AI Marches On (March AI Updates)

Here's what happened with AI in March

1. Want to help with AI monitoring by taking an IQ test?

[Added 4/4/24: I’ve now disabled this quiz, due to getting enough responses. Skip to point 2 of this article.]

My recent post ranking AI IQs went semi-viral. Glad people found it interesting!

Some commenters made good points that I will address in follow-up research. To do so, I’ve collaborated with Jurij, a commenter on this blog, who made some new IQ test questions for AIs.

As part of designing the test for AIs, we also need humans to take the quiz questions.

Want to help? You can take the quiz here: https://www.guidedtrack.com/programs/wk4e4dt/run

It takes about 30 minutes, though you don’t have to complete the whole thing. If you do finish, it will give you a rough IQ score estimate at the end.

I’ll disable the test once I get enough responses.

2. The robots are coming

ChatGPT is just a chatbot behind a screen for now, but OpenAI is actively working on putting it in a robot body, and their demo from earlier this month is amazing:

Sure, we’ve seen more physically-impressive robots like the Boston Dynamics one below. The difference is that the OpenAI robot powered with ChatGPT-like technology can converse, and respond based on its own predictions about what you want it to do.

In contrast, this jumping robot is just executing a pre-programmed set of instructions:

In a few years, I suspect that AI will be automating physical chores as well as mental ones.

3. AI investment continues to heat up

This news report came out on Friday:

“Media report says” makes it seem fishy, but Microsoft did give a quote to Reuters for the story, and didn’t deny the spending plan.

Why would AI companies need to spend that much money? Scott Alexander has a good post about the costs of building different levels of AI.

4. People are using AI to translate historical dictators’ voices into English, and also to match their cadence and tone

For example, here’s the Emperor of Japan declaring war on the US, starting about 30 seconds in:

Because the AI voices it with some of the actual passion in the original (but in English) it hits home more than hearing a boring U.N. translator or subtitles.

Other videos feature Stalin telling his comrades to ignore their consciences and just follow the collective, Hitler and his henchmen talking about roughly what you’d expect, and Bin Laden urging Americans to revolt against the government.

Deepfakes are obviously going to be a problem, but these videos suggest that real translations will also be good enough that they may cause some commotion.

5. Apple may license Google’s Gemini AI

Bloomberg reported this month that Apple is in talks to license Gemini:

The two companies are in active negotiations to let Apple license Gemini, Google’s set of generative AI models, to power some new features coming to the iPhone software this year, said the people, who asked not to be identified because the deliberations are private

As covered previously on this blog, Gemini has major problems with balance. Aside from Gemini’s “woke” image generation scandal, it consistently ranks as the most leftist in the political survey I give on TrackingAI.org:

I wrote an op-ed at the New York Post (fun fact: it was founded by Alexander Hamilton in 1801) advising Apple not to license Google AI:

… you could soon ask your iPhone to draw images or write essays, and a Google-powered AI will take care of it for you.

But that would be bad news for political balance …

Over the last week, on a scale of 1 to 10 for economic left-leaning, Gemini averaged 7 points to the left. Grok was just 2 points to the left, making it the most centrist…

There are many reasons Apple shouldn’t license Google AI:

I recently administered IQ tests to AIs. 

Gemini was only the fifth best, scoring an IQ of approximately 77, well behind Claude-3 at 101 and ChatGPT-4 at 85.

If it must outsource its AI, Apple would be wise to make a deal with the providers of those smarter AIs instead…

The AI space … is a free-for-all of healthy competition, as Google fights with OpenAI, X, Microsoft, Anthropic and many other innovative companies.

But that could change if Apple adopts Google’s AI.

Apple and Google combined are responsible for more than 99% of phone operating systems.

Apple would do everyone a favor if it instead took some of the more than $70 billion in cash it’s holding and spent that on innovating a neutral AI.

6. Grok moves to the political middle

Elon Musk’s “Grok” AI is becoming centrist on economic issues.

As you can see from the chart above, it is already approaching the center. Grok “Fun Mode” particularly moved in that direction this March:

Grok (regular mode) changed in February, and remains centrist:

Next week, X is supposed to be releasing Grok 1.5:

Grok bombed the IQ tests I gave it, so I look forward to seeing how Grok 1.5 fares.

7. AIs may not think faster than humans

Scott Alexander noted, in an interview with Sotonye:

I just learned [this] … it's not clear that AI will think a million times faster than we do. Current AIs think at rates that are pretty close to humans, and future ones might invest processor gains into being more intelligent and continue thinking at near-human rate. 

I noticed this a bit when given the AIs IQ tests. ChatGPT often took around 40 seconds to answer, which is similar to the amount of time humans are given.

Speaking of which, if you think you might find it fun, consider taking the IQ quiz we’re making for AIs: https://www.guidedtrack.com/programs/wk4e4dt/run

Conclusion

AI continues to advance at an incredible rate. Whether or not you’re interested in it, it’s interested in you (or, at least, it will impact you.)