One Useful Thing

Sorry, when I say I think I have a balanced take on AI what I mean is that I do my best to weigh both the pros and cons of this technology as opposed to a more extreme behavior like spending all day chatting with LLMs or posting all day on X about how AI is already better than me at everything and that jobs are over.

If I had to assign a confidence score for whether agents will change the way we all work and many aspects of how we live, I would put it at a 7/10, maybe 8/10. I felt about the same about the smartphone. While many things we do look the same way they did in 2005 (we still drive on roads, kids still go to school), at the same time it's undeniable that much of our lives are intermediated through a small screen and many societal dynamics have shifted due to that technology's existence.

I will concede that you should read my post with that context and draw your own conclusions about the veracity of my perspective — but I think it is more well-reasoned than what people generally attribute to "LLM hype". (Of course it's a bit tautological that I believe that, but I try to surround myself with people of all kinds technical and non-technical and like to think I stay reasonably grounded.)

All that said, I think the code from a leading company being bad and yet delivering good results is more a sign of the technology's jagged frontier[^1]. Calculators can't write sonnets the same way that LLMs are bad at math, but that doesn't make them useless — it just makes them a…

LLMs can definitely produce original other stuff: ask it to create an original poem and on an extremely specific niche subject and it will do so. You can specify the niche subject to the point where it is incredibly unlikely that there is a poem on that subject in its training data, and it will still produce an original poem on that subject [0]. The well-known "otter using wifi on a plane" series of images [1] is another example: this is not in the training data (well, it is now, because well-known, but you get the idea).

Is there something unique about code, that is different from language (or images), that would make it impossible for an LLM to produce original code? I don't believe so, but I'm willing to be convinced.

I think this switches the burden of proof: we know LLMs can produce original content in other contexts. Why would they not be able to create original code?

[0] Ever curious, I tested this assumption. I got Claude to write an original limerick about goats oiling their beards with olive oil, which was the first reasonable thing I could think of as a suitably niche subject. I googled the result and could not find anything close to it. I then asked it to produce another limerick on the same subject, and it produced a different limerick, so obviously not just repeating training data.

[1] https://www.oneusefulthing.org/p/the-recent-history-of-ai-in-32-otters

If you don't want to read the full thing, Mollick's substack has more accessible summary of this paper:

https://www.oneusefulthing.org/p/centaurs-and-cyborgs-on-the-jagged

There are two sides to this thing. The first author of this paper has also written another paper titles "Falling Asleep at the Wheel: Human/AI Collaboration in a Field Experiment". Abstract:

"As AI quality increases, humans have fewer incentives to exert effort and remain attentive, allowing the AI to substitute, rather than augment their performance... I found that subjects with higher quality AI were less accurate in their assessments of job applications than subjects with lower quality AI. On average, recruiters receiving lower quality AI exerted more effort and spent more time evaluating the resumes, and were less likely to automatically select the AI-recommended candidate. The recruiters collaborating with low-quality AI learned to interact better with their assigned AI and improved their performance. Crucially, these effects were driven by more experienced recruiters. Overall, the results show that maximizing human/AI performance may require lower quality AI, depending on the effort, learning, and skillset of the humans involved."

https://static1.squarespace.com/static/604b23e38c22a96e9c78879e/t/62d5d9448d061f7327e8a7e7/1658181956291/Falling+Asleep+at+the+Wheel+-+Fabrizio+DellAcqua.pdf

Huh. They're not "Pelicanmaxxing"... they're Ottermaxxing.

Take a look at the GLM 5.2 and Deepseek V4 "animal on a plane" examples. In every case, the animal is standing on top of the plane, a clear misunderstanding of the concept of "animal on a plane"... with the exception of the Otter. The otters are sitting in a seat on the plane, looking out the window.

That's Ethan Mollick's "Otter On A Plane Using WiFi" image benchmark.

https://www.oneusefulthing.org/p/the-recent-history-of-ai-in-32-otters

(Sometimes the Racoon is sitting inside the plane as well, but the racoon is a common backup benchmark. I'm surprised it wasn't also holding a sign saying that it loves trash.)

Also, Grok seemed to really really enjoy "whale on a plane" in that second round, and kudos to GPT Terra for deciding after 3 rounds that the user was terrible at spelling and generated "Antelope On A Plain".

EDIT: I promise I'm a human, but I did just notice my "that's not x... that's y" construction at the start. I am rather Claudepilled — my apologies.

This is just me, and I find that I go through periods of listening to podcasts and then it falls off my routine.

When I would think about it, listening to podcasts was low intensity edutainment, it feels like you are learning something but you really aren't, or the learning density is so low that most 45 minute podcasts could be summarized in 10 bullet points. It was not unlike watching a BBC planet earth nature show. Afterwards you know that a bird somewhere can spin its head around twice without breaking its neck but not much more.

Oxford Researchers Discovered How to Use AI To Learn Like A Genius https://www.youtube.com/watch?v=TPLPpz6dD3A

Use of generative AI tools to support learning https://www.ox.ac.uk/students/academic/guidance/skills/ai-study

https://www.oneusefulthing.org/p/assigning-ai-seven-ways-of-using

https://www.sciencedirect.com/science/article/pii/S0099133323000599?via%3Dihub

Bloom's Taxonomy https://www.phind.com/search/cm6o98lki00002e6g5anos0bv

• Bard when launched was using one of the PaLM models, which is inferior to GPT-3.5 (the free version of ChatGPT).

• Bard since a couple of months ago was using Gemini Pro, which is roughly comparable (a bit worse or better depending on whom you ask) to GPT-3.5.

• Bard is now (today) called Gemini, available as "Gemini" and "Gemini Advanced". The former is still comparable to the free version of ChatGPT, and the latter version costs $20/month and uses the "Gemini Ultra" model, and is meant to be roughly comparable to GPT 4 / the paid version of ChatGPT. (Their paper claimed it to be better than GPT 4 on some benchmarks, but real-world usage will show which way it goes — but it should be significantly better than Bard from recently. Edit: See https://www.oneusefulthing.org/p/google-gemini-advanced-tasting-notes from someone who's been using it for six weeks.)