaps so, now that the board has veto power over launching new systems.

While it feels like there’s another shoe waiting to drop here, there is a widespread recognition that innovation in the AI space needs to be tempered to the speed of safe deployment. 

Elsewhere …

• LISTEN: Redefining PR in the Age of AI
• Pakistan’s Former Prime Minister Uses AI to Campaign From Prison
• REVIEW: Google’s New AI Music Generator, MusicFX
• GPT and Other AI Models Can’t Analyze an SEC Filing, Researchers Find

Tips and Tricks

 Prompt engineering, according to OpenAI 

What’s happening: The term “prompt engineering” may seem better suited for April or May in this year of swift development for AI. After all, getting good output from chatbots is less about formulaic instructions than plain ‘ol solid context these days. But OpenAI is throwing it back to a few months ago, using that title for a blog post published in the documentation section of its website. 

OpenAI says: The blog outlines six strategies, and if you’ve been visiting this space each week, many will look familiar. Here’s one we haven’t tackled: Giving the model time to “think.” OpenAI puts it this way: “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.”

Our take: This is an interesting way to reduce hallucinations. Sometimes, even with reference material, AI doesn’t pick up on the context you’ve laid out for it and goes down the wrong path. This is why it’s so important to review outputs thoroughly — not all hallucinations will be obvious. 

If you recognize a part of the content you’re building isn’t going the way you want, ask the AI to talk through just that specific section. Working in bite-size chunks generates better responses from the AI anyway, and makes editing a draft less cumbersome. In addition to asking for a chain of thought, you can also ask the AI to take another run at that section with the specific context that it was missing.

Quote of the Week

“If I had to make a prediction, in high-income countries like the United States, I would guess that we are 18–24 months away from significant levels of AI use by the general population. In African countries, I expect to see a comparable level of use in three years or so. That’s still a gap, but it’s much shorter than the lag times we’ve seen with other innovations.”

— Bill Gates’ 2024 predictions on his blog, GatesNotes